Hospital Medicine

Interstitial lung disease (ILD) encompasses a diverse group of disorders that cause inflammation and fibrosis of the lung parenchyma. The clinical manifestations, disease course, management and prognosis of ILD vary depending on the underlying subtype, making accurate classification and diagnosis an important initial step. While a comprehensive list of ILD contains dozens of disorders, the majority of patients will fall into 1 of 3 categories: exposure-related ILD, connective tissue disease-related ILD (CT-ILD), and the idiopathic interstitial pneumonias (Table).

An essential first step in the evaluation of every hospitalized patient with ILD is establishing a diagnosis. A common mistake among clinicians who diagnose patients with ILD is not realizing that ILD is a collection of diseases with different etiologies, natural histories, and treatments. A careful evaluation should be performed in every hospitalized patient with ILD to ensure an accurate diagnosis, ideally in the context of a multidisciplinary conference with pulmonary, radiology, pathology, and other specialties, as appropriate. A multidisciplinary panel of the American Thoracic Society/European Respiratory Society recently published a revised classification of ILD based on a combination of clinical, radiologic, and histopathologic findings, which may aid in refining the diagnosis.¹

There are 3 main scenarios in which the hospital physician will encounter patients with ILD.

Acute presentation of new-onset disease. While many ILDs present insidiously, some cases present acutely and require hospitalization. The most common of these are acute hypersensitivity pneumonitis (HP), CT-ILD (in particular, myositis-related and systemic lupus erythematosus-related), drug-induced ILD (eg, amiodarone, nitrofurantoin), cryptogenic organizing pneumonia (COP), acute eosinophilic pneumonia (AEP), and acute interstitial pneumonia (AIP).

Acute presentation of established (chronic) disease. Patients with chronic forms of ILD can present to the hospital with an acute exacerbation of disease. This can be caused by extra-parenchymal complications, including pulmonary embolism, pneumothorax, and pleural effusion; parenchymal complications such as infectious pneumonia, aspiration pneumonitis, and congestive heart failure; or without an identifiable cause. This latter presentation is most commonly seen in idiopathic pulmonary fibrosis (IPF).^2,3

Elective hospitalization for diagnostic surgical lung biopsy. Patients with ILD may be hospitalized electively for a laparoscopic surgical lung biopsy as part of their diagnostic evaluations.

Physicians caring for a hospitalized ILD patient must be familiar with the clinical presentations, diagnostic approach, medical management, and outpatient follow-up recommended in these 3 settings. We will summarize these areas and provide answers to commonly encountered clinical questions in the hospitalized patient with ILD.

Acute onset (or worsening) of dyspnea is the primary presenting symptom in most patients hospitalized for ILD. This symptom should be further characterized by assessing the degree of dyspnea and the extent of exercise limitation, as both impact overall disease severity and prognosis.⁴ Cough is the second most common symptom, and can be nonproductive, as is common in IPF, or be associated with secretions if parenchymal infection or acute bronchitis is present.⁵ Pleuritic chest pain, pleural effusion, and/or the presence of extrapulmonary features, including dysphagia, joint pain and swelling, or cutaneous thickening may suggest the presence of a CT-ILD. Because most forms of ILD present with only nonspecific symptoms, a careful history and physical examination are essential.

History

A comprehensive patient history is the backbone of diagnosing any ILD. History-taking should focus on severity and temporal progression of symptoms, presence of pre-existing systemic conditions associated with ILD, symptoms of extrapulmonary disease, and exposures to substances that can cause pulmonary injury, including a detailed history of occupations and hobbies, medications, smoking, and familial lung disease.^6-9 Physicians must try to exclude other diagnoses that could result in a similar acute presentation, including congestive heart failure and infection. Considering the complex and extensive recommended history-taking, physicians may find it helpful to use a standardized questionnaire, as provided by the American College of Chest Physicians.¹⁰

Laboratory Testing

All patients presenting to the hospital with a suspected ILD should undergo careful assessment for the presence of connective tissue disease, including patients without clear symptoms because ILD can be the presenting manifestation. We routinely test for antinuclear antibody titer and pattern, rheumatoid factor, anticyclic citrullinated peptide, creatinine kinase, and aldolase as the initial screening panel in most patients, with further testing directed by the findings on history and physical examination. Pulmonary function tests are used routinely to monitor disease progression in the outpatient setting; however, in the hospitalized ILD patient, they are often difficult to perform and have no real diagnostic value. Similarly, arterial blood gas is not routinely used as part of the initial inpatient evaluation.

Imaging

All hospitalized patients with a known or suspected ILD should undergo chest imaging, assuming they are stable enough to do so. While the chest radiograph can provide a low-cost initial assessment of the degree of lung involvement and presence of accompanying abnormalities, computed tomography (CT) scanning is the diagnostic test of choice.¹¹ The pattern and distribution of abnormalities on CT scan can greatly assist with the differential diagnosis in patients presenting with a new ILD, while the presence and pattern of new opacities superimposed on chronic changes can inform the differential and the prognosis of an ILD exacerbation.¹² High-resolution CT provides the most sensitive imaging modality for diffuse ILD. The addition of prone and expiratory images are helpful in differentiating mild lung disease from atelectasis and detecting air trapping, respectively.¹³ However, since pulmonary embolism is a common extraparenchymal finding routinely considered in the differential of a patient presenting with a known or suspected ILD, physicians should consider ordering a CT pulmonary angiogram with additional high-resolution images. Most important, radiographic evaluation should include a review of all available prior chest imaging to assess both the tempo and the nature of radiographic findings.

Bronchoscopy

Bronchoscopy (with bronchoalveolar lavage [BAL], transbronchial lung biopsy [TBLB] and/or transbronchial needle aspiration [TBNA]) is not a routinely used diagnostic tool in the hospitalized ILD patient. However, it should be considered in certain circumstances.⁷ Cell count and differential can be helpful in diagnosing AEP (greater than 40% eosinophilia) or acute HP (greater than 50% lymphocytosis), while the addition of microbiologic and cytologic analysis can assist with the diagnosis of infectious etiologies (including pneumocystis pneumonia) or malignancy.^14,15 Bronchoscopy with BAL has limited sensitivity for many infections and the procedure is associated with a small risk of worsened hypoxemia. Transbronchial lung biopsy, and to a lesser extent TBNA, carry the added risk of pneumothorax and bleeding. In the majority of cases of ILD, TBLB and TBNA have limited diagnostic utility given the small amount of lung tissue sampled. In cases of suspected IPF, where the identification of the histologic pattern is needed for definitive diagnosis, tissue from TBLB cannot be used to make a conclusive diagnosis.^16,17 However, both TBNA and TBLB are useful in the diagnosis of granulomatous disorders, such as sarcoidosis, where the diagnostic yield ranges from 80% to 90% and 50% to 75%, respectively.^18,19

A newer bronchoscopic approach to sampling the lung using a bronchoscopically-placed cryoprobe (termed transbronchial cryobiopsy) has uncertain diagnostic utility and safety in the acute setting. This procedure involves intubation, sedation, and bronchoscopy allowing for the passage of an endobronchial cryoprobe through the bronchoscope and into the periphery of the lung. Several cryobiopies are generally taken from the same pulmonary subsegment. Despite a large number of recent publications on this topic, none of them have provided a clear sense of the diagnostic yield and safety.^20,21 Transbronchial cryobiopsy remains a highly controversial procedure in the clinical setting, and we would not recommend its use until further data are available.²²

Surgical Lung Biopsy

In the outpatient setting, a surgical lung biopsy is often useful when the ILD diagnosis cannot be made from the clinical context and imaging. However, patients presenting with acute respiratory failure from ILD are at greatly increased risk of complications from nonelective biopsy including pneumothorax, hemothorax, acute exacerbation of ILD, ICU admission, mechanical ventilation, and in-hospital mortality.^23,24 Acute histological findings can also make it difficult to appreciate the underlying pattern of fibrosis, reducing the diagnostic utility.^25-27 In our experience, surgical lung biopsy rarely alters the treatment of ILD patients presenting in acute respiratory failure. We believe that surgical lung biopsy should be reserved for the rare hospitalized patients in whom the clinician believes the results would clearly change management and that the substantial risk is worth taking.^5,28

The inpatient management of ILD is a large topic and difficult to comprehensively cover in a single review. Therefore, in this section, we will review 6 key management questions that address both general and specific treatment decisions that frequently arise in the care of hospitalized ILD patients (Figure).

When should hospitalized ILD patients be treated with antibiotics?

Infection and acute presentations of ILD have many similar clinical and radiographic features, making it difficult to distinguish between the two, or exclude infection as the causative role in an acute exacerbation.² In many ILD patients, the risk of infection is higher than in the general population, due to the acute and chronic use of immunosuppression. Until firm guidelines on the use of antibiotics in hospitalized patients with acute respiratory symptoms are available, we recommend considering the empiric use of antibiotics in ILD patients in respiratory failure, in addition to a thorough infectious workup.

When should hospitalized ILD patients be treated with corticosteroids?

Clinical experience supports the use of corticosteroids in the acute management of most rapidly progressive ILDs presenting with respiratory failure, including AEP, COP, acute HP, drug-induced ILD, and some cases of CT-ILD. Patients with AEP tend to respond rapidly to corticosteroids. In a series of 137 patients with AEP, 127 (92%) received corticosteroids, with defervescence and improved dyspnea within 48 to 72 hours and resolution of all symptoms after a median of 7 (4 to 10) days.²⁹ Cryptogenic organizing pneumonia is similarly corticosteroid-responsive, with patients typically started on doses of 1mg/kg of prednisone followed by a slow taper due to the risk of relapse.³⁰ For the majority of acute CT-ILD, oral prednisone is the initial treatment, often in combination with a second immunosuppressive agent such as mycophenolate.

No proven therapies are available for acute exacerbations of IPF (AE-IPF), including the use of corticosteroids. The most recent international guidelines on the management of AE-IPF conditionally recommends the use of corticosteroids, although this recommendation is largely based on anecdotal reports and clearly states that randomized studies are needed.³ When corticosteroids are used, we recommend high doses (eg, 1 to 2 mg/kg of prednisone) with close clinical monitoring. Consider stopping corticosteroids after 3 to 5 days if there is no evidence of clinical improvement. Prolonged courses of corticosteroids should be avoided.

What additional pharmacologic therapies should be considered in the treatment of hospitalized ILD patients?

Immunomodulators. Patients presenting acutely with a new-onset ILD or with an acute exacerbation of a chronic ILD often receive corticosteroids, sometimes in concert with an immunomodulator. This is most commonly seen in the acute management of CTD- ILD and in chronic HP, where mycophenolate mofetil, and to a lesser extent, cyclophosphamide and azathioprine for CT-ILD are used in combination with corticosteroids. The rationale for this is both therapeutic synergy and a desire to limit the long-term exposure to corticosteroids. Similarly, multiple observational cohort studies have investigated the role of combination or tandem immunosuppression in the treatment AE-IPF. Although cyclosporine, cyclophosphamide, azathioprine, rituximab and tacrolimus have all been studied, their efficacy remains uncertain.³ Until these therapies are better studied, they have no routine role in the management of AE-IPF.

Antifibrotics. Nintedanib and pirfenidone are 2 antifibrotic agents approved for the treatment of IPF. Clinical trials suggest that, in addition to slowing disease progression, these therapies may help prevent AE-IPF. The data are most robust in studies of nintedanib. A phase 2 trial with 432 subjects demonstrated a delay in time to the first investigator-reported acute exacerbation.³¹ Two follow-up phase 3 trials showed a reduction in centrally adjudicated AE-IPF in the pooled nintedanib groups compared to placebo.³² An initial phase 2 trial of pirfenidone showed a reduction in acute exacerbations in patients on pirfenidone, but this finding was not replicated in follow-up studies.^33-35 Because of their potential role in preventing acute exacerbations and emerging evidence to suggest that continuation of antifibrotics may lead to better outcomes during an acute exacerbation, these drugs should not generally be stopped during a hospitalization for ILD. However, no evidence supports their initiation during acute exacerbations, and we do not recommend starting antifibrotics in the hospitalized setting for newly diagnosed patients. Starting and stopping antifibrotics should be reserved for outpatient management.

When should noninvasive and mechanical ventilation be considered?

We recommend carefully considering the use of noninvasive ventilation (NIV) and intubation in every ILD patient in respiratory distress, as an acutely reversible process may be present. In patients requiring mechanical ventilation, every effort should be made to minimize potential damage by reducing the fraction of inspired oxygen (to prevent potential hyperoxic injury) and reducing tidal volumes (to minimize barotrauma). Patients with a chronic ILD, particularly IPF, who require NIV or mechanical ventilation will generally have poor outcomes.

Studies suggest that NIV prevents mechanical ventilation in only the minority of patients presenting with an AE-IPF and is associated with high in-hospital mortality and a median survival following hospital discharge of only 60 days.^36-38 The majority of patients with IPF requiring mechanical ventilation will not survive the intensive care unit. In a series of 23 patients presenting with acute respiratory failure and IPF, 22 of the 23 patients died while receiving mechanical ventilation, with a median survival of 3 days. In a more recent study of 34 patients with acute respiratory failure and IPF, 15 subjects underwent mechanical ventilation with an in-hospital mortality rate of 100%.³⁹ Given the overall poor survival associated with AE-IPF, mechanical ventilation should be carefully considered with the patient and family as part of an overall goals-of-care conversation prior to initiation.

When should hospitalized ILD patients be referred for inpatient lung transplant evaluation?

A subset of hospitalized patients with ILD will not respond to supportive and pharmacologic care, particularly those with advanced lung fibrosis. In these cases, lung transplantation may be the only remaining treatment option. This is particularly true for patients presenting with IPF, and it is 1 of the most common indications for lung transplantation. Patients with respiratory failure and ILD should be evaluated early in the hospital course for transplantation or considered for transfer to a transplant center. General contraindications to transplant are age older than 70 years, underweight or elevated BMI (generally higher than 30), malignancy within the last 2 years (with the exception of cutaneous squamous and basal cell tumors), untreatable major organ dysfunction other than the lung, noncurable chronic extrapulmonary infection (chronic active viral hepatitis B, hepatitis C, human immunodeficiency virus), significant chest wall deformity, untreatable psychiatric or psychologic disease, substance addiction within the last 6 months, or lack of dependable social support.⁴⁰ In select patients with ILD and gas exchange abnormalities, mechanical ventilation or extracorporeal membrane oxygenation may be used to bridge a patient to lung transplantation.⁴¹

What should you tell your ILD patient to expect at discharge?

Accurate diagnosis is important not only for acute inpatient management, but for informing long-term prognosis. Acute-onset ILD tends to be more reversible, to be responsive to medical therapy, and to have a more favorable overall outcome. On the other hand, acute exacerbations of established ILD, particularly IPF, can have a more unfavorable and treatment-refractory course. Once a diagnosis is established, it is important both to provide patients with information and ensure appropriate outpatient follow-up. The Pulmonary Fibrosis Foundation (the largest U.S. advocacy and support organization for patients with ILD) provides information on ILD to patients and families and can serve as an important educational source.⁴² Prior to discharge, it is important to evaluate the oxygen needs of patients at rest and with exertion. Referral to an ILD center at discharge is important whenever possible, to monitor clinical symptoms and lung function, initiate or assess response to treatment, and provide supportive care, including oxygen therapy, pulmonary rehabilitation, and outpatient lung transplant referral.

ILD is a group of heterogeneous disorders characterized by lung inflammation and fibrosis. Although the onset of disease is typically insidious, patients can present acutely requiring hospitalization. Inpatient management varies significantly depending on ILD subtype, and, therefore, accurate diagnosis is key in determining treatment and prognosis. As we develop an improved understanding of the mechanisms of acute presentations of ILD, and our approaches to detection and treatment improve as a result of clinical trials, we anticipate continued modifications to this shared framework.

Disclosure

Dr. Collard reports personal fees from Alkermes, aTyr Pharmaceuticals, Bayer, Boehringer Ingelheim, Bristol-Myers Squibb, Global Blood Therapeutics, Genoa, ImmuneWorks, Moerae Matrix, Navitor, Parexel, Patara, Pharma Capital Partners, Prometic, Takeda, Toray, and Xfibra, outside the submitted work. Drs. Farrand and Shah report no financial conflicts of interest.

Interstitial lung disease (ILD) encompasses a diverse group of disorders that cause inflammation and fibrosis of the lung parenchyma. The clinical manifestations, disease course, management and prognosis of ILD vary depending on the underlying subtype, making accurate classification and diagnosis an important initial step. While a comprehensive list of ILD contains dozens of disorders, the majority of patients will fall into 1 of 3 categories: exposure-related ILD, connective tissue disease-related ILD (CT-ILD), and the idiopathic interstitial pneumonias (Table).

An essential first step in the evaluation of every hospitalized patient with ILD is establishing a diagnosis. A common mistake among clinicians who diagnose patients with ILD is not realizing that ILD is a collection of diseases with different etiologies, natural histories, and treatments. A careful evaluation should be performed in every hospitalized patient with ILD to ensure an accurate diagnosis, ideally in the context of a multidisciplinary conference with pulmonary, radiology, pathology, and other specialties, as appropriate. A multidisciplinary panel of the American Thoracic Society/European Respiratory Society recently published a revised classification of ILD based on a combination of clinical, radiologic, and histopathologic findings, which may aid in refining the diagnosis.¹

There are 3 main scenarios in which the hospital physician will encounter patients with ILD.

Acute presentation of new-onset disease. While many ILDs present insidiously, some cases present acutely and require hospitalization. The most common of these are acute hypersensitivity pneumonitis (HP), CT-ILD (in particular, myositis-related and systemic lupus erythematosus-related), drug-induced ILD (eg, amiodarone, nitrofurantoin), cryptogenic organizing pneumonia (COP), acute eosinophilic pneumonia (AEP), and acute interstitial pneumonia (AIP).

Acute presentation of established (chronic) disease. Patients with chronic forms of ILD can present to the hospital with an acute exacerbation of disease. This can be caused by extra-parenchymal complications, including pulmonary embolism, pneumothorax, and pleural effusion; parenchymal complications such as infectious pneumonia, aspiration pneumonitis, and congestive heart failure; or without an identifiable cause. This latter presentation is most commonly seen in idiopathic pulmonary fibrosis (IPF).^2,3

Elective hospitalization for diagnostic surgical lung biopsy. Patients with ILD may be hospitalized electively for a laparoscopic surgical lung biopsy as part of their diagnostic evaluations.

Physicians caring for a hospitalized ILD patient must be familiar with the clinical presentations, diagnostic approach, medical management, and outpatient follow-up recommended in these 3 settings. We will summarize these areas and provide answers to commonly encountered clinical questions in the hospitalized patient with ILD.

Acute onset (or worsening) of dyspnea is the primary presenting symptom in most patients hospitalized for ILD. This symptom should be further characterized by assessing the degree of dyspnea and the extent of exercise limitation, as both impact overall disease severity and prognosis.⁴ Cough is the second most common symptom, and can be nonproductive, as is common in IPF, or be associated with secretions if parenchymal infection or acute bronchitis is present.⁵ Pleuritic chest pain, pleural effusion, and/or the presence of extrapulmonary features, including dysphagia, joint pain and swelling, or cutaneous thickening may suggest the presence of a CT-ILD. Because most forms of ILD present with only nonspecific symptoms, a careful history and physical examination are essential.

History

A comprehensive patient history is the backbone of diagnosing any ILD. History-taking should focus on severity and temporal progression of symptoms, presence of pre-existing systemic conditions associated with ILD, symptoms of extrapulmonary disease, and exposures to substances that can cause pulmonary injury, including a detailed history of occupations and hobbies, medications, smoking, and familial lung disease.^6-9 Physicians must try to exclude other diagnoses that could result in a similar acute presentation, including congestive heart failure and infection. Considering the complex and extensive recommended history-taking, physicians may find it helpful to use a standardized questionnaire, as provided by the American College of Chest Physicians.¹⁰

Laboratory Testing

All patients presenting to the hospital with a suspected ILD should undergo careful assessment for the presence of connective tissue disease, including patients without clear symptoms because ILD can be the presenting manifestation. We routinely test for antinuclear antibody titer and pattern, rheumatoid factor, anticyclic citrullinated peptide, creatinine kinase, and aldolase as the initial screening panel in most patients, with further testing directed by the findings on history and physical examination. Pulmonary function tests are used routinely to monitor disease progression in the outpatient setting; however, in the hospitalized ILD patient, they are often difficult to perform and have no real diagnostic value. Similarly, arterial blood gas is not routinely used as part of the initial inpatient evaluation.

Imaging

All hospitalized patients with a known or suspected ILD should undergo chest imaging, assuming they are stable enough to do so. While the chest radiograph can provide a low-cost initial assessment of the degree of lung involvement and presence of accompanying abnormalities, computed tomography (CT) scanning is the diagnostic test of choice.¹¹ The pattern and distribution of abnormalities on CT scan can greatly assist with the differential diagnosis in patients presenting with a new ILD, while the presence and pattern of new opacities superimposed on chronic changes can inform the differential and the prognosis of an ILD exacerbation.¹² High-resolution CT provides the most sensitive imaging modality for diffuse ILD. The addition of prone and expiratory images are helpful in differentiating mild lung disease from atelectasis and detecting air trapping, respectively.¹³ However, since pulmonary embolism is a common extraparenchymal finding routinely considered in the differential of a patient presenting with a known or suspected ILD, physicians should consider ordering a CT pulmonary angiogram with additional high-resolution images. Most important, radiographic evaluation should include a review of all available prior chest imaging to assess both the tempo and the nature of radiographic findings.

Bronchoscopy

Bronchoscopy (with bronchoalveolar lavage [BAL], transbronchial lung biopsy [TBLB] and/or transbronchial needle aspiration [TBNA]) is not a routinely used diagnostic tool in the hospitalized ILD patient. However, it should be considered in certain circumstances.⁷ Cell count and differential can be helpful in diagnosing AEP (greater than 40% eosinophilia) or acute HP (greater than 50% lymphocytosis), while the addition of microbiologic and cytologic analysis can assist with the diagnosis of infectious etiologies (including pneumocystis pneumonia) or malignancy.^14,15 Bronchoscopy with BAL has limited sensitivity for many infections and the procedure is associated with a small risk of worsened hypoxemia. Transbronchial lung biopsy, and to a lesser extent TBNA, carry the added risk of pneumothorax and bleeding. In the majority of cases of ILD, TBLB and TBNA have limited diagnostic utility given the small amount of lung tissue sampled. In cases of suspected IPF, where the identification of the histologic pattern is needed for definitive diagnosis, tissue from TBLB cannot be used to make a conclusive diagnosis.^16,17 However, both TBNA and TBLB are useful in the diagnosis of granulomatous disorders, such as sarcoidosis, where the diagnostic yield ranges from 80% to 90% and 50% to 75%, respectively.^18,19

A newer bronchoscopic approach to sampling the lung using a bronchoscopically-placed cryoprobe (termed transbronchial cryobiopsy) has uncertain diagnostic utility and safety in the acute setting. This procedure involves intubation, sedation, and bronchoscopy allowing for the passage of an endobronchial cryoprobe through the bronchoscope and into the periphery of the lung. Several cryobiopies are generally taken from the same pulmonary subsegment. Despite a large number of recent publications on this topic, none of them have provided a clear sense of the diagnostic yield and safety.^20,21 Transbronchial cryobiopsy remains a highly controversial procedure in the clinical setting, and we would not recommend its use until further data are available.²²

Surgical Lung Biopsy

In the outpatient setting, a surgical lung biopsy is often useful when the ILD diagnosis cannot be made from the clinical context and imaging. However, patients presenting with acute respiratory failure from ILD are at greatly increased risk of complications from nonelective biopsy including pneumothorax, hemothorax, acute exacerbation of ILD, ICU admission, mechanical ventilation, and in-hospital mortality.^23,24 Acute histological findings can also make it difficult to appreciate the underlying pattern of fibrosis, reducing the diagnostic utility.^25-27 In our experience, surgical lung biopsy rarely alters the treatment of ILD patients presenting in acute respiratory failure. We believe that surgical lung biopsy should be reserved for the rare hospitalized patients in whom the clinician believes the results would clearly change management and that the substantial risk is worth taking.^5,28

The inpatient management of ILD is a large topic and difficult to comprehensively cover in a single review. Therefore, in this section, we will review 6 key management questions that address both general and specific treatment decisions that frequently arise in the care of hospitalized ILD patients (Figure).

When should hospitalized ILD patients be treated with antibiotics?

Infection and acute presentations of ILD have many similar clinical and radiographic features, making it difficult to distinguish between the two, or exclude infection as the causative role in an acute exacerbation.² In many ILD patients, the risk of infection is higher than in the general population, due to the acute and chronic use of immunosuppression. Until firm guidelines on the use of antibiotics in hospitalized patients with acute respiratory symptoms are available, we recommend considering the empiric use of antibiotics in ILD patients in respiratory failure, in addition to a thorough infectious workup.

When should hospitalized ILD patients be treated with corticosteroids?

Clinical experience supports the use of corticosteroids in the acute management of most rapidly progressive ILDs presenting with respiratory failure, including AEP, COP, acute HP, drug-induced ILD, and some cases of CT-ILD. Patients with AEP tend to respond rapidly to corticosteroids. In a series of 137 patients with AEP, 127 (92%) received corticosteroids, with defervescence and improved dyspnea within 48 to 72 hours and resolution of all symptoms after a median of 7 (4 to 10) days.²⁹ Cryptogenic organizing pneumonia is similarly corticosteroid-responsive, with patients typically started on doses of 1mg/kg of prednisone followed by a slow taper due to the risk of relapse.³⁰ For the majority of acute CT-ILD, oral prednisone is the initial treatment, often in combination with a second immunosuppressive agent such as mycophenolate.

No proven therapies are available for acute exacerbations of IPF (AE-IPF), including the use of corticosteroids. The most recent international guidelines on the management of AE-IPF conditionally recommends the use of corticosteroids, although this recommendation is largely based on anecdotal reports and clearly states that randomized studies are needed.³ When corticosteroids are used, we recommend high doses (eg, 1 to 2 mg/kg of prednisone) with close clinical monitoring. Consider stopping corticosteroids after 3 to 5 days if there is no evidence of clinical improvement. Prolonged courses of corticosteroids should be avoided.

What additional pharmacologic therapies should be considered in the treatment of hospitalized ILD patients?

Immunomodulators. Patients presenting acutely with a new-onset ILD or with an acute exacerbation of a chronic ILD often receive corticosteroids, sometimes in concert with an immunomodulator. This is most commonly seen in the acute management of CTD- ILD and in chronic HP, where mycophenolate mofetil, and to a lesser extent, cyclophosphamide and azathioprine for CT-ILD are used in combination with corticosteroids. The rationale for this is both therapeutic synergy and a desire to limit the long-term exposure to corticosteroids. Similarly, multiple observational cohort studies have investigated the role of combination or tandem immunosuppression in the treatment AE-IPF. Although cyclosporine, cyclophosphamide, azathioprine, rituximab and tacrolimus have all been studied, their efficacy remains uncertain.³ Until these therapies are better studied, they have no routine role in the management of AE-IPF.

Antifibrotics. Nintedanib and pirfenidone are 2 antifibrotic agents approved for the treatment of IPF. Clinical trials suggest that, in addition to slowing disease progression, these therapies may help prevent AE-IPF. The data are most robust in studies of nintedanib. A phase 2 trial with 432 subjects demonstrated a delay in time to the first investigator-reported acute exacerbation.³¹ Two follow-up phase 3 trials showed a reduction in centrally adjudicated AE-IPF in the pooled nintedanib groups compared to placebo.³² An initial phase 2 trial of pirfenidone showed a reduction in acute exacerbations in patients on pirfenidone, but this finding was not replicated in follow-up studies.^33-35 Because of their potential role in preventing acute exacerbations and emerging evidence to suggest that continuation of antifibrotics may lead to better outcomes during an acute exacerbation, these drugs should not generally be stopped during a hospitalization for ILD. However, no evidence supports their initiation during acute exacerbations, and we do not recommend starting antifibrotics in the hospitalized setting for newly diagnosed patients. Starting and stopping antifibrotics should be reserved for outpatient management.

When should noninvasive and mechanical ventilation be considered?

We recommend carefully considering the use of noninvasive ventilation (NIV) and intubation in every ILD patient in respiratory distress, as an acutely reversible process may be present. In patients requiring mechanical ventilation, every effort should be made to minimize potential damage by reducing the fraction of inspired oxygen (to prevent potential hyperoxic injury) and reducing tidal volumes (to minimize barotrauma). Patients with a chronic ILD, particularly IPF, who require NIV or mechanical ventilation will generally have poor outcomes.

Studies suggest that NIV prevents mechanical ventilation in only the minority of patients presenting with an AE-IPF and is associated with high in-hospital mortality and a median survival following hospital discharge of only 60 days.^36-38 The majority of patients with IPF requiring mechanical ventilation will not survive the intensive care unit. In a series of 23 patients presenting with acute respiratory failure and IPF, 22 of the 23 patients died while receiving mechanical ventilation, with a median survival of 3 days. In a more recent study of 34 patients with acute respiratory failure and IPF, 15 subjects underwent mechanical ventilation with an in-hospital mortality rate of 100%.³⁹ Given the overall poor survival associated with AE-IPF, mechanical ventilation should be carefully considered with the patient and family as part of an overall goals-of-care conversation prior to initiation.

When should hospitalized ILD patients be referred for inpatient lung transplant evaluation?

A subset of hospitalized patients with ILD will not respond to supportive and pharmacologic care, particularly those with advanced lung fibrosis. In these cases, lung transplantation may be the only remaining treatment option. This is particularly true for patients presenting with IPF, and it is 1 of the most common indications for lung transplantation. Patients with respiratory failure and ILD should be evaluated early in the hospital course for transplantation or considered for transfer to a transplant center. General contraindications to transplant are age older than 70 years, underweight or elevated BMI (generally higher than 30), malignancy within the last 2 years (with the exception of cutaneous squamous and basal cell tumors), untreatable major organ dysfunction other than the lung, noncurable chronic extrapulmonary infection (chronic active viral hepatitis B, hepatitis C, human immunodeficiency virus), significant chest wall deformity, untreatable psychiatric or psychologic disease, substance addiction within the last 6 months, or lack of dependable social support.⁴⁰ In select patients with ILD and gas exchange abnormalities, mechanical ventilation or extracorporeal membrane oxygenation may be used to bridge a patient to lung transplantation.⁴¹

What should you tell your ILD patient to expect at discharge?

Accurate diagnosis is important not only for acute inpatient management, but for informing long-term prognosis. Acute-onset ILD tends to be more reversible, to be responsive to medical therapy, and to have a more favorable overall outcome. On the other hand, acute exacerbations of established ILD, particularly IPF, can have a more unfavorable and treatment-refractory course. Once a diagnosis is established, it is important both to provide patients with information and ensure appropriate outpatient follow-up. The Pulmonary Fibrosis Foundation (the largest U.S. advocacy and support organization for patients with ILD) provides information on ILD to patients and families and can serve as an important educational source.⁴² Prior to discharge, it is important to evaluate the oxygen needs of patients at rest and with exertion. Referral to an ILD center at discharge is important whenever possible, to monitor clinical symptoms and lung function, initiate or assess response to treatment, and provide supportive care, including oxygen therapy, pulmonary rehabilitation, and outpatient lung transplant referral.

ILD is a group of heterogeneous disorders characterized by lung inflammation and fibrosis. Although the onset of disease is typically insidious, patients can present acutely requiring hospitalization. Inpatient management varies significantly depending on ILD subtype, and, therefore, accurate diagnosis is key in determining treatment and prognosis. As we develop an improved understanding of the mechanisms of acute presentations of ILD, and our approaches to detection and treatment improve as a result of clinical trials, we anticipate continued modifications to this shared framework.

Disclosure

Dr. Collard reports personal fees from Alkermes, aTyr Pharmaceuticals, Bayer, Boehringer Ingelheim, Bristol-Myers Squibb, Global Blood Therapeutics, Genoa, ImmuneWorks, Moerae Matrix, Navitor, Parexel, Patara, Pharma Capital Partners, Prometic, Takeda, Toray, and Xfibra, outside the submitted work. Drs. Farrand and Shah report no financial conflicts of interest.

Interstitial lung disease (ILD) encompasses a diverse group of disorders that cause inflammation and fibrosis of the lung parenchyma. The clinical manifestations, disease course, management and prognosis of ILD vary depending on the underlying subtype, making accurate classification and diagnosis an important initial step. While a comprehensive list of ILD contains dozens of disorders, the majority of patients will fall into 1 of 3 categories: exposure-related ILD, connective tissue disease-related ILD (CT-ILD), and the idiopathic interstitial pneumonias (Table).

An essential first step in the evaluation of every hospitalized patient with ILD is establishing a diagnosis. A common mistake among clinicians who diagnose patients with ILD is not realizing that ILD is a collection of diseases with different etiologies, natural histories, and treatments. A careful evaluation should be performed in every hospitalized patient with ILD to ensure an accurate diagnosis, ideally in the context of a multidisciplinary conference with pulmonary, radiology, pathology, and other specialties, as appropriate. A multidisciplinary panel of the American Thoracic Society/European Respiratory Society recently published a revised classification of ILD based on a combination of clinical, radiologic, and histopathologic findings, which may aid in refining the diagnosis.¹

There are 3 main scenarios in which the hospital physician will encounter patients with ILD.

Acute presentation of new-onset disease. While many ILDs present insidiously, some cases present acutely and require hospitalization. The most common of these are acute hypersensitivity pneumonitis (HP), CT-ILD (in particular, myositis-related and systemic lupus erythematosus-related), drug-induced ILD (eg, amiodarone, nitrofurantoin), cryptogenic organizing pneumonia (COP), acute eosinophilic pneumonia (AEP), and acute interstitial pneumonia (AIP).

Acute presentation of established (chronic) disease. Patients with chronic forms of ILD can present to the hospital with an acute exacerbation of disease. This can be caused by extra-parenchymal complications, including pulmonary embolism, pneumothorax, and pleural effusion; parenchymal complications such as infectious pneumonia, aspiration pneumonitis, and congestive heart failure; or without an identifiable cause. This latter presentation is most commonly seen in idiopathic pulmonary fibrosis (IPF).^2,3

Elective hospitalization for diagnostic surgical lung biopsy. Patients with ILD may be hospitalized electively for a laparoscopic surgical lung biopsy as part of their diagnostic evaluations.

Physicians caring for a hospitalized ILD patient must be familiar with the clinical presentations, diagnostic approach, medical management, and outpatient follow-up recommended in these 3 settings. We will summarize these areas and provide answers to commonly encountered clinical questions in the hospitalized patient with ILD.

Acute onset (or worsening) of dyspnea is the primary presenting symptom in most patients hospitalized for ILD. This symptom should be further characterized by assessing the degree of dyspnea and the extent of exercise limitation, as both impact overall disease severity and prognosis.⁴ Cough is the second most common symptom, and can be nonproductive, as is common in IPF, or be associated with secretions if parenchymal infection or acute bronchitis is present.⁵ Pleuritic chest pain, pleural effusion, and/or the presence of extrapulmonary features, including dysphagia, joint pain and swelling, or cutaneous thickening may suggest the presence of a CT-ILD. Because most forms of ILD present with only nonspecific symptoms, a careful history and physical examination are essential.

History

A comprehensive patient history is the backbone of diagnosing any ILD. History-taking should focus on severity and temporal progression of symptoms, presence of pre-existing systemic conditions associated with ILD, symptoms of extrapulmonary disease, and exposures to substances that can cause pulmonary injury, including a detailed history of occupations and hobbies, medications, smoking, and familial lung disease.^6-9 Physicians must try to exclude other diagnoses that could result in a similar acute presentation, including congestive heart failure and infection. Considering the complex and extensive recommended history-taking, physicians may find it helpful to use a standardized questionnaire, as provided by the American College of Chest Physicians.¹⁰

Laboratory Testing

All patients presenting to the hospital with a suspected ILD should undergo careful assessment for the presence of connective tissue disease, including patients without clear symptoms because ILD can be the presenting manifestation. We routinely test for antinuclear antibody titer and pattern, rheumatoid factor, anticyclic citrullinated peptide, creatinine kinase, and aldolase as the initial screening panel in most patients, with further testing directed by the findings on history and physical examination. Pulmonary function tests are used routinely to monitor disease progression in the outpatient setting; however, in the hospitalized ILD patient, they are often difficult to perform and have no real diagnostic value. Similarly, arterial blood gas is not routinely used as part of the initial inpatient evaluation.

Imaging

All hospitalized patients with a known or suspected ILD should undergo chest imaging, assuming they are stable enough to do so. While the chest radiograph can provide a low-cost initial assessment of the degree of lung involvement and presence of accompanying abnormalities, computed tomography (CT) scanning is the diagnostic test of choice.¹¹ The pattern and distribution of abnormalities on CT scan can greatly assist with the differential diagnosis in patients presenting with a new ILD, while the presence and pattern of new opacities superimposed on chronic changes can inform the differential and the prognosis of an ILD exacerbation.¹² High-resolution CT provides the most sensitive imaging modality for diffuse ILD. The addition of prone and expiratory images are helpful in differentiating mild lung disease from atelectasis and detecting air trapping, respectively.¹³ However, since pulmonary embolism is a common extraparenchymal finding routinely considered in the differential of a patient presenting with a known or suspected ILD, physicians should consider ordering a CT pulmonary angiogram with additional high-resolution images. Most important, radiographic evaluation should include a review of all available prior chest imaging to assess both the tempo and the nature of radiographic findings.

Bronchoscopy

Bronchoscopy (with bronchoalveolar lavage [BAL], transbronchial lung biopsy [TBLB] and/or transbronchial needle aspiration [TBNA]) is not a routinely used diagnostic tool in the hospitalized ILD patient. However, it should be considered in certain circumstances.⁷ Cell count and differential can be helpful in diagnosing AEP (greater than 40% eosinophilia) or acute HP (greater than 50% lymphocytosis), while the addition of microbiologic and cytologic analysis can assist with the diagnosis of infectious etiologies (including pneumocystis pneumonia) or malignancy.^14,15 Bronchoscopy with BAL has limited sensitivity for many infections and the procedure is associated with a small risk of worsened hypoxemia. Transbronchial lung biopsy, and to a lesser extent TBNA, carry the added risk of pneumothorax and bleeding. In the majority of cases of ILD, TBLB and TBNA have limited diagnostic utility given the small amount of lung tissue sampled. In cases of suspected IPF, where the identification of the histologic pattern is needed for definitive diagnosis, tissue from TBLB cannot be used to make a conclusive diagnosis.^16,17 However, both TBNA and TBLB are useful in the diagnosis of granulomatous disorders, such as sarcoidosis, where the diagnostic yield ranges from 80% to 90% and 50% to 75%, respectively.^18,19

A newer bronchoscopic approach to sampling the lung using a bronchoscopically-placed cryoprobe (termed transbronchial cryobiopsy) has uncertain diagnostic utility and safety in the acute setting. This procedure involves intubation, sedation, and bronchoscopy allowing for the passage of an endobronchial cryoprobe through the bronchoscope and into the periphery of the lung. Several cryobiopies are generally taken from the same pulmonary subsegment. Despite a large number of recent publications on this topic, none of them have provided a clear sense of the diagnostic yield and safety.^20,21 Transbronchial cryobiopsy remains a highly controversial procedure in the clinical setting, and we would not recommend its use until further data are available.²²

Surgical Lung Biopsy

In the outpatient setting, a surgical lung biopsy is often useful when the ILD diagnosis cannot be made from the clinical context and imaging. However, patients presenting with acute respiratory failure from ILD are at greatly increased risk of complications from nonelective biopsy including pneumothorax, hemothorax, acute exacerbation of ILD, ICU admission, mechanical ventilation, and in-hospital mortality.^23,24 Acute histological findings can also make it difficult to appreciate the underlying pattern of fibrosis, reducing the diagnostic utility.^25-27 In our experience, surgical lung biopsy rarely alters the treatment of ILD patients presenting in acute respiratory failure. We believe that surgical lung biopsy should be reserved for the rare hospitalized patients in whom the clinician believes the results would clearly change management and that the substantial risk is worth taking.^5,28

The inpatient management of ILD is a large topic and difficult to comprehensively cover in a single review. Therefore, in this section, we will review 6 key management questions that address both general and specific treatment decisions that frequently arise in the care of hospitalized ILD patients (Figure).

When should hospitalized ILD patients be treated with antibiotics?

Infection and acute presentations of ILD have many similar clinical and radiographic features, making it difficult to distinguish between the two, or exclude infection as the causative role in an acute exacerbation.² In many ILD patients, the risk of infection is higher than in the general population, due to the acute and chronic use of immunosuppression. Until firm guidelines on the use of antibiotics in hospitalized patients with acute respiratory symptoms are available, we recommend considering the empiric use of antibiotics in ILD patients in respiratory failure, in addition to a thorough infectious workup.

When should hospitalized ILD patients be treated with corticosteroids?

Clinical experience supports the use of corticosteroids in the acute management of most rapidly progressive ILDs presenting with respiratory failure, including AEP, COP, acute HP, drug-induced ILD, and some cases of CT-ILD. Patients with AEP tend to respond rapidly to corticosteroids. In a series of 137 patients with AEP, 127 (92%) received corticosteroids, with defervescence and improved dyspnea within 48 to 72 hours and resolution of all symptoms after a median of 7 (4 to 10) days.²⁹ Cryptogenic organizing pneumonia is similarly corticosteroid-responsive, with patients typically started on doses of 1mg/kg of prednisone followed by a slow taper due to the risk of relapse.³⁰ For the majority of acute CT-ILD, oral prednisone is the initial treatment, often in combination with a second immunosuppressive agent such as mycophenolate.

No proven therapies are available for acute exacerbations of IPF (AE-IPF), including the use of corticosteroids. The most recent international guidelines on the management of AE-IPF conditionally recommends the use of corticosteroids, although this recommendation is largely based on anecdotal reports and clearly states that randomized studies are needed.³ When corticosteroids are used, we recommend high doses (eg, 1 to 2 mg/kg of prednisone) with close clinical monitoring. Consider stopping corticosteroids after 3 to 5 days if there is no evidence of clinical improvement. Prolonged courses of corticosteroids should be avoided.

What additional pharmacologic therapies should be considered in the treatment of hospitalized ILD patients?

Immunomodulators. Patients presenting acutely with a new-onset ILD or with an acute exacerbation of a chronic ILD often receive corticosteroids, sometimes in concert with an immunomodulator. This is most commonly seen in the acute management of CTD- ILD and in chronic HP, where mycophenolate mofetil, and to a lesser extent, cyclophosphamide and azathioprine for CT-ILD are used in combination with corticosteroids. The rationale for this is both therapeutic synergy and a desire to limit the long-term exposure to corticosteroids. Similarly, multiple observational cohort studies have investigated the role of combination or tandem immunosuppression in the treatment AE-IPF. Although cyclosporine, cyclophosphamide, azathioprine, rituximab and tacrolimus have all been studied, their efficacy remains uncertain.³ Until these therapies are better studied, they have no routine role in the management of AE-IPF.

Antifibrotics. Nintedanib and pirfenidone are 2 antifibrotic agents approved for the treatment of IPF. Clinical trials suggest that, in addition to slowing disease progression, these therapies may help prevent AE-IPF. The data are most robust in studies of nintedanib. A phase 2 trial with 432 subjects demonstrated a delay in time to the first investigator-reported acute exacerbation.³¹ Two follow-up phase 3 trials showed a reduction in centrally adjudicated AE-IPF in the pooled nintedanib groups compared to placebo.³² An initial phase 2 trial of pirfenidone showed a reduction in acute exacerbations in patients on pirfenidone, but this finding was not replicated in follow-up studies.^33-35 Because of their potential role in preventing acute exacerbations and emerging evidence to suggest that continuation of antifibrotics may lead to better outcomes during an acute exacerbation, these drugs should not generally be stopped during a hospitalization for ILD. However, no evidence supports their initiation during acute exacerbations, and we do not recommend starting antifibrotics in the hospitalized setting for newly diagnosed patients. Starting and stopping antifibrotics should be reserved for outpatient management.

When should noninvasive and mechanical ventilation be considered?

We recommend carefully considering the use of noninvasive ventilation (NIV) and intubation in every ILD patient in respiratory distress, as an acutely reversible process may be present. In patients requiring mechanical ventilation, every effort should be made to minimize potential damage by reducing the fraction of inspired oxygen (to prevent potential hyperoxic injury) and reducing tidal volumes (to minimize barotrauma). Patients with a chronic ILD, particularly IPF, who require NIV or mechanical ventilation will generally have poor outcomes.

Studies suggest that NIV prevents mechanical ventilation in only the minority of patients presenting with an AE-IPF and is associated with high in-hospital mortality and a median survival following hospital discharge of only 60 days.^36-38 The majority of patients with IPF requiring mechanical ventilation will not survive the intensive care unit. In a series of 23 patients presenting with acute respiratory failure and IPF, 22 of the 23 patients died while receiving mechanical ventilation, with a median survival of 3 days. In a more recent study of 34 patients with acute respiratory failure and IPF, 15 subjects underwent mechanical ventilation with an in-hospital mortality rate of 100%.³⁹ Given the overall poor survival associated with AE-IPF, mechanical ventilation should be carefully considered with the patient and family as part of an overall goals-of-care conversation prior to initiation.

When should hospitalized ILD patients be referred for inpatient lung transplant evaluation?

A subset of hospitalized patients with ILD will not respond to supportive and pharmacologic care, particularly those with advanced lung fibrosis. In these cases, lung transplantation may be the only remaining treatment option. This is particularly true for patients presenting with IPF, and it is 1 of the most common indications for lung transplantation. Patients with respiratory failure and ILD should be evaluated early in the hospital course for transplantation or considered for transfer to a transplant center. General contraindications to transplant are age older than 70 years, underweight or elevated BMI (generally higher than 30), malignancy within the last 2 years (with the exception of cutaneous squamous and basal cell tumors), untreatable major organ dysfunction other than the lung, noncurable chronic extrapulmonary infection (chronic active viral hepatitis B, hepatitis C, human immunodeficiency virus), significant chest wall deformity, untreatable psychiatric or psychologic disease, substance addiction within the last 6 months, or lack of dependable social support.⁴⁰ In select patients with ILD and gas exchange abnormalities, mechanical ventilation or extracorporeal membrane oxygenation may be used to bridge a patient to lung transplantation.⁴¹

What should you tell your ILD patient to expect at discharge?

Accurate diagnosis is important not only for acute inpatient management, but for informing long-term prognosis. Acute-onset ILD tends to be more reversible, to be responsive to medical therapy, and to have a more favorable overall outcome. On the other hand, acute exacerbations of established ILD, particularly IPF, can have a more unfavorable and treatment-refractory course. Once a diagnosis is established, it is important both to provide patients with information and ensure appropriate outpatient follow-up. The Pulmonary Fibrosis Foundation (the largest U.S. advocacy and support organization for patients with ILD) provides information on ILD to patients and families and can serve as an important educational source.⁴² Prior to discharge, it is important to evaluate the oxygen needs of patients at rest and with exertion. Referral to an ILD center at discharge is important whenever possible, to monitor clinical symptoms and lung function, initiate or assess response to treatment, and provide supportive care, including oxygen therapy, pulmonary rehabilitation, and outpatient lung transplant referral.

ILD is a group of heterogeneous disorders characterized by lung inflammation and fibrosis. Although the onset of disease is typically insidious, patients can present acutely requiring hospitalization. Inpatient management varies significantly depending on ILD subtype, and, therefore, accurate diagnosis is key in determining treatment and prognosis. As we develop an improved understanding of the mechanisms of acute presentations of ILD, and our approaches to detection and treatment improve as a result of clinical trials, we anticipate continued modifications to this shared framework.

Disclosure

Dr. Collard reports personal fees from Alkermes, aTyr Pharmaceuticals, Bayer, Boehringer Ingelheim, Bristol-Myers Squibb, Global Blood Therapeutics, Genoa, ImmuneWorks, Moerae Matrix, Navitor, Parexel, Patara, Pharma Capital Partners, Prometic, Takeda, Toray, and Xfibra, outside the submitted work. Drs. Farrand and Shah report no financial conflicts of interest.

The approach to clinical conundrums by an expert clinician is revealed through the presentation of an actual patient’s case in an approach typical of a morning report. Similarly to patient care, sequential pieces of information are provided to the clinician, who is unfamiliar with the case. The focus is on the thought processes of both the clinical team caring for the patient and the discussant.

After losing consciousness at a supermarket, a 70-year-old man was brought to the emergency department by paramedics. He subsequently developed chest pain.

Syncope can be difficult to evaluate, but chest pain may help narrow an otherwise broad differential diagnosis. If this patient has aortic stenosis or hypertrophic cardiomyopathy, effort syncope is the culprit. Cardiac dysrhythmia (eg, ventricular tachycardia), complete heart block, and supraventricular tachycardia each can cause syncope along with chest pain. Myocardial infarction and associated ventricular arrhythmia might also explain both chest pain and syncope. The paramedics might have noted an arrhythmia on the cardiac monitor; if possible, the rhythm strip should be reviewed. A pulmonary embolus can cause chest pain and, if large enough to cause right ventricular compromise, syncope.

According to witnesses at the supermarket, the patient dropped to the ground, lost consciousness, and convulsed for 30 seconds. He had no head trauma, tongue biting, urinary incontinence, or confusion afterward. Electrocardiogram (ECG) performed at the scene showed ST elevations in leads V1 to V3 with ST depressions in the inferior leads. On arrival in the emergency department, the patient described nonradiating substernal chest pressure exacerbated by deep inhalation. The pain did not improve with nitroglycerin. He recalled feeling light-headed before the syncope.

He had not received medical care for 20 years and had no known illnesses other than hypertension. He was not taking any medications. He previously worked as a welder and never smoked tobacco, drank alcohol, or used illicit drugs. The patient’s temperature was 36.4°C. Heart rate was 88 beats per minute, blood pressure 128/72 mm Hg, oxygen saturation 100% on room air, and respiratory rate 22 breaths per minute. The patient had conjunctival pallor. There was a grade 3/6 crescendo- decrescendo systolic murmur loudest at the right upper sternal border without radiation to the carotids. There was no jugular venous distention. Lungs were clear to auscultation bilaterally. There was no peripheral edema, rash, or lymphadenopathy.

Convulsive movements commonly occur during episodes of unconsciousness lasting more than 15 seconds—a phenomenon termed convulsive syncope and often is confused with seizures. These movements are usually clonic jerks of the extremities and trunk and slight twitching of the face, and occasionally tonic extension of the trunk and clenching of the jaw. Absence of tongue biting, urinary incontinence, and confusion in this patient’s case makes seizures less likely.

The distribution of ST segment changes on his ECG are concerning for myocardial infarction in the septal and inferior regions. Right-sided ECG should be performed to assess for right ventricular infarction. Although myocardial ischemia is the primary concern, some features warrant consideration of other etiologies of syncope. First, syncope is an unusual presentation of cardiac ischemia or infarct. The complaint of chest pressure exacerbated by deep inhalation is another atypical feature for myocardial ischemia. Although the patient’s oxygen saturation and heart rate are normal, pulmonary embolism remains a possibility.

The prominent crescendo-decrescendo systolic murmur at the right upper sternal border could indicate aortic stenosis; the carotids should be palpated to assess for pulsus parvus et tardus. A high-flow state associated with anemia could also lead to a midsystolic murmur. Conjunctival pallor typically is seen with hemoglobin levels of 6 g/dL or less. This finding may indicate severe anemia, which has the potential to cause myocardial ischemia and syncope.

Laboratory testing revealed a troponin of 0.04 ng/dL, hemoglobin 4.1 g/dL with MCV of 84.7 fL, white blood cell count 6,500/μL and platelet count 179,000/μL. Serum sodium was 130 mEq/L, urea nitrogen 16 mg/dL, creatinine 1.6 mg/dL, calcium 7.8 mg/dL, total protein 11.4 g/dL (reference range, 6.0-8.2), and albumin 2.2 g/dL. Erythrocyte sedimentation rate (ESR) was 20 mm/h. Serum iron was 48 μg/mL, total iron binding capacity 275 μg/dL, percent iron saturation 17% (reference range, 20-55), and ferritin 10 ng/mL (reference range, 30-400). The international normalized ratio (INR) was 1.5, prothrombin time 15.5 sec (reference range, 9.4-11.6), and partial thromboplastin time 24.7 sec (reference range, 22.9-30.6). Hepatitis C and HIV antibodies were negative as was the urine toxicology screen. Urine protein to creatinine ratio was 0.07. His hemoglobin rose to 7.9 g/dL with transfusion of 4 units packed red blood cells (RBC). His chest pain improved and inferior ST depressions resolved on follow-up ECG. Further history revealed multiple episodes of melena and hematochezia in the preceding weeks without nausea, vomiting, or abdominal pain.

The patient has a strikingly large gamma gap: 9.2. A gap larger than 4 is concerning for the presence of paraproteins. Given the possibility of a paraproteinemia (eg, multiple myeloma, plasmacytoma, Waldenström macroglobulinemia), the first step is to check serum and urine protein electrophoresis. The patient’s anemia is significant and reticulocyte index low. The low ferritin level combined with the inappropriately low reticulocyte count could result from iron deficiency anemia, another bone marrow process, or both. The patient’s syncope likely resulted from severe anemia and hypovolemia associated with hematochezia. The prolonged prothrombin time could be caused by a coagulation factor production problem, from vitamin K deficiency or underlying liver disease, or by a consumptive problem, from low-grade disseminated intravascular coagulation. It is controversial whether inhaling welding fumes causes cancer, but the patient’s age alone makes malignancy a definite possibility.

Aortic stenosis is intriguing in light of the patient’s painless melena and hematochezia. Heyde syndrome is a phenomenon in which high shear stress causes a reduction in the size of von Willebrand factor predisposing to bleeding from submucosal angiodysplasia. However, Heyde syndrome has been reported to occur in the setting of severe or critical aortic stenosis and is unlikely to be the cause here. The patient needs to be examined with both upper and lower endoscopy to rule out gastrointestinal (GI) malignancy, gastric and esophageal varices, and painless peptic ulcers. High right ventricular systolic blood pressures along with echodense material in the inferior vena cava and right atrium suggest the possibility of malignancy with vascular invasion. Tricuspid regurgitation is consistent with high right-sided pressures. As left ventricular ejection fraction is reduced, some of the high right-sided pressures could also be attributable to left heart failure. CT findings of rectal wall thickening and perirectal lymph nodes could be attributable to cancer with locally metastatic disease. Blood loss caused by this cancer would explain the severe anemia on admission as well as the low ferritin level and the iron deficiency anemia. In this 70-year-old man who has not had routine health care maintenance, the leading diagnosis is colorectal cancer. However, the markedly elevated globulin gap and elevated INR strongly suggest another process (eg, multiple myeloma, other paraproteinemia) is also present.

INR remained elevated (1.5) despite vitamin K supplementation. Peripheral blood smear showed hypochromic and normocytic RBCs with moderate rouleaux formation (Figure 1). Intravenous pantoprazole and octreotide were started. Upper and lower endoscopy revealed multiple esophageal erosions and both a polypoid mass and an ulcer within the rectum with evidence of prior bleeding. The mass was resected and the ulcer biopsied.

On hospital  day 3, the patient was transfused another unit of packed RBCs. Hemoglobin level increased from 7.4 g/dL to 8.2 g/dL, but he began to complain of headache, blurred vision, and worsened chest pain. He did not have weakness, numbness, diplopia, dysphagia, or dysarthria. External examination and extraocular movements of both eyes were normal. Visual acuity was 20/25 bilaterally. Funduscopic examination revealed mild dilation of retinal veins and retinal hemorrhages (Figure 2).

Rouleaux formation occurs as excess cathodal proteins, such as immunoglobulins or fibrinogen, adhere to RBCs and cause the cells to stack together in long chains. Classically this is associated with multiple myeloma, but can occur with Waldenström’s macroglobulinemia and other cancers or infections. It can also occur as an artifact in smear preparation but the large globulin gap in this patient supports pathologic rouleaux formation.

Venous retinopathy with hemorrhages may occur with occlusion of the arterial supply (eg, as with carotid artery obstruction) but also with hyperviscosity syndrome (HVS). Vascular disturbances throughout the body play a major role in HVS, but these changes are most easily visualized in the retina. It is interesting that the patient’s headache and blurred vision began after he received additional blood transfusions. Spuriously low hemoglobin and hematocrit levels may stem from increased plasma volume from high immunoglobulin M (IgM) concentrations in Waldenström macroglobulinemia; thus, RBC transfusions can exacerbate symptoms by elevating total RBC mass. Normocytic, normochromic anemia is characteristic of both multiple myeloma and Waldenström’s macroglobulinemia. That the patient’s chest pain recurred coincidentally with blurred vision and headache suggests the likely cause is cardiac ischemia from hyperviscosity. The serum viscosity level should be checked, and, if it is elevated, urgent serum plasmapheresis should be considered. Determining the source of excess globulin production and treating the underlying disease are crucial at this juncture.

In the general population, rectal adenocarcinoma is the most common cause of a rectal mass. In this patient, presence of a paraproteinemia may point to a different diagnosis. Extramedullary colorectal plasmacytoma can occur in the rectum but is exceedingly rare. Waldenström’s macroglobulinemia, a subtype of lymphoplasmacytic lymphoma, can be associated with a rectal lymphoma. At this point, it is not possible to confidently predict the etiology of the mass.

He was started on cyclophosphamide, bortezomib and dexamethasone for IgG κ myeloma with improvement in his headache, blurred vision, chest pain, and plasma viscosity (4 to 1.8). His hemoglobin remained stable at 10 g/dL. Neoadjuvant Capecitabine and radiation therapy were initiated for his rectal cancer.

Multiple myeloma is characterized by monoclonal proliferation of plasma cells, elevated circulating monoclonal immunoglobulin, and end-organ damage.¹ It accounts for approximately 0.8% of all new cancer diagnoses; average age at onset is 70 years. The patient described here had an unusual presentation, with GI bleeding and progression to HVS, and known risk factors for multiple myeloma (male sex, low socioeconomic status, welding career).^2,3

An early clue in the diagnosis was the patient’s large gamma gap and concurrent anemia. Gamma gap, calculated by subtracting serum albumin from serum total protein, is so named because it often reflects an elevated gamma globulin concentration. However, it actually reflects all nonalbumin serum protein. A gamma gap larger than 3.1 g/dL is an independent risk factor for death⁴ and may be associated with infection, autoimmunity, and malignancy. Although there are no screening guidelines for multiple myeloma, 73% of cases are brought to attention by anemia discovered on routine laboratory investigation.⁵ This patient’s lack of prior medical care likely contributed to his atypical presentation. Screening colonoscopy, recommended at age 50, might have identified his rectal cancer at an earlier stage.

The patient’s anemia was likely secondary to GI hemorrhage and bone marrow suppression. His hematochezia might have been partly related to the pathophysiologic interaction of paraproteins with platelets, coagulation factors, and blood vessels.⁶ Amyloidosis of the GI tract is seen in 8% of AL amyloidosis⁷ and most frequently manifests as gastrointestinal bleeding, which is thought to be due to ischemia, vascular friability, or mucosal lesions. It less commonly presents as malabsorption or dysmotility.⁸ Although gastrointestinal amyloid is not typically associated with radiologic abnormalities, occasionally it may cause luminal wall thickening, adenopathy, and inflammatory stranding.⁹ The gold standard for diagnosis is tissue biopsy. However, presence of amyloidosis does not change the overall treatment strategy for multiple myeloma.

An interesting feature of this case is the development of HVS, which typically manifests with mucosal bleeding, blurred vision, and headache.¹⁰ HVS can be diagnosed on retinal examination with findings of venous tortuosity, dilatation, and intraretinal hemorrhage, as occurred in this case,¹¹ and is confirmed with serum viscosity measurement. The first evidence of HVS in this case might have been the spontaneous echo contrast, or “smoke,” detected on echocardiogram. Spontaneous echo contrast represents increased RBC aggregation, from interaction of RBCs and plasma proteins, at low shear rates,¹² and is associated with conditions that result in left atrial stasis, such as atrial arrhythmias and mitral stenosis. This patient did not have valvular pathology or arrhythmia, and thus the “smoke” likely reflected HVS.

Of the paraproteinemias, Waldenström’s macroglobulinemia is most often associated with HVS, likely because of the pentameric structure of IgM¹³ and the consequential large size that predisposes to vascular occlusion. Whereas HVS can occur with IgM levels as low as 3 g/dL, it typically does not occur with IgG concentrations under 15 g/dL. This patient presented with an IgG level of 8 g/dL and developed HVS symptoms only after multiple packed RBC transfusions. Elevated IgG level likely made him susceptible to HVS, which ultimately was precipitated by blood transfusion. Therefore, this patient’s initial chest pain most likely was caused by demand cardiac ischemia secondary to anemia, whereas his subsequent, posttransfusion chest pain likely resulted from hyperviscosity angina. Hyperviscosity angina—cardiac ischemia resulting from poor coronary perfusion caused by hyperviscous blood—has been described in polycythemia and connective tissue disorders.^14,15 To our knowledge, however, hyperviscosity angina has not been reported in patients with multiple myeloma. Treatment of hyperviscosity with end-organ damage typically consists of plasmapheresis, but this patient was started on urgent chemotherapy, and his symptoms improved. Untreated HVS can lead to end-organ ischemia and death.

This patient had a multitude of seemingly disparate symptoms and abnormalities that ultimately were united in a diagnosis of IgG κ multiple myeloma. Subsequently diagnosed rectal adenocarcinoma may have led to ongoing blood loss, which worsened the anemia, but had no evident relation to the primary diagnosis of multiple myeloma. This case exemplifies the fact that HVS is a rare but important iatrogenic complication of multiple myeloma treated with blood transfusion. As this patient’s hospital course progressed, the plot, and his blood, thickened.

• Multiple myeloma is occasionally associated with HVS, which manifests with mucosal bleeding, blurred vision, and headache.
• Hyperviscosity angina—cardiac ischemia resulting from poor coronary perfusion caused by hyperviscous blood—should be considered in patients with paraproteinemias and chest pain.
• Plasmapheresis reverses the clinical manifestations of HVS but not the underlying disease process (eg, Waldenström’s macroglobulinemia, multiple myeloma, leukemia, polycythemia).
• Spontaneous echo contrast represents increased RBC aggregation, from interaction of RBCs and plasma proteins, at low shear rates, and is associated with left atrial stasis, commonly from atrial fibrillation or mitral stenosis, but might be present in HVS.

Acknowledgment

The authors thank Peter Campochiaro, MD, and Whitney Green, MD, for their contributions to the images used in this article.

Disclosure

Dr. Sedighi Manesh is supported by the Jeremiah A. Barondess Fellowship in the Clinical Transaction of the New York Academy of Medicine, in collaboration with the Accreditation Council for Graduate Medical Education (ACGME). The other authors have nothing to report.

The approach to clinical conundrums by an expert clinician is revealed through the presentation of an actual patient’s case in an approach typical of a morning report. Similarly to patient care, sequential pieces of information are provided to the clinician, who is unfamiliar with the case. The focus is on the thought processes of both the clinical team caring for the patient and the discussant.

After losing consciousness at a supermarket, a 70-year-old man was brought to the emergency department by paramedics. He subsequently developed chest pain.

Syncope can be difficult to evaluate, but chest pain may help narrow an otherwise broad differential diagnosis. If this patient has aortic stenosis or hypertrophic cardiomyopathy, effort syncope is the culprit. Cardiac dysrhythmia (eg, ventricular tachycardia), complete heart block, and supraventricular tachycardia each can cause syncope along with chest pain. Myocardial infarction and associated ventricular arrhythmia might also explain both chest pain and syncope. The paramedics might have noted an arrhythmia on the cardiac monitor; if possible, the rhythm strip should be reviewed. A pulmonary embolus can cause chest pain and, if large enough to cause right ventricular compromise, syncope.

According to witnesses at the supermarket, the patient dropped to the ground, lost consciousness, and convulsed for 30 seconds. He had no head trauma, tongue biting, urinary incontinence, or confusion afterward. Electrocardiogram (ECG) performed at the scene showed ST elevations in leads V1 to V3 with ST depressions in the inferior leads. On arrival in the emergency department, the patient described nonradiating substernal chest pressure exacerbated by deep inhalation. The pain did not improve with nitroglycerin. He recalled feeling light-headed before the syncope.

He had not received medical care for 20 years and had no known illnesses other than hypertension. He was not taking any medications. He previously worked as a welder and never smoked tobacco, drank alcohol, or used illicit drugs. The patient’s temperature was 36.4°C. Heart rate was 88 beats per minute, blood pressure 128/72 mm Hg, oxygen saturation 100% on room air, and respiratory rate 22 breaths per minute. The patient had conjunctival pallor. There was a grade 3/6 crescendo- decrescendo systolic murmur loudest at the right upper sternal border without radiation to the carotids. There was no jugular venous distention. Lungs were clear to auscultation bilaterally. There was no peripheral edema, rash, or lymphadenopathy.

Convulsive movements commonly occur during episodes of unconsciousness lasting more than 15 seconds—a phenomenon termed convulsive syncope and often is confused with seizures. These movements are usually clonic jerks of the extremities and trunk and slight twitching of the face, and occasionally tonic extension of the trunk and clenching of the jaw. Absence of tongue biting, urinary incontinence, and confusion in this patient’s case makes seizures less likely.

The distribution of ST segment changes on his ECG are concerning for myocardial infarction in the septal and inferior regions. Right-sided ECG should be performed to assess for right ventricular infarction. Although myocardial ischemia is the primary concern, some features warrant consideration of other etiologies of syncope. First, syncope is an unusual presentation of cardiac ischemia or infarct. The complaint of chest pressure exacerbated by deep inhalation is another atypical feature for myocardial ischemia. Although the patient’s oxygen saturation and heart rate are normal, pulmonary embolism remains a possibility.

The prominent crescendo-decrescendo systolic murmur at the right upper sternal border could indicate aortic stenosis; the carotids should be palpated to assess for pulsus parvus et tardus. A high-flow state associated with anemia could also lead to a midsystolic murmur. Conjunctival pallor typically is seen with hemoglobin levels of 6 g/dL or less. This finding may indicate severe anemia, which has the potential to cause myocardial ischemia and syncope.

Laboratory testing revealed a troponin of 0.04 ng/dL, hemoglobin 4.1 g/dL with MCV of 84.7 fL, white blood cell count 6,500/μL and platelet count 179,000/μL. Serum sodium was 130 mEq/L, urea nitrogen 16 mg/dL, creatinine 1.6 mg/dL, calcium 7.8 mg/dL, total protein 11.4 g/dL (reference range, 6.0-8.2), and albumin 2.2 g/dL. Erythrocyte sedimentation rate (ESR) was 20 mm/h. Serum iron was 48 μg/mL, total iron binding capacity 275 μg/dL, percent iron saturation 17% (reference range, 20-55), and ferritin 10 ng/mL (reference range, 30-400). The international normalized ratio (INR) was 1.5, prothrombin time 15.5 sec (reference range, 9.4-11.6), and partial thromboplastin time 24.7 sec (reference range, 22.9-30.6). Hepatitis C and HIV antibodies were negative as was the urine toxicology screen. Urine protein to creatinine ratio was 0.07. His hemoglobin rose to 7.9 g/dL with transfusion of 4 units packed red blood cells (RBC). His chest pain improved and inferior ST depressions resolved on follow-up ECG. Further history revealed multiple episodes of melena and hematochezia in the preceding weeks without nausea, vomiting, or abdominal pain.

The patient has a strikingly large gamma gap: 9.2. A gap larger than 4 is concerning for the presence of paraproteins. Given the possibility of a paraproteinemia (eg, multiple myeloma, plasmacytoma, Waldenström macroglobulinemia), the first step is to check serum and urine protein electrophoresis. The patient’s anemia is significant and reticulocyte index low. The low ferritin level combined with the inappropriately low reticulocyte count could result from iron deficiency anemia, another bone marrow process, or both. The patient’s syncope likely resulted from severe anemia and hypovolemia associated with hematochezia. The prolonged prothrombin time could be caused by a coagulation factor production problem, from vitamin K deficiency or underlying liver disease, or by a consumptive problem, from low-grade disseminated intravascular coagulation. It is controversial whether inhaling welding fumes causes cancer, but the patient’s age alone makes malignancy a definite possibility.

Aortic stenosis is intriguing in light of the patient’s painless melena and hematochezia. Heyde syndrome is a phenomenon in which high shear stress causes a reduction in the size of von Willebrand factor predisposing to bleeding from submucosal angiodysplasia. However, Heyde syndrome has been reported to occur in the setting of severe or critical aortic stenosis and is unlikely to be the cause here. The patient needs to be examined with both upper and lower endoscopy to rule out gastrointestinal (GI) malignancy, gastric and esophageal varices, and painless peptic ulcers. High right ventricular systolic blood pressures along with echodense material in the inferior vena cava and right atrium suggest the possibility of malignancy with vascular invasion. Tricuspid regurgitation is consistent with high right-sided pressures. As left ventricular ejection fraction is reduced, some of the high right-sided pressures could also be attributable to left heart failure. CT findings of rectal wall thickening and perirectal lymph nodes could be attributable to cancer with locally metastatic disease. Blood loss caused by this cancer would explain the severe anemia on admission as well as the low ferritin level and the iron deficiency anemia. In this 70-year-old man who has not had routine health care maintenance, the leading diagnosis is colorectal cancer. However, the markedly elevated globulin gap and elevated INR strongly suggest another process (eg, multiple myeloma, other paraproteinemia) is also present.

INR remained elevated (1.5) despite vitamin K supplementation. Peripheral blood smear showed hypochromic and normocytic RBCs with moderate rouleaux formation (Figure 1). Intravenous pantoprazole and octreotide were started. Upper and lower endoscopy revealed multiple esophageal erosions and both a polypoid mass and an ulcer within the rectum with evidence of prior bleeding. The mass was resected and the ulcer biopsied.

On hospital  day 3, the patient was transfused another unit of packed RBCs. Hemoglobin level increased from 7.4 g/dL to 8.2 g/dL, but he began to complain of headache, blurred vision, and worsened chest pain. He did not have weakness, numbness, diplopia, dysphagia, or dysarthria. External examination and extraocular movements of both eyes were normal. Visual acuity was 20/25 bilaterally. Funduscopic examination revealed mild dilation of retinal veins and retinal hemorrhages (Figure 2).

Rouleaux formation occurs as excess cathodal proteins, such as immunoglobulins or fibrinogen, adhere to RBCs and cause the cells to stack together in long chains. Classically this is associated with multiple myeloma, but can occur with Waldenström’s macroglobulinemia and other cancers or infections. It can also occur as an artifact in smear preparation but the large globulin gap in this patient supports pathologic rouleaux formation.

Venous retinopathy with hemorrhages may occur with occlusion of the arterial supply (eg, as with carotid artery obstruction) but also with hyperviscosity syndrome (HVS). Vascular disturbances throughout the body play a major role in HVS, but these changes are most easily visualized in the retina. It is interesting that the patient’s headache and blurred vision began after he received additional blood transfusions. Spuriously low hemoglobin and hematocrit levels may stem from increased plasma volume from high immunoglobulin M (IgM) concentrations in Waldenström macroglobulinemia; thus, RBC transfusions can exacerbate symptoms by elevating total RBC mass. Normocytic, normochromic anemia is characteristic of both multiple myeloma and Waldenström’s macroglobulinemia. That the patient’s chest pain recurred coincidentally with blurred vision and headache suggests the likely cause is cardiac ischemia from hyperviscosity. The serum viscosity level should be checked, and, if it is elevated, urgent serum plasmapheresis should be considered. Determining the source of excess globulin production and treating the underlying disease are crucial at this juncture.

In the general population, rectal adenocarcinoma is the most common cause of a rectal mass. In this patient, presence of a paraproteinemia may point to a different diagnosis. Extramedullary colorectal plasmacytoma can occur in the rectum but is exceedingly rare. Waldenström’s macroglobulinemia, a subtype of lymphoplasmacytic lymphoma, can be associated with a rectal lymphoma. At this point, it is not possible to confidently predict the etiology of the mass.

He was started on cyclophosphamide, bortezomib and dexamethasone for IgG κ myeloma with improvement in his headache, blurred vision, chest pain, and plasma viscosity (4 to 1.8). His hemoglobin remained stable at 10 g/dL. Neoadjuvant Capecitabine and radiation therapy were initiated for his rectal cancer.

Multiple myeloma is characterized by monoclonal proliferation of plasma cells, elevated circulating monoclonal immunoglobulin, and end-organ damage.¹ It accounts for approximately 0.8% of all new cancer diagnoses; average age at onset is 70 years. The patient described here had an unusual presentation, with GI bleeding and progression to HVS, and known risk factors for multiple myeloma (male sex, low socioeconomic status, welding career).^2,3

An early clue in the diagnosis was the patient’s large gamma gap and concurrent anemia. Gamma gap, calculated by subtracting serum albumin from serum total protein, is so named because it often reflects an elevated gamma globulin concentration. However, it actually reflects all nonalbumin serum protein. A gamma gap larger than 3.1 g/dL is an independent risk factor for death⁴ and may be associated with infection, autoimmunity, and malignancy. Although there are no screening guidelines for multiple myeloma, 73% of cases are brought to attention by anemia discovered on routine laboratory investigation.⁵ This patient’s lack of prior medical care likely contributed to his atypical presentation. Screening colonoscopy, recommended at age 50, might have identified his rectal cancer at an earlier stage.

The patient’s anemia was likely secondary to GI hemorrhage and bone marrow suppression. His hematochezia might have been partly related to the pathophysiologic interaction of paraproteins with platelets, coagulation factors, and blood vessels.⁶ Amyloidosis of the GI tract is seen in 8% of AL amyloidosis⁷ and most frequently manifests as gastrointestinal bleeding, which is thought to be due to ischemia, vascular friability, or mucosal lesions. It less commonly presents as malabsorption or dysmotility.⁸ Although gastrointestinal amyloid is not typically associated with radiologic abnormalities, occasionally it may cause luminal wall thickening, adenopathy, and inflammatory stranding.⁹ The gold standard for diagnosis is tissue biopsy. However, presence of amyloidosis does not change the overall treatment strategy for multiple myeloma.

An interesting feature of this case is the development of HVS, which typically manifests with mucosal bleeding, blurred vision, and headache.¹⁰ HVS can be diagnosed on retinal examination with findings of venous tortuosity, dilatation, and intraretinal hemorrhage, as occurred in this case,¹¹ and is confirmed with serum viscosity measurement. The first evidence of HVS in this case might have been the spontaneous echo contrast, or “smoke,” detected on echocardiogram. Spontaneous echo contrast represents increased RBC aggregation, from interaction of RBCs and plasma proteins, at low shear rates,¹² and is associated with conditions that result in left atrial stasis, such as atrial arrhythmias and mitral stenosis. This patient did not have valvular pathology or arrhythmia, and thus the “smoke” likely reflected HVS.

Of the paraproteinemias, Waldenström’s macroglobulinemia is most often associated with HVS, likely because of the pentameric structure of IgM¹³ and the consequential large size that predisposes to vascular occlusion. Whereas HVS can occur with IgM levels as low as 3 g/dL, it typically does not occur with IgG concentrations under 15 g/dL. This patient presented with an IgG level of 8 g/dL and developed HVS symptoms only after multiple packed RBC transfusions. Elevated IgG level likely made him susceptible to HVS, which ultimately was precipitated by blood transfusion. Therefore, this patient’s initial chest pain most likely was caused by demand cardiac ischemia secondary to anemia, whereas his subsequent, posttransfusion chest pain likely resulted from hyperviscosity angina. Hyperviscosity angina—cardiac ischemia resulting from poor coronary perfusion caused by hyperviscous blood—has been described in polycythemia and connective tissue disorders.^14,15 To our knowledge, however, hyperviscosity angina has not been reported in patients with multiple myeloma. Treatment of hyperviscosity with end-organ damage typically consists of plasmapheresis, but this patient was started on urgent chemotherapy, and his symptoms improved. Untreated HVS can lead to end-organ ischemia and death.

This patient had a multitude of seemingly disparate symptoms and abnormalities that ultimately were united in a diagnosis of IgG κ multiple myeloma. Subsequently diagnosed rectal adenocarcinoma may have led to ongoing blood loss, which worsened the anemia, but had no evident relation to the primary diagnosis of multiple myeloma. This case exemplifies the fact that HVS is a rare but important iatrogenic complication of multiple myeloma treated with blood transfusion. As this patient’s hospital course progressed, the plot, and his blood, thickened.

• Multiple myeloma is occasionally associated with HVS, which manifests with mucosal bleeding, blurred vision, and headache.
• Hyperviscosity angina—cardiac ischemia resulting from poor coronary perfusion caused by hyperviscous blood—should be considered in patients with paraproteinemias and chest pain.
• Plasmapheresis reverses the clinical manifestations of HVS but not the underlying disease process (eg, Waldenström’s macroglobulinemia, multiple myeloma, leukemia, polycythemia).
• Spontaneous echo contrast represents increased RBC aggregation, from interaction of RBCs and plasma proteins, at low shear rates, and is associated with left atrial stasis, commonly from atrial fibrillation or mitral stenosis, but might be present in HVS.

Acknowledgment

The authors thank Peter Campochiaro, MD, and Whitney Green, MD, for their contributions to the images used in this article.

Disclosure

Dr. Sedighi Manesh is supported by the Jeremiah A. Barondess Fellowship in the Clinical Transaction of the New York Academy of Medicine, in collaboration with the Accreditation Council for Graduate Medical Education (ACGME). The other authors have nothing to report.

The approach to clinical conundrums by an expert clinician is revealed through the presentation of an actual patient’s case in an approach typical of a morning report. Similarly to patient care, sequential pieces of information are provided to the clinician, who is unfamiliar with the case. The focus is on the thought processes of both the clinical team caring for the patient and the discussant.

After losing consciousness at a supermarket, a 70-year-old man was brought to the emergency department by paramedics. He subsequently developed chest pain.

Syncope can be difficult to evaluate, but chest pain may help narrow an otherwise broad differential diagnosis. If this patient has aortic stenosis or hypertrophic cardiomyopathy, effort syncope is the culprit. Cardiac dysrhythmia (eg, ventricular tachycardia), complete heart block, and supraventricular tachycardia each can cause syncope along with chest pain. Myocardial infarction and associated ventricular arrhythmia might also explain both chest pain and syncope. The paramedics might have noted an arrhythmia on the cardiac monitor; if possible, the rhythm strip should be reviewed. A pulmonary embolus can cause chest pain and, if large enough to cause right ventricular compromise, syncope.

According to witnesses at the supermarket, the patient dropped to the ground, lost consciousness, and convulsed for 30 seconds. He had no head trauma, tongue biting, urinary incontinence, or confusion afterward. Electrocardiogram (ECG) performed at the scene showed ST elevations in leads V1 to V3 with ST depressions in the inferior leads. On arrival in the emergency department, the patient described nonradiating substernal chest pressure exacerbated by deep inhalation. The pain did not improve with nitroglycerin. He recalled feeling light-headed before the syncope.

He had not received medical care for 20 years and had no known illnesses other than hypertension. He was not taking any medications. He previously worked as a welder and never smoked tobacco, drank alcohol, or used illicit drugs. The patient’s temperature was 36.4°C. Heart rate was 88 beats per minute, blood pressure 128/72 mm Hg, oxygen saturation 100% on room air, and respiratory rate 22 breaths per minute. The patient had conjunctival pallor. There was a grade 3/6 crescendo- decrescendo systolic murmur loudest at the right upper sternal border without radiation to the carotids. There was no jugular venous distention. Lungs were clear to auscultation bilaterally. There was no peripheral edema, rash, or lymphadenopathy.

Convulsive movements commonly occur during episodes of unconsciousness lasting more than 15 seconds—a phenomenon termed convulsive syncope and often is confused with seizures. These movements are usually clonic jerks of the extremities and trunk and slight twitching of the face, and occasionally tonic extension of the trunk and clenching of the jaw. Absence of tongue biting, urinary incontinence, and confusion in this patient’s case makes seizures less likely.

The distribution of ST segment changes on his ECG are concerning for myocardial infarction in the septal and inferior regions. Right-sided ECG should be performed to assess for right ventricular infarction. Although myocardial ischemia is the primary concern, some features warrant consideration of other etiologies of syncope. First, syncope is an unusual presentation of cardiac ischemia or infarct. The complaint of chest pressure exacerbated by deep inhalation is another atypical feature for myocardial ischemia. Although the patient’s oxygen saturation and heart rate are normal, pulmonary embolism remains a possibility.

The prominent crescendo-decrescendo systolic murmur at the right upper sternal border could indicate aortic stenosis; the carotids should be palpated to assess for pulsus parvus et tardus. A high-flow state associated with anemia could also lead to a midsystolic murmur. Conjunctival pallor typically is seen with hemoglobin levels of 6 g/dL or less. This finding may indicate severe anemia, which has the potential to cause myocardial ischemia and syncope.

Laboratory testing revealed a troponin of 0.04 ng/dL, hemoglobin 4.1 g/dL with MCV of 84.7 fL, white blood cell count 6,500/μL and platelet count 179,000/μL. Serum sodium was 130 mEq/L, urea nitrogen 16 mg/dL, creatinine 1.6 mg/dL, calcium 7.8 mg/dL, total protein 11.4 g/dL (reference range, 6.0-8.2), and albumin 2.2 g/dL. Erythrocyte sedimentation rate (ESR) was 20 mm/h. Serum iron was 48 μg/mL, total iron binding capacity 275 μg/dL, percent iron saturation 17% (reference range, 20-55), and ferritin 10 ng/mL (reference range, 30-400). The international normalized ratio (INR) was 1.5, prothrombin time 15.5 sec (reference range, 9.4-11.6), and partial thromboplastin time 24.7 sec (reference range, 22.9-30.6). Hepatitis C and HIV antibodies were negative as was the urine toxicology screen. Urine protein to creatinine ratio was 0.07. His hemoglobin rose to 7.9 g/dL with transfusion of 4 units packed red blood cells (RBC). His chest pain improved and inferior ST depressions resolved on follow-up ECG. Further history revealed multiple episodes of melena and hematochezia in the preceding weeks without nausea, vomiting, or abdominal pain.

The patient has a strikingly large gamma gap: 9.2. A gap larger than 4 is concerning for the presence of paraproteins. Given the possibility of a paraproteinemia (eg, multiple myeloma, plasmacytoma, Waldenström macroglobulinemia), the first step is to check serum and urine protein electrophoresis. The patient’s anemia is significant and reticulocyte index low. The low ferritin level combined with the inappropriately low reticulocyte count could result from iron deficiency anemia, another bone marrow process, or both. The patient’s syncope likely resulted from severe anemia and hypovolemia associated with hematochezia. The prolonged prothrombin time could be caused by a coagulation factor production problem, from vitamin K deficiency or underlying liver disease, or by a consumptive problem, from low-grade disseminated intravascular coagulation. It is controversial whether inhaling welding fumes causes cancer, but the patient’s age alone makes malignancy a definite possibility.

Aortic stenosis is intriguing in light of the patient’s painless melena and hematochezia. Heyde syndrome is a phenomenon in which high shear stress causes a reduction in the size of von Willebrand factor predisposing to bleeding from submucosal angiodysplasia. However, Heyde syndrome has been reported to occur in the setting of severe or critical aortic stenosis and is unlikely to be the cause here. The patient needs to be examined with both upper and lower endoscopy to rule out gastrointestinal (GI) malignancy, gastric and esophageal varices, and painless peptic ulcers. High right ventricular systolic blood pressures along with echodense material in the inferior vena cava and right atrium suggest the possibility of malignancy with vascular invasion. Tricuspid regurgitation is consistent with high right-sided pressures. As left ventricular ejection fraction is reduced, some of the high right-sided pressures could also be attributable to left heart failure. CT findings of rectal wall thickening and perirectal lymph nodes could be attributable to cancer with locally metastatic disease. Blood loss caused by this cancer would explain the severe anemia on admission as well as the low ferritin level and the iron deficiency anemia. In this 70-year-old man who has not had routine health care maintenance, the leading diagnosis is colorectal cancer. However, the markedly elevated globulin gap and elevated INR strongly suggest another process (eg, multiple myeloma, other paraproteinemia) is also present.

INR remained elevated (1.5) despite vitamin K supplementation. Peripheral blood smear showed hypochromic and normocytic RBCs with moderate rouleaux formation (Figure 1). Intravenous pantoprazole and octreotide were started. Upper and lower endoscopy revealed multiple esophageal erosions and both a polypoid mass and an ulcer within the rectum with evidence of prior bleeding. The mass was resected and the ulcer biopsied.

On hospital  day 3, the patient was transfused another unit of packed RBCs. Hemoglobin level increased from 7.4 g/dL to 8.2 g/dL, but he began to complain of headache, blurred vision, and worsened chest pain. He did not have weakness, numbness, diplopia, dysphagia, or dysarthria. External examination and extraocular movements of both eyes were normal. Visual acuity was 20/25 bilaterally. Funduscopic examination revealed mild dilation of retinal veins and retinal hemorrhages (Figure 2).

Rouleaux formation occurs as excess cathodal proteins, such as immunoglobulins or fibrinogen, adhere to RBCs and cause the cells to stack together in long chains. Classically this is associated with multiple myeloma, but can occur with Waldenström’s macroglobulinemia and other cancers or infections. It can also occur as an artifact in smear preparation but the large globulin gap in this patient supports pathologic rouleaux formation.

Venous retinopathy with hemorrhages may occur with occlusion of the arterial supply (eg, as with carotid artery obstruction) but also with hyperviscosity syndrome (HVS). Vascular disturbances throughout the body play a major role in HVS, but these changes are most easily visualized in the retina. It is interesting that the patient’s headache and blurred vision began after he received additional blood transfusions. Spuriously low hemoglobin and hematocrit levels may stem from increased plasma volume from high immunoglobulin M (IgM) concentrations in Waldenström macroglobulinemia; thus, RBC transfusions can exacerbate symptoms by elevating total RBC mass. Normocytic, normochromic anemia is characteristic of both multiple myeloma and Waldenström’s macroglobulinemia. That the patient’s chest pain recurred coincidentally with blurred vision and headache suggests the likely cause is cardiac ischemia from hyperviscosity. The serum viscosity level should be checked, and, if it is elevated, urgent serum plasmapheresis should be considered. Determining the source of excess globulin production and treating the underlying disease are crucial at this juncture.

In the general population, rectal adenocarcinoma is the most common cause of a rectal mass. In this patient, presence of a paraproteinemia may point to a different diagnosis. Extramedullary colorectal plasmacytoma can occur in the rectum but is exceedingly rare. Waldenström’s macroglobulinemia, a subtype of lymphoplasmacytic lymphoma, can be associated with a rectal lymphoma. At this point, it is not possible to confidently predict the etiology of the mass.

He was started on cyclophosphamide, bortezomib and dexamethasone for IgG κ myeloma with improvement in his headache, blurred vision, chest pain, and plasma viscosity (4 to 1.8). His hemoglobin remained stable at 10 g/dL. Neoadjuvant Capecitabine and radiation therapy were initiated for his rectal cancer.

Multiple myeloma is characterized by monoclonal proliferation of plasma cells, elevated circulating monoclonal immunoglobulin, and end-organ damage.¹ It accounts for approximately 0.8% of all new cancer diagnoses; average age at onset is 70 years. The patient described here had an unusual presentation, with GI bleeding and progression to HVS, and known risk factors for multiple myeloma (male sex, low socioeconomic status, welding career).^2,3

An early clue in the diagnosis was the patient’s large gamma gap and concurrent anemia. Gamma gap, calculated by subtracting serum albumin from serum total protein, is so named because it often reflects an elevated gamma globulin concentration. However, it actually reflects all nonalbumin serum protein. A gamma gap larger than 3.1 g/dL is an independent risk factor for death⁴ and may be associated with infection, autoimmunity, and malignancy. Although there are no screening guidelines for multiple myeloma, 73% of cases are brought to attention by anemia discovered on routine laboratory investigation.⁵ This patient’s lack of prior medical care likely contributed to his atypical presentation. Screening colonoscopy, recommended at age 50, might have identified his rectal cancer at an earlier stage.

The patient’s anemia was likely secondary to GI hemorrhage and bone marrow suppression. His hematochezia might have been partly related to the pathophysiologic interaction of paraproteins with platelets, coagulation factors, and blood vessels.⁶ Amyloidosis of the GI tract is seen in 8% of AL amyloidosis⁷ and most frequently manifests as gastrointestinal bleeding, which is thought to be due to ischemia, vascular friability, or mucosal lesions. It less commonly presents as malabsorption or dysmotility.⁸ Although gastrointestinal amyloid is not typically associated with radiologic abnormalities, occasionally it may cause luminal wall thickening, adenopathy, and inflammatory stranding.⁹ The gold standard for diagnosis is tissue biopsy. However, presence of amyloidosis does not change the overall treatment strategy for multiple myeloma.

An interesting feature of this case is the development of HVS, which typically manifests with mucosal bleeding, blurred vision, and headache.¹⁰ HVS can be diagnosed on retinal examination with findings of venous tortuosity, dilatation, and intraretinal hemorrhage, as occurred in this case,¹¹ and is confirmed with serum viscosity measurement. The first evidence of HVS in this case might have been the spontaneous echo contrast, or “smoke,” detected on echocardiogram. Spontaneous echo contrast represents increased RBC aggregation, from interaction of RBCs and plasma proteins, at low shear rates,¹² and is associated with conditions that result in left atrial stasis, such as atrial arrhythmias and mitral stenosis. This patient did not have valvular pathology or arrhythmia, and thus the “smoke” likely reflected HVS.

Of the paraproteinemias, Waldenström’s macroglobulinemia is most often associated with HVS, likely because of the pentameric structure of IgM¹³ and the consequential large size that predisposes to vascular occlusion. Whereas HVS can occur with IgM levels as low as 3 g/dL, it typically does not occur with IgG concentrations under 15 g/dL. This patient presented with an IgG level of 8 g/dL and developed HVS symptoms only after multiple packed RBC transfusions. Elevated IgG level likely made him susceptible to HVS, which ultimately was precipitated by blood transfusion. Therefore, this patient’s initial chest pain most likely was caused by demand cardiac ischemia secondary to anemia, whereas his subsequent, posttransfusion chest pain likely resulted from hyperviscosity angina. Hyperviscosity angina—cardiac ischemia resulting from poor coronary perfusion caused by hyperviscous blood—has been described in polycythemia and connective tissue disorders.^14,15 To our knowledge, however, hyperviscosity angina has not been reported in patients with multiple myeloma. Treatment of hyperviscosity with end-organ damage typically consists of plasmapheresis, but this patient was started on urgent chemotherapy, and his symptoms improved. Untreated HVS can lead to end-organ ischemia and death.

This patient had a multitude of seemingly disparate symptoms and abnormalities that ultimately were united in a diagnosis of IgG κ multiple myeloma. Subsequently diagnosed rectal adenocarcinoma may have led to ongoing blood loss, which worsened the anemia, but had no evident relation to the primary diagnosis of multiple myeloma. This case exemplifies the fact that HVS is a rare but important iatrogenic complication of multiple myeloma treated with blood transfusion. As this patient’s hospital course progressed, the plot, and his blood, thickened.

• Multiple myeloma is occasionally associated with HVS, which manifests with mucosal bleeding, blurred vision, and headache.
• Hyperviscosity angina—cardiac ischemia resulting from poor coronary perfusion caused by hyperviscous blood—should be considered in patients with paraproteinemias and chest pain.
• Plasmapheresis reverses the clinical manifestations of HVS but not the underlying disease process (eg, Waldenström’s macroglobulinemia, multiple myeloma, leukemia, polycythemia).
• Spontaneous echo contrast represents increased RBC aggregation, from interaction of RBCs and plasma proteins, at low shear rates, and is associated with left atrial stasis, commonly from atrial fibrillation or mitral stenosis, but might be present in HVS.

Acknowledgment

The authors thank Peter Campochiaro, MD, and Whitney Green, MD, for their contributions to the images used in this article.

Disclosure

Dr. Sedighi Manesh is supported by the Jeremiah A. Barondess Fellowship in the Clinical Transaction of the New York Academy of Medicine, in collaboration with the Accreditation Council for Graduate Medical Education (ACGME). The other authors have nothing to report.

For over 3 decades, AcademyHealth and its predecessor organizations and members have been studying how the healthcare system works and ways to improve health outcomes. The value of health services research (HSR) training programs that emphasize learning opportunities in delivery system settings was initially articulated at the 2009 AcademyHealth Summit on the Future of HSR Data and Methods.¹ Two years later, the need for such programs was reiterated as a priority for AcademyHealth’s HSR Learning Consortium in their strategic plan.² While HSR methods have become increasingly sophisticated, historical approaches largely relied on extant (usually academic-based) researchers.

To realize the goal of building learning health systems (considered here to be entities where applied, operationally relevant research is systematically designed, generated, and translated into high-quality care delivery), many healthcare organizations have begun to use researchers as an internal resource to inform and support higher quality and more efficient care delivery operations. However, there is a current dearth of scientists trained in research disciplines (eg, comparative effectiveness research, patient-centered outcomes research, implementation science) more directly applicable to operational settings.^3,4 Conducting research within these “real-world” environments is challenging for a variety of well-documented reasons,^5,6 and many important questions cannot be answered using traditional study designs and/or methodologies. Until more researchers are trained in research approaches that align better with care delivery needs, the field will continue to fall short of addressing topics identified by the National Academy of Medicine (NAM; formerly the Institute of Medicine) as priorities for real system improvement.

While researchers in academic and consulting settings play critical roles in knowledge generation, a substantial area for expansion is support for “embedded researchers” who work more directly with operational leaders and understand local context, data, and organizational-level goals of delivery systems. AcademyHealth’s Delivery System Science Fellowship (DSSF) was developed to forge a stronger link between rigorous research practice and pragmatic aspects of care delivery.

Since inception of the DSSF, national attention has further emphasized the need for specialized, experiential learning with specific competencies that extend formal HSR training. The Agency for Healthcare Research and Quality (AHRQ) is currently convening a Technical Expert Panel to guide Training the Next Generation of Learning Health System Researchers as part of its larger effort to provide support for evidence generation and uptake in these applied settings (see http://www.ahrq.gov/news/blog/ahrqviews/supporting-learning-health-systems.html for more information). This work is reinforced by an ongoing project at AcademyHealth focused on Understanding the Current Health Services Research Workforce and Maximizing Its Future,⁷ partially sponsored by AHRQ. These efforts aim to inform AHRQ-funded career and training program requirements in order to build a workforce in partnership with academia and delivery organizations in support of developing high-functioning researchers directly positioned to drive progress toward learning healthcare systems.

The AcademyHealth DSSF Program is a pioneering effort designed to meet the needs of learning healthcare systems for a human capital resource capable of generating insights from operational data and deploying this knowledge effectively. Established in 2012 in partnership with 3 initial host sites, the DSSF provides a paid, postdoctoral training opportunity to help highly qualified, early-career researchers gain applied experience in delivery system settings.⁸ The goal is to provide hands-on experience and professional leadership opportunities to enhance the array of skills needed to generate and apply evidence in delivery systems. The long-term program objective is for graduates to employ the methods and training garnered during the DSSF to produce new, practical insights required to transform healthcare delivery and achieve the “Triple Aim.”⁹

Sixteen delivery systems have participated in the DSSF program over the course of 5 years. These organizations represent a diverse group of innovative, high-performing systems that serve populations across the rural-urban continuum. Host site preceptors are nationally recognized experts in areas such as health economics, comparative effectiveness, pragmatic trials, clinical decision support, and implementation science. The fellowship is guided by an advisory committee that provides strategic direction and plays a key role in selecting fellows. AcademyHealth partners with 8 to 10 health systems annually, accepting new partnerships as interest and availability allows. The Figure summarizes program inputs, elements, antecedents, outputs, and intended outcomes. As a relatively new and evolving program, the DSSF is using a disciplined approach to assess both mid- and long-term outcomes. The Table is a comparative matrix that depicts the range in areas of investment offered by a sample of host sites during the fellowship program. Complementing development of the common core competencies and exposures presented in the Table, specific project choice at a host site is mutually determined according to system needs and the fellow’s interests.

To apply for DSSF, researchers must hold a doctoral degree in any relevant aspect of clinical medicine, HSR, or a related field. The review committee evaluates applicants based on their qualifications, a clear statement of professional goals, appropriateness for placement, the quality of a writing sample, and letters of recommendation. Host sites select a fellow based on their individual preferences and recommendations from the review committee. The minimum duration of the fellowship is 1 year, during which the fellow works full time at the host site. Host sites provide the fellows’ salary and benefits, financial support to attend AcademyHealth’s Annual Research Meeting, and mentorship and/or training. To ensure general continuity across sites, each fellow’s work focuses on “delivery system science (DSS),” with a significant part of the work intended for public dissemination (eg, a conference presentation) or publication in a peer-reviewed journal.

Host sites additionally provide AcademyHealth financial support to manage the process of recruiting promising candidates. AcademyHealth manages and convenes the advisory committee proceedings, facilitates the application cycle (including peer review of applicants), markets the fellowship, guides the interview and match process, and promotes placement of fellows upon program completion. AcademyHealth also convenes active and graduate fellows to foster engagement and professional development.

Program Level

AcademyHealth administers an annual evaluation to assess the program, understand impact on the fellow’s professional development and growth, track publications, and inform programmatic goals. To date, we have identified over 50 peer-reviewed publications resulting from work conducted through the fellowship (visit http://www.academyhealth.org/dssfpublications for a full listing of publications stemming from the DSSF program). As an example of continuous program improvement, this year staff implemented a fellow-led monthly call in response to requests to connect fellows. This has proven to be a useful response for fellows to understand how to enhance their own experience by learning from fellows in other systems to incorporate focused areas of development via cross-system sharing. As an indicator of continued value, of the 16 sites that have participated to date, 10 have participated in the DSSF for more than 1 cycle; 3 are new host sites currently participating in their first year of the program; and 3 host sites participated for 1 year. The 3 inaugural host sites that helped launch the DSSF continue to serve as host sites to date.

As a marker of longitudinal success, staff will continue to follow up with preceptors and fellows to understand fellows’ contributions to the host site and the field, as well as impact on the fellow career trajectory. Of the 19 fellow alumni to date, 8 have moved on to academic or research positions, and 11 have remained in care delivery systems to provide local expertise in study design, execution, and evidence uptake.

The program has also made some general contributions to advance the discipline of delivery system science, including:

1. Defining DSS and clarifying training needs for “embedded researchers” and health system analysts. To characterize the fellowship, AcademyHealth and the program advisory committee jointly developed the following definition for delivery system science (DSS):

“DSS includes research that seeks to understand how delivery systems operate, influence, change, and respond to external stimulus, among other topics. DSS may include efforts to examine how and under what circumstances interventions work and how delivery systems effectively implement evidence-based innovations. For the DSSF, DSS is conducted by researchers who are ‘embedded’ in delivery systems and respond to the decision-making needs of those systems.”

Additionally, closer connections with DSS leaders have led to a better understanding of challenges, opportunities, and needs of delivery systems.^10,11

2. Cultivating a network of delivery systems and system leaders interested in expanding the cadre of embedded researchers, and trainees who intend to build careers in DSS. The DSSF also aims to enhance fellows’ skills and knowledge base, career opportunities, and professional network. To extend these relationships and support delivery system analytics, AcademyHealth worked with preceptors and fellows to inform creation of a new Community of Practice supported by AcademyHealth’s EDM Forum and guide planning for AcademyHealth’s Concordium conference to provide a national meeting to showcase DSS.

3. Strategic planning to ensure sustainable support for embedded research within delivery systems. Substantial interest in the program developed quickly, with rapid learning over the first few cycles to refine the program to meet host sites’ and fellows’ needs. Both efforts were critical to demonstrate that the DSSF fulfills an important need for our health system partners and members. As indicated previously, strong, sustained interest from prospective host sites and applicants demonstrates the program has created a win-win to jointly assess fit while building skills and supporting continuous learning.

Likewise, Lisa Simpson, President and CEO of AcademyHealth, and Lucy Savitz, DSSF host site preceptor at Intermountain Healthcare, participated in the Canadian Institutes of Health Research (CIHR) Invitational Workshop, “Modernizing Health Services and Policy Research Training in Canada” in March 2016. Shared learning largely informed by the DSSF led to CIHR creating a similar fellowship program with initial awards to be made in 2017 (see https://www.researchnet-recherchenet.ca/rnr16/vwOpprtntyDtls.do?prog=2540&view=browseActive&sponsor=CIHR-8&type=EXACT&resultCount=25 for more information). We are working to thread these efforts together in a way to leverage our learning community of government agencies, academia, and employers as a long-term funding stream for training in delivery science.

Participant Host Site Level

Two selected examples of how DSSF researchers have engaged high-priority topics that contributed to health system operations are provided here.

Kaiser Permanente Southern California: Assessing adherence with “Choosing Wisely” recommendations in oncology. In partnership with preceptor Dr. Michael Gould, 2013-2014 DSSF fellow Dr. Erin Hahn worked with the Kaiser Permanente Southern California (KPSC) Care Improvement Research Team to lead a project addressing several KPSC priority areas. Focusing on “Choosing Wisely” recommendations from the American Society of Clinical Oncology,¹² the project evaluated appropriateness of imaging and laboratory services for early-stage cancer patients and survivors between and within 2 integrated health systems, Kaiser Permanente (KP) and Intermountain Healthcare.^13,14 Results were presented to KP national leaders, including an external health policy advisory board. In close collaboration with clinical and operational leaders in medical oncology, this multiregional, multisystem project is contributing to targeted quality improvement efforts and improved healthcare value, including audit and feedback of nonrecommended labs.

Dr. Hahn subsequently received a KPSC Incubator Award, a competitive internal grant, to further study factors associated with use of nonrecommended surveillance lab tests for early-stage breast cancer patients. The study focused on medical oncologists within KPSC, categorizing them as high or low utilizers of the tests.¹⁵ Results indicate that high utilizers perceive that the tests help manage patient anxiety about recurrence, while acknowledging that the tests do not provide clinical utility. These findings are contributing to the development of targeted survivorship services across the organization.

Intermountain Healthcare: Formative evaluation of large-scale implementation of shared decision-making. Preceptor Dr. Lucy Savitz assembled a team to conduct a formative evaluation of Intermountain Healthcare’s efforts to implement shared decision-making (SDM) as part of its Center for Medicare and Medicaid Innovation Challenge Award. The 2015-2017 DSSF fellow at Intermountain Healthcare, Dr. Kim Brunisholz, served as a core member of the project team, focusing primarily on a mixed-methods evaluation of the SDM program.

Dr. Brunisholz engaged operational leads, clinical teams, patient and family advisory councils, and senior executives to conduct the program assessment. Results demonstrated significant variation in invited participation in SDM among eligible patient populations: preference sensitive conditions (1 in 30 patients), oncology-related diagnosis (1 in 3 patients), and chronic conditions (1 in 74 patients). Provisional analysis of patient-level clinical outcomes demonstrated that among those invited to the SDM program compared to those that were not, total joint replacement was decreased (10.1% vs 17.3%; P < 0.001) and a trend towards breast conservation emerged (61.8% vs 56.4%; P = 0.10). No difference in treatment choice for lower back pain was observed. Qualitative program analysis suggested need for improvements in the areas of (1) routine and continuous staff training, (2) workflow standardization, and (3) active data monitoring with meaningful, actionable feedback to caregivers. In response to these results, a chartered SDM Steering Committee was created (Dr. Brunisholz is a member of that group) to develop a strategic plan for SDM, with an accompanying organizational response to reimplement SDM in a targeted manner. Learning from this program is being leveraged to support a subproject analysis on a large scale using data from the High Value Healthcare Collaborative as part of an AHRQ-funded Center of Excellence award. (See https://www.ahrq.gov/news/newsroom/press-releases/2015/pcorawards.html. For more information on the High Value Healthcare Collaborative, please visit: https://www.highvaluehealthcare.org.)

Moving forward, the DSSF will continue working with progressive delivery systems. Partnerships between organizations that are interested in integrating rigorous research practice to drive continuous system improvement and maximize the value of care will have substantial need for technical skills and analytic capacity. They will also need to ensure that researchers working in their systems have sufficient understanding of cultural and political context within the organization to be effective leaders who can manage change.

AcademyHealth created the DSSF in response to the field’s request to build a research workforce that reflects the vision for a 21st Century Health System, as laid out by the NAM.¹⁶ We anticipate that as the US Department of Health & Human Services’ goals for payment reform and new measures to promote quality and high-value care are implemented, the DSSF trainees’ skill set will be increasingly valuable and will provide needed thought leadership on strategies to generate and apply evidence in practice.

Disclosure

Ms. Kanani received funding from Intermountain Healthcare, Kaiser Permanente Southern California for support for the Delivery System Science Fellowship. Drs. Hahn, Gould, and Brunisholz have no conflicts to disclose. Dr. Savitz has received funding from HRQ COE, PCORI LHSNet; received funding for lectures from the Institute from Healthcare Improvement, Department of Epidemiology, University of Utah; received funding for travel, accommodations, and meeting expenses from AHRQ NAC, EDM Forum, AH CAPP, AARP NPC, and PROM TEP; and received additional funding from Dartmouth University. Dr. Holve received funding from Intermountain Healthcare, Kaiser Permanente Southern California for support for the Delivery System Science Fellowship, provided by our delivery system partners, several of whom are coauthors on this manuscript.

For over 3 decades, AcademyHealth and its predecessor organizations and members have been studying how the healthcare system works and ways to improve health outcomes. The value of health services research (HSR) training programs that emphasize learning opportunities in delivery system settings was initially articulated at the 2009 AcademyHealth Summit on the Future of HSR Data and Methods.¹ Two years later, the need for such programs was reiterated as a priority for AcademyHealth’s HSR Learning Consortium in their strategic plan.² While HSR methods have become increasingly sophisticated, historical approaches largely relied on extant (usually academic-based) researchers.

To realize the goal of building learning health systems (considered here to be entities where applied, operationally relevant research is systematically designed, generated, and translated into high-quality care delivery), many healthcare organizations have begun to use researchers as an internal resource to inform and support higher quality and more efficient care delivery operations. However, there is a current dearth of scientists trained in research disciplines (eg, comparative effectiveness research, patient-centered outcomes research, implementation science) more directly applicable to operational settings.^3,4 Conducting research within these “real-world” environments is challenging for a variety of well-documented reasons,^5,6 and many important questions cannot be answered using traditional study designs and/or methodologies. Until more researchers are trained in research approaches that align better with care delivery needs, the field will continue to fall short of addressing topics identified by the National Academy of Medicine (NAM; formerly the Institute of Medicine) as priorities for real system improvement.

While researchers in academic and consulting settings play critical roles in knowledge generation, a substantial area for expansion is support for “embedded researchers” who work more directly with operational leaders and understand local context, data, and organizational-level goals of delivery systems. AcademyHealth’s Delivery System Science Fellowship (DSSF) was developed to forge a stronger link between rigorous research practice and pragmatic aspects of care delivery.

Since inception of the DSSF, national attention has further emphasized the need for specialized, experiential learning with specific competencies that extend formal HSR training. The Agency for Healthcare Research and Quality (AHRQ) is currently convening a Technical Expert Panel to guide Training the Next Generation of Learning Health System Researchers as part of its larger effort to provide support for evidence generation and uptake in these applied settings (see http://www.ahrq.gov/news/blog/ahrqviews/supporting-learning-health-systems.html for more information). This work is reinforced by an ongoing project at AcademyHealth focused on Understanding the Current Health Services Research Workforce and Maximizing Its Future,⁷ partially sponsored by AHRQ. These efforts aim to inform AHRQ-funded career and training program requirements in order to build a workforce in partnership with academia and delivery organizations in support of developing high-functioning researchers directly positioned to drive progress toward learning healthcare systems.

The AcademyHealth DSSF Program is a pioneering effort designed to meet the needs of learning healthcare systems for a human capital resource capable of generating insights from operational data and deploying this knowledge effectively. Established in 2012 in partnership with 3 initial host sites, the DSSF provides a paid, postdoctoral training opportunity to help highly qualified, early-career researchers gain applied experience in delivery system settings.⁸ The goal is to provide hands-on experience and professional leadership opportunities to enhance the array of skills needed to generate and apply evidence in delivery systems. The long-term program objective is for graduates to employ the methods and training garnered during the DSSF to produce new, practical insights required to transform healthcare delivery and achieve the “Triple Aim.”⁹

Sixteen delivery systems have participated in the DSSF program over the course of 5 years. These organizations represent a diverse group of innovative, high-performing systems that serve populations across the rural-urban continuum. Host site preceptors are nationally recognized experts in areas such as health economics, comparative effectiveness, pragmatic trials, clinical decision support, and implementation science. The fellowship is guided by an advisory committee that provides strategic direction and plays a key role in selecting fellows. AcademyHealth partners with 8 to 10 health systems annually, accepting new partnerships as interest and availability allows. The Figure summarizes program inputs, elements, antecedents, outputs, and intended outcomes. As a relatively new and evolving program, the DSSF is using a disciplined approach to assess both mid- and long-term outcomes. The Table is a comparative matrix that depicts the range in areas of investment offered by a sample of host sites during the fellowship program. Complementing development of the common core competencies and exposures presented in the Table, specific project choice at a host site is mutually determined according to system needs and the fellow’s interests.

To apply for DSSF, researchers must hold a doctoral degree in any relevant aspect of clinical medicine, HSR, or a related field. The review committee evaluates applicants based on their qualifications, a clear statement of professional goals, appropriateness for placement, the quality of a writing sample, and letters of recommendation. Host sites select a fellow based on their individual preferences and recommendations from the review committee. The minimum duration of the fellowship is 1 year, during which the fellow works full time at the host site. Host sites provide the fellows’ salary and benefits, financial support to attend AcademyHealth’s Annual Research Meeting, and mentorship and/or training. To ensure general continuity across sites, each fellow’s work focuses on “delivery system science (DSS),” with a significant part of the work intended for public dissemination (eg, a conference presentation) or publication in a peer-reviewed journal.

Host sites additionally provide AcademyHealth financial support to manage the process of recruiting promising candidates. AcademyHealth manages and convenes the advisory committee proceedings, facilitates the application cycle (including peer review of applicants), markets the fellowship, guides the interview and match process, and promotes placement of fellows upon program completion. AcademyHealth also convenes active and graduate fellows to foster engagement and professional development.

Program Level

AcademyHealth administers an annual evaluation to assess the program, understand impact on the fellow’s professional development and growth, track publications, and inform programmatic goals. To date, we have identified over 50 peer-reviewed publications resulting from work conducted through the fellowship (visit http://www.academyhealth.org/dssfpublications for a full listing of publications stemming from the DSSF program). As an example of continuous program improvement, this year staff implemented a fellow-led monthly call in response to requests to connect fellows. This has proven to be a useful response for fellows to understand how to enhance their own experience by learning from fellows in other systems to incorporate focused areas of development via cross-system sharing. As an indicator of continued value, of the 16 sites that have participated to date, 10 have participated in the DSSF for more than 1 cycle; 3 are new host sites currently participating in their first year of the program; and 3 host sites participated for 1 year. The 3 inaugural host sites that helped launch the DSSF continue to serve as host sites to date.

As a marker of longitudinal success, staff will continue to follow up with preceptors and fellows to understand fellows’ contributions to the host site and the field, as well as impact on the fellow career trajectory. Of the 19 fellow alumni to date, 8 have moved on to academic or research positions, and 11 have remained in care delivery systems to provide local expertise in study design, execution, and evidence uptake.

The program has also made some general contributions to advance the discipline of delivery system science, including:

1. Defining DSS and clarifying training needs for “embedded researchers” and health system analysts. To characterize the fellowship, AcademyHealth and the program advisory committee jointly developed the following definition for delivery system science (DSS):

“DSS includes research that seeks to understand how delivery systems operate, influence, change, and respond to external stimulus, among other topics. DSS may include efforts to examine how and under what circumstances interventions work and how delivery systems effectively implement evidence-based innovations. For the DSSF, DSS is conducted by researchers who are ‘embedded’ in delivery systems and respond to the decision-making needs of those systems.”

Additionally, closer connections with DSS leaders have led to a better understanding of challenges, opportunities, and needs of delivery systems.^10,11

2. Cultivating a network of delivery systems and system leaders interested in expanding the cadre of embedded researchers, and trainees who intend to build careers in DSS. The DSSF also aims to enhance fellows’ skills and knowledge base, career opportunities, and professional network. To extend these relationships and support delivery system analytics, AcademyHealth worked with preceptors and fellows to inform creation of a new Community of Practice supported by AcademyHealth’s EDM Forum and guide planning for AcademyHealth’s Concordium conference to provide a national meeting to showcase DSS.

3. Strategic planning to ensure sustainable support for embedded research within delivery systems. Substantial interest in the program developed quickly, with rapid learning over the first few cycles to refine the program to meet host sites’ and fellows’ needs. Both efforts were critical to demonstrate that the DSSF fulfills an important need for our health system partners and members. As indicated previously, strong, sustained interest from prospective host sites and applicants demonstrates the program has created a win-win to jointly assess fit while building skills and supporting continuous learning.

Likewise, Lisa Simpson, President and CEO of AcademyHealth, and Lucy Savitz, DSSF host site preceptor at Intermountain Healthcare, participated in the Canadian Institutes of Health Research (CIHR) Invitational Workshop, “Modernizing Health Services and Policy Research Training in Canada” in March 2016. Shared learning largely informed by the DSSF led to CIHR creating a similar fellowship program with initial awards to be made in 2017 (see https://www.researchnet-recherchenet.ca/rnr16/vwOpprtntyDtls.do?prog=2540&view=browseActive&sponsor=CIHR-8&type=EXACT&resultCount=25 for more information). We are working to thread these efforts together in a way to leverage our learning community of government agencies, academia, and employers as a long-term funding stream for training in delivery science.

Participant Host Site Level

Two selected examples of how DSSF researchers have engaged high-priority topics that contributed to health system operations are provided here.

Kaiser Permanente Southern California: Assessing adherence with “Choosing Wisely” recommendations in oncology. In partnership with preceptor Dr. Michael Gould, 2013-2014 DSSF fellow Dr. Erin Hahn worked with the Kaiser Permanente Southern California (KPSC) Care Improvement Research Team to lead a project addressing several KPSC priority areas. Focusing on “Choosing Wisely” recommendations from the American Society of Clinical Oncology,¹² the project evaluated appropriateness of imaging and laboratory services for early-stage cancer patients and survivors between and within 2 integrated health systems, Kaiser Permanente (KP) and Intermountain Healthcare.^13,14 Results were presented to KP national leaders, including an external health policy advisory board. In close collaboration with clinical and operational leaders in medical oncology, this multiregional, multisystem project is contributing to targeted quality improvement efforts and improved healthcare value, including audit and feedback of nonrecommended labs.

Dr. Hahn subsequently received a KPSC Incubator Award, a competitive internal grant, to further study factors associated with use of nonrecommended surveillance lab tests for early-stage breast cancer patients. The study focused on medical oncologists within KPSC, categorizing them as high or low utilizers of the tests.¹⁵ Results indicate that high utilizers perceive that the tests help manage patient anxiety about recurrence, while acknowledging that the tests do not provide clinical utility. These findings are contributing to the development of targeted survivorship services across the organization.

Intermountain Healthcare: Formative evaluation of large-scale implementation of shared decision-making. Preceptor Dr. Lucy Savitz assembled a team to conduct a formative evaluation of Intermountain Healthcare’s efforts to implement shared decision-making (SDM) as part of its Center for Medicare and Medicaid Innovation Challenge Award. The 2015-2017 DSSF fellow at Intermountain Healthcare, Dr. Kim Brunisholz, served as a core member of the project team, focusing primarily on a mixed-methods evaluation of the SDM program.

Dr. Brunisholz engaged operational leads, clinical teams, patient and family advisory councils, and senior executives to conduct the program assessment. Results demonstrated significant variation in invited participation in SDM among eligible patient populations: preference sensitive conditions (1 in 30 patients), oncology-related diagnosis (1 in 3 patients), and chronic conditions (1 in 74 patients). Provisional analysis of patient-level clinical outcomes demonstrated that among those invited to the SDM program compared to those that were not, total joint replacement was decreased (10.1% vs 17.3%; P < 0.001) and a trend towards breast conservation emerged (61.8% vs 56.4%; P = 0.10). No difference in treatment choice for lower back pain was observed. Qualitative program analysis suggested need for improvements in the areas of (1) routine and continuous staff training, (2) workflow standardization, and (3) active data monitoring with meaningful, actionable feedback to caregivers. In response to these results, a chartered SDM Steering Committee was created (Dr. Brunisholz is a member of that group) to develop a strategic plan for SDM, with an accompanying organizational response to reimplement SDM in a targeted manner. Learning from this program is being leveraged to support a subproject analysis on a large scale using data from the High Value Healthcare Collaborative as part of an AHRQ-funded Center of Excellence award. (See https://www.ahrq.gov/news/newsroom/press-releases/2015/pcorawards.html. For more information on the High Value Healthcare Collaborative, please visit: https://www.highvaluehealthcare.org.)

Moving forward, the DSSF will continue working with progressive delivery systems. Partnerships between organizations that are interested in integrating rigorous research practice to drive continuous system improvement and maximize the value of care will have substantial need for technical skills and analytic capacity. They will also need to ensure that researchers working in their systems have sufficient understanding of cultural and political context within the organization to be effective leaders who can manage change.

AcademyHealth created the DSSF in response to the field’s request to build a research workforce that reflects the vision for a 21st Century Health System, as laid out by the NAM.¹⁶ We anticipate that as the US Department of Health & Human Services’ goals for payment reform and new measures to promote quality and high-value care are implemented, the DSSF trainees’ skill set will be increasingly valuable and will provide needed thought leadership on strategies to generate and apply evidence in practice.

Disclosure

Ms. Kanani received funding from Intermountain Healthcare, Kaiser Permanente Southern California for support for the Delivery System Science Fellowship. Drs. Hahn, Gould, and Brunisholz have no conflicts to disclose. Dr. Savitz has received funding from HRQ COE, PCORI LHSNet; received funding for lectures from the Institute from Healthcare Improvement, Department of Epidemiology, University of Utah; received funding for travel, accommodations, and meeting expenses from AHRQ NAC, EDM Forum, AH CAPP, AARP NPC, and PROM TEP; and received additional funding from Dartmouth University. Dr. Holve received funding from Intermountain Healthcare, Kaiser Permanente Southern California for support for the Delivery System Science Fellowship, provided by our delivery system partners, several of whom are coauthors on this manuscript.

For over 3 decades, AcademyHealth and its predecessor organizations and members have been studying how the healthcare system works and ways to improve health outcomes. The value of health services research (HSR) training programs that emphasize learning opportunities in delivery system settings was initially articulated at the 2009 AcademyHealth Summit on the Future of HSR Data and Methods.¹ Two years later, the need for such programs was reiterated as a priority for AcademyHealth’s HSR Learning Consortium in their strategic plan.² While HSR methods have become increasingly sophisticated, historical approaches largely relied on extant (usually academic-based) researchers.

To realize the goal of building learning health systems (considered here to be entities where applied, operationally relevant research is systematically designed, generated, and translated into high-quality care delivery), many healthcare organizations have begun to use researchers as an internal resource to inform and support higher quality and more efficient care delivery operations. However, there is a current dearth of scientists trained in research disciplines (eg, comparative effectiveness research, patient-centered outcomes research, implementation science) more directly applicable to operational settings.^3,4 Conducting research within these “real-world” environments is challenging for a variety of well-documented reasons,^5,6 and many important questions cannot be answered using traditional study designs and/or methodologies. Until more researchers are trained in research approaches that align better with care delivery needs, the field will continue to fall short of addressing topics identified by the National Academy of Medicine (NAM; formerly the Institute of Medicine) as priorities for real system improvement.

While researchers in academic and consulting settings play critical roles in knowledge generation, a substantial area for expansion is support for “embedded researchers” who work more directly with operational leaders and understand local context, data, and organizational-level goals of delivery systems. AcademyHealth’s Delivery System Science Fellowship (DSSF) was developed to forge a stronger link between rigorous research practice and pragmatic aspects of care delivery.

Since inception of the DSSF, national attention has further emphasized the need for specialized, experiential learning with specific competencies that extend formal HSR training. The Agency for Healthcare Research and Quality (AHRQ) is currently convening a Technical Expert Panel to guide Training the Next Generation of Learning Health System Researchers as part of its larger effort to provide support for evidence generation and uptake in these applied settings (see http://www.ahrq.gov/news/blog/ahrqviews/supporting-learning-health-systems.html for more information). This work is reinforced by an ongoing project at AcademyHealth focused on Understanding the Current Health Services Research Workforce and Maximizing Its Future,⁷ partially sponsored by AHRQ. These efforts aim to inform AHRQ-funded career and training program requirements in order to build a workforce in partnership with academia and delivery organizations in support of developing high-functioning researchers directly positioned to drive progress toward learning healthcare systems.

The AcademyHealth DSSF Program is a pioneering effort designed to meet the needs of learning healthcare systems for a human capital resource capable of generating insights from operational data and deploying this knowledge effectively. Established in 2012 in partnership with 3 initial host sites, the DSSF provides a paid, postdoctoral training opportunity to help highly qualified, early-career researchers gain applied experience in delivery system settings.⁸ The goal is to provide hands-on experience and professional leadership opportunities to enhance the array of skills needed to generate and apply evidence in delivery systems. The long-term program objective is for graduates to employ the methods and training garnered during the DSSF to produce new, practical insights required to transform healthcare delivery and achieve the “Triple Aim.”⁹

Sixteen delivery systems have participated in the DSSF program over the course of 5 years. These organizations represent a diverse group of innovative, high-performing systems that serve populations across the rural-urban continuum. Host site preceptors are nationally recognized experts in areas such as health economics, comparative effectiveness, pragmatic trials, clinical decision support, and implementation science. The fellowship is guided by an advisory committee that provides strategic direction and plays a key role in selecting fellows. AcademyHealth partners with 8 to 10 health systems annually, accepting new partnerships as interest and availability allows. The Figure summarizes program inputs, elements, antecedents, outputs, and intended outcomes. As a relatively new and evolving program, the DSSF is using a disciplined approach to assess both mid- and long-term outcomes. The Table is a comparative matrix that depicts the range in areas of investment offered by a sample of host sites during the fellowship program. Complementing development of the common core competencies and exposures presented in the Table, specific project choice at a host site is mutually determined according to system needs and the fellow’s interests.

To apply for DSSF, researchers must hold a doctoral degree in any relevant aspect of clinical medicine, HSR, or a related field. The review committee evaluates applicants based on their qualifications, a clear statement of professional goals, appropriateness for placement, the quality of a writing sample, and letters of recommendation. Host sites select a fellow based on their individual preferences and recommendations from the review committee. The minimum duration of the fellowship is 1 year, during which the fellow works full time at the host site. Host sites provide the fellows’ salary and benefits, financial support to attend AcademyHealth’s Annual Research Meeting, and mentorship and/or training. To ensure general continuity across sites, each fellow’s work focuses on “delivery system science (DSS),” with a significant part of the work intended for public dissemination (eg, a conference presentation) or publication in a peer-reviewed journal.

Host sites additionally provide AcademyHealth financial support to manage the process of recruiting promising candidates. AcademyHealth manages and convenes the advisory committee proceedings, facilitates the application cycle (including peer review of applicants), markets the fellowship, guides the interview and match process, and promotes placement of fellows upon program completion. AcademyHealth also convenes active and graduate fellows to foster engagement and professional development.

Program Level

AcademyHealth administers an annual evaluation to assess the program, understand impact on the fellow’s professional development and growth, track publications, and inform programmatic goals. To date, we have identified over 50 peer-reviewed publications resulting from work conducted through the fellowship (visit http://www.academyhealth.org/dssfpublications for a full listing of publications stemming from the DSSF program). As an example of continuous program improvement, this year staff implemented a fellow-led monthly call in response to requests to connect fellows. This has proven to be a useful response for fellows to understand how to enhance their own experience by learning from fellows in other systems to incorporate focused areas of development via cross-system sharing. As an indicator of continued value, of the 16 sites that have participated to date, 10 have participated in the DSSF for more than 1 cycle; 3 are new host sites currently participating in their first year of the program; and 3 host sites participated for 1 year. The 3 inaugural host sites that helped launch the DSSF continue to serve as host sites to date.

As a marker of longitudinal success, staff will continue to follow up with preceptors and fellows to understand fellows’ contributions to the host site and the field, as well as impact on the fellow career trajectory. Of the 19 fellow alumni to date, 8 have moved on to academic or research positions, and 11 have remained in care delivery systems to provide local expertise in study design, execution, and evidence uptake.

The program has also made some general contributions to advance the discipline of delivery system science, including:

1. Defining DSS and clarifying training needs for “embedded researchers” and health system analysts. To characterize the fellowship, AcademyHealth and the program advisory committee jointly developed the following definition for delivery system science (DSS):

“DSS includes research that seeks to understand how delivery systems operate, influence, change, and respond to external stimulus, among other topics. DSS may include efforts to examine how and under what circumstances interventions work and how delivery systems effectively implement evidence-based innovations. For the DSSF, DSS is conducted by researchers who are ‘embedded’ in delivery systems and respond to the decision-making needs of those systems.”

Additionally, closer connections with DSS leaders have led to a better understanding of challenges, opportunities, and needs of delivery systems.^10,11

2. Cultivating a network of delivery systems and system leaders interested in expanding the cadre of embedded researchers, and trainees who intend to build careers in DSS. The DSSF also aims to enhance fellows’ skills and knowledge base, career opportunities, and professional network. To extend these relationships and support delivery system analytics, AcademyHealth worked with preceptors and fellows to inform creation of a new Community of Practice supported by AcademyHealth’s EDM Forum and guide planning for AcademyHealth’s Concordium conference to provide a national meeting to showcase DSS.

3. Strategic planning to ensure sustainable support for embedded research within delivery systems. Substantial interest in the program developed quickly, with rapid learning over the first few cycles to refine the program to meet host sites’ and fellows’ needs. Both efforts were critical to demonstrate that the DSSF fulfills an important need for our health system partners and members. As indicated previously, strong, sustained interest from prospective host sites and applicants demonstrates the program has created a win-win to jointly assess fit while building skills and supporting continuous learning.

Likewise, Lisa Simpson, President and CEO of AcademyHealth, and Lucy Savitz, DSSF host site preceptor at Intermountain Healthcare, participated in the Canadian Institutes of Health Research (CIHR) Invitational Workshop, “Modernizing Health Services and Policy Research Training in Canada” in March 2016. Shared learning largely informed by the DSSF led to CIHR creating a similar fellowship program with initial awards to be made in 2017 (see https://www.researchnet-recherchenet.ca/rnr16/vwOpprtntyDtls.do?prog=2540&view=browseActive&sponsor=CIHR-8&type=EXACT&resultCount=25 for more information). We are working to thread these efforts together in a way to leverage our learning community of government agencies, academia, and employers as a long-term funding stream for training in delivery science.

Participant Host Site Level

Two selected examples of how DSSF researchers have engaged high-priority topics that contributed to health system operations are provided here.

Kaiser Permanente Southern California: Assessing adherence with “Choosing Wisely” recommendations in oncology. In partnership with preceptor Dr. Michael Gould, 2013-2014 DSSF fellow Dr. Erin Hahn worked with the Kaiser Permanente Southern California (KPSC) Care Improvement Research Team to lead a project addressing several KPSC priority areas. Focusing on “Choosing Wisely” recommendations from the American Society of Clinical Oncology,¹² the project evaluated appropriateness of imaging and laboratory services for early-stage cancer patients and survivors between and within 2 integrated health systems, Kaiser Permanente (KP) and Intermountain Healthcare.^13,14 Results were presented to KP national leaders, including an external health policy advisory board. In close collaboration with clinical and operational leaders in medical oncology, this multiregional, multisystem project is contributing to targeted quality improvement efforts and improved healthcare value, including audit and feedback of nonrecommended labs.

Dr. Hahn subsequently received a KPSC Incubator Award, a competitive internal grant, to further study factors associated with use of nonrecommended surveillance lab tests for early-stage breast cancer patients. The study focused on medical oncologists within KPSC, categorizing them as high or low utilizers of the tests.¹⁵ Results indicate that high utilizers perceive that the tests help manage patient anxiety about recurrence, while acknowledging that the tests do not provide clinical utility. These findings are contributing to the development of targeted survivorship services across the organization.

Intermountain Healthcare: Formative evaluation of large-scale implementation of shared decision-making. Preceptor Dr. Lucy Savitz assembled a team to conduct a formative evaluation of Intermountain Healthcare’s efforts to implement shared decision-making (SDM) as part of its Center for Medicare and Medicaid Innovation Challenge Award. The 2015-2017 DSSF fellow at Intermountain Healthcare, Dr. Kim Brunisholz, served as a core member of the project team, focusing primarily on a mixed-methods evaluation of the SDM program.

Dr. Brunisholz engaged operational leads, clinical teams, patient and family advisory councils, and senior executives to conduct the program assessment. Results demonstrated significant variation in invited participation in SDM among eligible patient populations: preference sensitive conditions (1 in 30 patients), oncology-related diagnosis (1 in 3 patients), and chronic conditions (1 in 74 patients). Provisional analysis of patient-level clinical outcomes demonstrated that among those invited to the SDM program compared to those that were not, total joint replacement was decreased (10.1% vs 17.3%; P < 0.001) and a trend towards breast conservation emerged (61.8% vs 56.4%; P = 0.10). No difference in treatment choice for lower back pain was observed. Qualitative program analysis suggested need for improvements in the areas of (1) routine and continuous staff training, (2) workflow standardization, and (3) active data monitoring with meaningful, actionable feedback to caregivers. In response to these results, a chartered SDM Steering Committee was created (Dr. Brunisholz is a member of that group) to develop a strategic plan for SDM, with an accompanying organizational response to reimplement SDM in a targeted manner. Learning from this program is being leveraged to support a subproject analysis on a large scale using data from the High Value Healthcare Collaborative as part of an AHRQ-funded Center of Excellence award. (See https://www.ahrq.gov/news/newsroom/press-releases/2015/pcorawards.html. For more information on the High Value Healthcare Collaborative, please visit: https://www.highvaluehealthcare.org.)

Moving forward, the DSSF will continue working with progressive delivery systems. Partnerships between organizations that are interested in integrating rigorous research practice to drive continuous system improvement and maximize the value of care will have substantial need for technical skills and analytic capacity. They will also need to ensure that researchers working in their systems have sufficient understanding of cultural and political context within the organization to be effective leaders who can manage change.

AcademyHealth created the DSSF in response to the field’s request to build a research workforce that reflects the vision for a 21st Century Health System, as laid out by the NAM.¹⁶ We anticipate that as the US Department of Health & Human Services’ goals for payment reform and new measures to promote quality and high-value care are implemented, the DSSF trainees’ skill set will be increasingly valuable and will provide needed thought leadership on strategies to generate and apply evidence in practice.

Disclosure

Ms. Kanani received funding from Intermountain Healthcare, Kaiser Permanente Southern California for support for the Delivery System Science Fellowship. Drs. Hahn, Gould, and Brunisholz have no conflicts to disclose. Dr. Savitz has received funding from HRQ COE, PCORI LHSNet; received funding for lectures from the Institute from Healthcare Improvement, Department of Epidemiology, University of Utah; received funding for travel, accommodations, and meeting expenses from AHRQ NAC, EDM Forum, AH CAPP, AARP NPC, and PROM TEP; and received additional funding from Dartmouth University. Dr. Holve received funding from Intermountain Healthcare, Kaiser Permanente Southern California for support for the Delivery System Science Fellowship, provided by our delivery system partners, several of whom are coauthors on this manuscript.

Hospitalists must create rapport and communicate large amounts of information in a short amount of time without having a prior relationship with the patient.¹ High-quality communication can improve satisfaction and compliance, while poor communication leaves patients ill prepared to transition back to the community.^2–10

Many medical schools use standardized patients (SPs) to both train and evaluate their students’ communication skills. To our knowledge, no published studies describe using SPs to assess or teach communication skills for hospitalists.

Our objective in this study was to use SPs to assess for deficits in our hospitalists’ communication skills and to determine whether feedback provided by SPs could improve hospitalist confidence in and performance of optimal communication behaviors.

Setting and Participants

Scenario and Checklist Development

We developed 3 SP encounters around common hospitalist-patient interactions: daily rounding, discharge, and interacting with a difficult patient. In order to assess communication skills, we developed a checklist with 3 core domains: Courtesy and Respect, Listen, and Explain. Each domain corresponded to 1 of 3 questions on the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey that pertained to doctor’s communications skills: (1) How often did doctors treat you with courtesy and respect? (2) How often did doctors listen carefully to you? (3) How often did doctors explain things in a way you could understand? We then developed checklist items that corresponded to essential communication skills within each of the 3 domains. These communication skills were based on best practices and published literature.

Discharge Encounter (Table 2): Patient admitted the night before with abdominal pain by another hospitalist. The checklist was based on AIDET^®, an effective communication skills training protocol that our hospitalist group had been trained on.¹¹

Daily Rounding Encounter (Table 3): Patient being discharged after an admission for congestive heart failure. The checklist was developed from the Society of Hospital Medicine discharge toolkit.¹²

Simulated Encounters

All 3 encounters occurred on the same day and each one lasted 1 hour (20 minutes for the encounter, 10 minutes for a posttest survey, and 30 minutes of feedback from the SP). For each case, a task list was given to the hospitalist before walking into the room (Appendix 1). During the feedback session, the SP gave the hospitalist feedback using the case checklist items. They then watched a video of the encounter and the SP further emphasized areas for improvement.

SP Training

SP training consisted of three 3-hour training sessions, which included review of the case, script, guidance on scoring the checklist items, role plays with attending hospitalists, and feedback training. Each SP was assigned to only 1 case.

Seven of the 24 encounters for each SP were reviewed independently by 2 investigators who created a final score for each checklist item which was compared to the SP’s checklist item score. The kappa (k) statistic was used to evaluate inter-observer reliability using the SAS system software (SAS Institute Inc.).

Analysis

The percent of hospitalists who performed each checklist item adequately within in each of the 3 domains (Courtesy and Respect, Listen, and Explain) was calculated. To compare the 3 domains, t tests were used.

We calculated the percent that our hospitalist group received on the 3 HCAHPS doctor’s questions 1 year prior to our SP exercise and 1 year after the SP exercise.

Twenty-three hospitalists completed all 3 encounters. For the 3 domains (Courtesy and Respect, Listen, and Explain), hospitalists performed significantly better in the Listen domain compared to the other 2 domains, with a mean percent adequate score of 90.2 % (95% confidence interval [CI], 72.2%-100%; P < 0.05), and significantly worse in the Explain domain compared to the other 2 domains, with a mean percent adequate score of 65.0% (95% CI, 49.2%-83.6%; P < 0.05). The mean percent adequate score for the Courtesy and Respect domain was 81.6% (95% CI, 56%-100%). This was significantly higher than the Explain domain and significantly lower than the Listen domain.

Posttest survey results showed that hospitalists had an increased level of confidence in their bedside manner, patient satisfaction skills, and high-quality discharge discussion skills.

Inter-Rater Reliability

Inter-rater reliability for the discharge encounter, the daily rounding encounter, and the difficult patient encounter were 0.74 (95% CI, 0.64-0.84), 0.73 (95% CI, 0.63-0.82), and 0.73 (95% CI, 0.63-0.83), respectively.

HCAHPS

Four hundred sixteen HCAHPS surveys were returned in the year prior to our SP exercise, and the percent of patients who answered always to the questions on Courtesy and Respect, Listen, and Explain were 80.4%, 74.2 %, and 69.4 %, respectively. In the year after our SP exercise, 492 surveys were returned, and there was no significant change in HCAHP scores for the group (80.9% for Courtesy and Respect, 70.2% for the Listen question, and 70.5% for Explain).

We have shown that SPs can be used to assess deficits in hospitalist communication skills and provide feedback that can improve hospitalist confidence in performing optimal communication behaviors. We have also shown that hospitalists perceive the exercise as beneficial in improving their communication skills and perceive them as similar to their real patient encounters.

The Explain domain was significantly worse than the Courtesy and Respect and Listen domains for our hospitalists. Analysis of the checklist items within the Explain domain found that the items within this domain that were most problematic for hospitalists were summarizing information at the end of the encounter, using teach-back (a communication confirmation method where a healthcare provider asks a patient to repeat what was said to confirm understanding), encouraging additional questions by using open-ended statements (What questions do you have?) instead of close ended statements (Do you have any questions?), managing team and self-up, setting expectations on length of stay, and timing of tests. This correlated with our patient satisfaction HCAHPS data, which showed that patients consistently rated our hospitalists’ ability to explain things in a way they could understand lowest among the 3 questions. HCAHPS scores did not change after our SP exercise, and this lack of improvement may indicate that meaningful improvement in communication skills requires longitudinal interventions and real-time feedback rather than a single exercise, as was shown in a recent study looking at daily patient satisfaction score feedback given to internal medicine residents.¹⁴

Our study had several limitations. First, hospitalists knew they were being videotaped and observed, which may have altered their behaviors and may not reflect our hospitalists’ actual behaviors with patients. Furthermore, we did not examine whether the feedback given was incorporated into our hospitalists’ daily patient communications and whether this impacted our patients care other than examining HCAHPS scores.

SPs can be used to identify deficiencies in communication skills and provide specific guidance that improves hospitalist confidence in their communication skills.

Acknowledgment

This trial was funded by a grant from The Doctor’s Company Foundation.

Disclosure

None of the authors report any conflicts of interest.

Hospitalists must create rapport and communicate large amounts of information in a short amount of time without having a prior relationship with the patient.¹ High-quality communication can improve satisfaction and compliance, while poor communication leaves patients ill prepared to transition back to the community.^2–10

Many medical schools use standardized patients (SPs) to both train and evaluate their students’ communication skills. To our knowledge, no published studies describe using SPs to assess or teach communication skills for hospitalists.

Our objective in this study was to use SPs to assess for deficits in our hospitalists’ communication skills and to determine whether feedback provided by SPs could improve hospitalist confidence in and performance of optimal communication behaviors.

Setting and Participants

Scenario and Checklist Development

We developed 3 SP encounters around common hospitalist-patient interactions: daily rounding, discharge, and interacting with a difficult patient. In order to assess communication skills, we developed a checklist with 3 core domains: Courtesy and Respect, Listen, and Explain. Each domain corresponded to 1 of 3 questions on the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey that pertained to doctor’s communications skills: (1) How often did doctors treat you with courtesy and respect? (2) How often did doctors listen carefully to you? (3) How often did doctors explain things in a way you could understand? We then developed checklist items that corresponded to essential communication skills within each of the 3 domains. These communication skills were based on best practices and published literature.

Discharge Encounter (Table 2): Patient admitted the night before with abdominal pain by another hospitalist. The checklist was based on AIDET^®, an effective communication skills training protocol that our hospitalist group had been trained on.¹¹

Daily Rounding Encounter (Table 3): Patient being discharged after an admission for congestive heart failure. The checklist was developed from the Society of Hospital Medicine discharge toolkit.¹²

Simulated Encounters

All 3 encounters occurred on the same day and each one lasted 1 hour (20 minutes for the encounter, 10 minutes for a posttest survey, and 30 minutes of feedback from the SP). For each case, a task list was given to the hospitalist before walking into the room (Appendix 1). During the feedback session, the SP gave the hospitalist feedback using the case checklist items. They then watched a video of the encounter and the SP further emphasized areas for improvement.

SP Training

SP training consisted of three 3-hour training sessions, which included review of the case, script, guidance on scoring the checklist items, role plays with attending hospitalists, and feedback training. Each SP was assigned to only 1 case.

Seven of the 24 encounters for each SP were reviewed independently by 2 investigators who created a final score for each checklist item which was compared to the SP’s checklist item score. The kappa (k) statistic was used to evaluate inter-observer reliability using the SAS system software (SAS Institute Inc.).

Analysis

The percent of hospitalists who performed each checklist item adequately within in each of the 3 domains (Courtesy and Respect, Listen, and Explain) was calculated. To compare the 3 domains, t tests were used.

We calculated the percent that our hospitalist group received on the 3 HCAHPS doctor’s questions 1 year prior to our SP exercise and 1 year after the SP exercise.

Twenty-three hospitalists completed all 3 encounters. For the 3 domains (Courtesy and Respect, Listen, and Explain), hospitalists performed significantly better in the Listen domain compared to the other 2 domains, with a mean percent adequate score of 90.2 % (95% confidence interval [CI], 72.2%-100%; P < 0.05), and significantly worse in the Explain domain compared to the other 2 domains, with a mean percent adequate score of 65.0% (95% CI, 49.2%-83.6%; P < 0.05). The mean percent adequate score for the Courtesy and Respect domain was 81.6% (95% CI, 56%-100%). This was significantly higher than the Explain domain and significantly lower than the Listen domain.

Posttest survey results showed that hospitalists had an increased level of confidence in their bedside manner, patient satisfaction skills, and high-quality discharge discussion skills.

Inter-Rater Reliability

Inter-rater reliability for the discharge encounter, the daily rounding encounter, and the difficult patient encounter were 0.74 (95% CI, 0.64-0.84), 0.73 (95% CI, 0.63-0.82), and 0.73 (95% CI, 0.63-0.83), respectively.

HCAHPS

Four hundred sixteen HCAHPS surveys were returned in the year prior to our SP exercise, and the percent of patients who answered always to the questions on Courtesy and Respect, Listen, and Explain were 80.4%, 74.2 %, and 69.4 %, respectively. In the year after our SP exercise, 492 surveys were returned, and there was no significant change in HCAHP scores for the group (80.9% for Courtesy and Respect, 70.2% for the Listen question, and 70.5% for Explain).

We have shown that SPs can be used to assess deficits in hospitalist communication skills and provide feedback that can improve hospitalist confidence in performing optimal communication behaviors. We have also shown that hospitalists perceive the exercise as beneficial in improving their communication skills and perceive them as similar to their real patient encounters.

The Explain domain was significantly worse than the Courtesy and Respect and Listen domains for our hospitalists. Analysis of the checklist items within the Explain domain found that the items within this domain that were most problematic for hospitalists were summarizing information at the end of the encounter, using teach-back (a communication confirmation method where a healthcare provider asks a patient to repeat what was said to confirm understanding), encouraging additional questions by using open-ended statements (What questions do you have?) instead of close ended statements (Do you have any questions?), managing team and self-up, setting expectations on length of stay, and timing of tests. This correlated with our patient satisfaction HCAHPS data, which showed that patients consistently rated our hospitalists’ ability to explain things in a way they could understand lowest among the 3 questions. HCAHPS scores did not change after our SP exercise, and this lack of improvement may indicate that meaningful improvement in communication skills requires longitudinal interventions and real-time feedback rather than a single exercise, as was shown in a recent study looking at daily patient satisfaction score feedback given to internal medicine residents.¹⁴

Our study had several limitations. First, hospitalists knew they were being videotaped and observed, which may have altered their behaviors and may not reflect our hospitalists’ actual behaviors with patients. Furthermore, we did not examine whether the feedback given was incorporated into our hospitalists’ daily patient communications and whether this impacted our patients care other than examining HCAHPS scores.

SPs can be used to identify deficiencies in communication skills and provide specific guidance that improves hospitalist confidence in their communication skills.

Acknowledgment

This trial was funded by a grant from The Doctor’s Company Foundation.

Disclosure

None of the authors report any conflicts of interest.

Hospitalists must create rapport and communicate large amounts of information in a short amount of time without having a prior relationship with the patient.¹ High-quality communication can improve satisfaction and compliance, while poor communication leaves patients ill prepared to transition back to the community.^2–10

Many medical schools use standardized patients (SPs) to both train and evaluate their students’ communication skills. To our knowledge, no published studies describe using SPs to assess or teach communication skills for hospitalists.

Our objective in this study was to use SPs to assess for deficits in our hospitalists’ communication skills and to determine whether feedback provided by SPs could improve hospitalist confidence in and performance of optimal communication behaviors.

Setting and Participants

Scenario and Checklist Development

We developed 3 SP encounters around common hospitalist-patient interactions: daily rounding, discharge, and interacting with a difficult patient. In order to assess communication skills, we developed a checklist with 3 core domains: Courtesy and Respect, Listen, and Explain. Each domain corresponded to 1 of 3 questions on the Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey that pertained to doctor’s communications skills: (1) How often did doctors treat you with courtesy and respect? (2) How often did doctors listen carefully to you? (3) How often did doctors explain things in a way you could understand? We then developed checklist items that corresponded to essential communication skills within each of the 3 domains. These communication skills were based on best practices and published literature.

Discharge Encounter (Table 2): Patient admitted the night before with abdominal pain by another hospitalist. The checklist was based on AIDET^®, an effective communication skills training protocol that our hospitalist group had been trained on.¹¹

Daily Rounding Encounter (Table 3): Patient being discharged after an admission for congestive heart failure. The checklist was developed from the Society of Hospital Medicine discharge toolkit.¹²

Simulated Encounters

All 3 encounters occurred on the same day and each one lasted 1 hour (20 minutes for the encounter, 10 minutes for a posttest survey, and 30 minutes of feedback from the SP). For each case, a task list was given to the hospitalist before walking into the room (Appendix 1). During the feedback session, the SP gave the hospitalist feedback using the case checklist items. They then watched a video of the encounter and the SP further emphasized areas for improvement.

SP Training

SP training consisted of three 3-hour training sessions, which included review of the case, script, guidance on scoring the checklist items, role plays with attending hospitalists, and feedback training. Each SP was assigned to only 1 case.

Seven of the 24 encounters for each SP were reviewed independently by 2 investigators who created a final score for each checklist item which was compared to the SP’s checklist item score. The kappa (k) statistic was used to evaluate inter-observer reliability using the SAS system software (SAS Institute Inc.).

Analysis

The percent of hospitalists who performed each checklist item adequately within in each of the 3 domains (Courtesy and Respect, Listen, and Explain) was calculated. To compare the 3 domains, t tests were used.

We calculated the percent that our hospitalist group received on the 3 HCAHPS doctor’s questions 1 year prior to our SP exercise and 1 year after the SP exercise.

Twenty-three hospitalists completed all 3 encounters. For the 3 domains (Courtesy and Respect, Listen, and Explain), hospitalists performed significantly better in the Listen domain compared to the other 2 domains, with a mean percent adequate score of 90.2 % (95% confidence interval [CI], 72.2%-100%; P < 0.05), and significantly worse in the Explain domain compared to the other 2 domains, with a mean percent adequate score of 65.0% (95% CI, 49.2%-83.6%; P < 0.05). The mean percent adequate score for the Courtesy and Respect domain was 81.6% (95% CI, 56%-100%). This was significantly higher than the Explain domain and significantly lower than the Listen domain.

Posttest survey results showed that hospitalists had an increased level of confidence in their bedside manner, patient satisfaction skills, and high-quality discharge discussion skills.

Inter-Rater Reliability

Inter-rater reliability for the discharge encounter, the daily rounding encounter, and the difficult patient encounter were 0.74 (95% CI, 0.64-0.84), 0.73 (95% CI, 0.63-0.82), and 0.73 (95% CI, 0.63-0.83), respectively.

HCAHPS

Four hundred sixteen HCAHPS surveys were returned in the year prior to our SP exercise, and the percent of patients who answered always to the questions on Courtesy and Respect, Listen, and Explain were 80.4%, 74.2 %, and 69.4 %, respectively. In the year after our SP exercise, 492 surveys were returned, and there was no significant change in HCAHP scores for the group (80.9% for Courtesy and Respect, 70.2% for the Listen question, and 70.5% for Explain).

We have shown that SPs can be used to assess deficits in hospitalist communication skills and provide feedback that can improve hospitalist confidence in performing optimal communication behaviors. We have also shown that hospitalists perceive the exercise as beneficial in improving their communication skills and perceive them as similar to their real patient encounters.

The Explain domain was significantly worse than the Courtesy and Respect and Listen domains for our hospitalists. Analysis of the checklist items within the Explain domain found that the items within this domain that were most problematic for hospitalists were summarizing information at the end of the encounter, using teach-back (a communication confirmation method where a healthcare provider asks a patient to repeat what was said to confirm understanding), encouraging additional questions by using open-ended statements (What questions do you have?) instead of close ended statements (Do you have any questions?), managing team and self-up, setting expectations on length of stay, and timing of tests. This correlated with our patient satisfaction HCAHPS data, which showed that patients consistently rated our hospitalists’ ability to explain things in a way they could understand lowest among the 3 questions. HCAHPS scores did not change after our SP exercise, and this lack of improvement may indicate that meaningful improvement in communication skills requires longitudinal interventions and real-time feedback rather than a single exercise, as was shown in a recent study looking at daily patient satisfaction score feedback given to internal medicine residents.¹⁴

Our study had several limitations. First, hospitalists knew they were being videotaped and observed, which may have altered their behaviors and may not reflect our hospitalists’ actual behaviors with patients. Furthermore, we did not examine whether the feedback given was incorporated into our hospitalists’ daily patient communications and whether this impacted our patients care other than examining HCAHPS scores.

SPs can be used to identify deficiencies in communication skills and provide specific guidance that improves hospitalist confidence in their communication skills.

Acknowledgment

This trial was funded by a grant from The Doctor’s Company Foundation.

Disclosure

None of the authors report any conflicts of interest.

The “Things We Do for No Reason” (TWDFNR) series reviews practices which have become common parts of hospital care but which may provide little value to our patients. Practices reviewed in the TWDFNR series do not represent “black and white” conclusions or clinical practice standards, but are meant as a starting place for research and active discussions among hospitalists and patients. We invite you to be part of that discussion. https://www.choosingwisely.org/

A 47-year-old man with a history of alcohol abuse, cirrhosis, and grade II esophageal varices is admitted for treatment of alcohol withdrawal. He reports having some dark-colored stools a week prior to admission, but his stools since then have been normal in color. A repeat hemoglobin is stable, but a fecal occult blood test is positive. What should be done next?

The US Preventive Services Task Force and the American College of Gastroenterology recommend fecal occult blood testing (FOBT) as one method for colorectal cancer (CRC) screening in average risk populations.^1,2 FOBTs can be divided into guaiac-based tests (gFOBTs), which measure heme, and fecal immunochemical tests (FITs), which measure the globin portion of human hemoglobin (Hb). In gFOBTs, heme present in the sample reacts with a hydrogen peroxide developer to oxidize guaiac, producing a blue color.³ Screening gFOBT was shown to decrease mortality from CRC in several landmark studies in the 1990s, but its sensitivity is poor, ranging from 30% to 57%.⁴ Because the guaiac-induced color change is determined visually, interpretation of gFOBT results are subject to error. In a survey of 173 medical providers, 12% did not accurately interpret gFOBT results.⁵ In light of these limitations, recent guidelines support the use of newer FITs for CRC screening. FITs utilize antibodies directed against the human globin moiety and demonstrate an increased sensitivity when compared with gFOBTs (by 32% to 62%) for detecting neoplasm.⁶ While evidence supports the use of FOBTs in CRC screening, providers use these tests for nonvalidated purposes, including the evaluation of suspected acute upper gastrointestinal bleeding (UGIB).

Given the incidence (up to 100 per 100,000 persons per year) and high mortality of UGIB (up to 20,000 deaths annually in the United States),⁷ there would ideally be a noninvasive test available to help guide management. In evaluating a patient with possible acute UGIB, FOBT affords several theoretical benefits. FOBT is quick, inexpensive, and can be performed by any health professional. In contrast, the primary diagnostic procedure for UGIB, esophagogastroduodenoscopy (EGD), carries procedural and sedation-related risks, can be costly and time-consuming, and requires consultation from subspecialty providers.

While FOBTs are valuable as screening tests for CRC in the outpatient setting, their use has been extended to diagnose gastrointestinal (GI) bleeding in the inpatient setting without supporting data. As is true for many screening tests, FOBT is associated with a high incidence of false-positive results, or type I errors.^8,9 False-positive FOBT results can occur from ingested blood via extra-intestinal sources (eg, epistaxis, gingival bleeding, pharyngitis, hemoptysis), or in medical conditions with intestinal mucosal inflammation (eg, esophagitis, gastritis, inflammatory bowel disease). False-positive results can also be due to clinically insignificant GI blood loss induced by medications (eg, aspirin, nonsteroidal anti-inflammatory drugs), alcohol,¹⁰ or by ingestion of meats, fruits, or vegetables containing peroxidase (eg, broccoli, cauliflower).¹¹

Outpatients using FOBTs for cancer screening are advised to hold medications and avoid foods that may lead to false-positive results. Despite institution of these restrictions, false-positive rates are still high, as 37% to 53% of CRC screening patients with a positive FOBT have a subsequent negative colonoscopy, and only 11% to 21% of these patients have a source of bleeding identified on subsequent EGD.¹² False-positive results might be even higher in the inpatient setting, where patients typically do not adhere to these restrictions. A review of FOBTs performed in 3 acute care hospitals revealed that 65% of patients tested were on at least one medication that impacted the validity of gFOBT results, and 98% had no evidence of dietary restriction prior to testing.¹³

The use of FOBTs (particularly FITs) is also subject to false-negative results, or type II errors. While FITs have increased specificity for lower GI bleeding, their ability to detect UGIB is limited, because most Hb is digested in the small intestine and not present in rectal stool.¹⁴ In a study of more than 2,700 patients, FIT results were not correlated with the presence of upper GI pathology.¹⁵ False-negative results are less common with gFOBTs, although these may occur with low volume, slow or intermittent bleeding,¹⁶ or with ingestion of substances that inhibit oxidation, such as vitamin C.¹⁷

Beyond these test limitations, studies suggest that the majority of inpatient FOBT results do not impact immediate medical decision-making or management. In one study, only 34% of hospitalized patients with a positive FOBT underwent further GI studies, with the majority of those patients (60%) receiving endoscopy before the results of the FOBT were known.¹⁸ In another study of 201 FOBTs performed on hospitalized patients, those with negative results underwent further GI evaluation at a higher rate than those with positive results (41% vs 38%).⁸ This aligns with a study that revealed the majority of patients suspected of having a GI bleed underwent endoscopic evaluation regardless of the FOBT result.⁹

FOBT currently has a role in CRC screening and may have a role in the evaluation of anemia of unknown etiology to evaluate for occult GIB, although the yield is likely low.¹³ In one retrospective analysis of inpatients with unexplained anemia, 43.6% of FOBTs were positive, but a potential GI cause was found in only 6.8% of patients.⁹ Patients with anemia from an unknown etiology should have a workup based on the history, physical, and complete blood count indices. While iron deficiency anemia warrants eventual evaluation for occult blood loss, noncritical anemia in an otherwise stable patient does not require an inpatient evaluation. When FOBT is used in the outpatient setting, patients can be counseled on proper dietary and medication modifications prior to testing.

A careful history, physical examination, and visual inspection of the stool remain the foundation of establishing UGIB as the etiology of anemia. Observed melena (either by passed stool or a rectal examination) has a likelihood ratio (LR) of 25 for UGIB; a patient’s self-report of stools that sounds melenic (black or tarry) has an LR of 5-6.¹⁹ An upper GI source may be further supported by an elevated blood urea nitrogen (BUN) to creatinine ratio, as blood is absorbed through the small bowel and patients may have concomitant decreased renal perfusion. A BUN to creatinine ratio of >30 is associated with a positive LR (LR+) of 7.5 for UGIB.¹⁹ Recall that the higher the LR+, and the lower the negative LR (LR-), the better the test is at ruling in and out the diagnosis, respectively. LR+ of 2–10 and LR– of 0.1–0.5 represent a modestly helpful diagnostic test, whereas LR+ >10 and LR- <0.1 are considered robust. These are generalizations only, as value of LR+/LR- depends on pretest probability.

Clinical decision tools, such as the Glasgow-Blatchford and Rockall scores, utilize the history, physical examination, laboratory results, and pretest probability for high-grade peptic ulcer stigmata to estimate the severity of an UGIB and risk for adverse outcomes, respectively. Notably, these scoring systems do not include FOBT results. Despite the relatively inexpensive cost per FOBT ($3.03 per unit),²⁰ this test’s poor specificity when used in the inpatient setting has the potential to lead to significant, unnecessary downstream expense (as well as the potential for procedural risk and anxiety for patients). Given that the incidence of acute UGIB is approximately 100 per 100,000 persons per year,⁷ based on the United States population in 2016,²¹ there were 323,936 patients with UGIB. If each patient underwent an FOBT, the direct expense would be nearly a million dollars. Nonetheless, the number of patients getting a FOBT in the inpatient setting for a suspected UGIB (or for other indications) is unknown, and the direct costs of the tests itself likely represent a fraction of the healthcare expenditures associated with this practice. Allowing that only a third of patients with positive FOBTs in the inpatient setting typically undergo EGD,²² overuse of this test would lead to a high number of unnecessary EGDs, and potentially colonoscopies or additional diagnostic procedures (eg, capsule endoscopy). In light of the false-positive results associated with FOBT, and lack of diagnostic utility, this brief cost analysis suggests FOBT is a low-value test for suspected UGIB in the inpatient setting, and there are potential significant cost savings if FOBTs are withheld.

Although Gastroccult²³ may be considered for the detection of occult blood in gastric juice, its package insert states: “As with any occult blood test, results with the Gastroccult test cannot be considered conclusive evidence of the presence or absence of upper gastrointestinal bleeding or pathology.” As with any diagnostic evaluation, we would only recommend this test if it would change management.

• FOBT should not be performed to diagnose UGIB.
• When there is clinical suspicion of acute GI bleeding, the best diagnostic tools are a good history, physical examination, and visual inspection of the stool by the clinician to determine the presence of hematochezia or melena.
• Deferring FOBT to the ambulatory setting may improve test performance characteristics.

Revisiting our patient, for all of the reasons discussed above, there is no indication for FOBT as it would not affect management. Based on a careful history and physical examination, our patient would likely require upper endoscopy either as an inpatient or an outpatient depending on his clinical course.

FOBT is validated as an outpatient colon cancer screening tool in asymptomatic patients, not for inpatient evaluation of acute GIB. Given the poor positive predictive value for a positive FOBT in an acute GIB scenario, the potential risk for unnecessary treatments or procedures is real. Conversely, a negative FOBT (particularly FIT) does not rule out GI bleeding and risks a false sense of security that may result in under-treatment. In most scenarios in which FOBT is performed, clinicians can make decisions based on a composite of history, physical exam, visual inspection of the stool, and laboratory investigation. Until further research substantiates the utility of FOBT for this purpose, we would recommend against the routine use of FOBT for evaluating UGIB in hospitalized patients.

Acknowledgment

The authors would like to thank and acknowledge the team from Orlando Health for their review of this manuscript.

Disclosure: The authors do not have any relevant financial disclosures to report. The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs.

Do you think this is a low-value practice? Is this truly a “Thing We Do for No Reason”? Let us know what you do in your practice and propose ideas for other “Things We Do for No Reason” topics. Please join in the conversation online at Twitter (#TWDFNR)/Facebook and don’t forget to “Like It” on Facebook or retweet it on Twitter. We invite you to propose ideas for other “Things We Do for No Reason” topics by [email protected].

The “Things We Do for No Reason” (TWDFNR) series reviews practices which have become common parts of hospital care but which may provide little value to our patients. Practices reviewed in the TWDFNR series do not represent “black and white” conclusions or clinical practice standards, but are meant as a starting place for research and active discussions among hospitalists and patients. We invite you to be part of that discussion. https://www.choosingwisely.org/

A 47-year-old man with a history of alcohol abuse, cirrhosis, and grade II esophageal varices is admitted for treatment of alcohol withdrawal. He reports having some dark-colored stools a week prior to admission, but his stools since then have been normal in color. A repeat hemoglobin is stable, but a fecal occult blood test is positive. What should be done next?

The US Preventive Services Task Force and the American College of Gastroenterology recommend fecal occult blood testing (FOBT) as one method for colorectal cancer (CRC) screening in average risk populations.^1,2 FOBTs can be divided into guaiac-based tests (gFOBTs), which measure heme, and fecal immunochemical tests (FITs), which measure the globin portion of human hemoglobin (Hb). In gFOBTs, heme present in the sample reacts with a hydrogen peroxide developer to oxidize guaiac, producing a blue color.³ Screening gFOBT was shown to decrease mortality from CRC in several landmark studies in the 1990s, but its sensitivity is poor, ranging from 30% to 57%.⁴ Because the guaiac-induced color change is determined visually, interpretation of gFOBT results are subject to error. In a survey of 173 medical providers, 12% did not accurately interpret gFOBT results.⁵ In light of these limitations, recent guidelines support the use of newer FITs for CRC screening. FITs utilize antibodies directed against the human globin moiety and demonstrate an increased sensitivity when compared with gFOBTs (by 32% to 62%) for detecting neoplasm.⁶ While evidence supports the use of FOBTs in CRC screening, providers use these tests for nonvalidated purposes, including the evaluation of suspected acute upper gastrointestinal bleeding (UGIB).

Given the incidence (up to 100 per 100,000 persons per year) and high mortality of UGIB (up to 20,000 deaths annually in the United States),⁷ there would ideally be a noninvasive test available to help guide management. In evaluating a patient with possible acute UGIB, FOBT affords several theoretical benefits. FOBT is quick, inexpensive, and can be performed by any health professional. In contrast, the primary diagnostic procedure for UGIB, esophagogastroduodenoscopy (EGD), carries procedural and sedation-related risks, can be costly and time-consuming, and requires consultation from subspecialty providers.

While FOBTs are valuable as screening tests for CRC in the outpatient setting, their use has been extended to diagnose gastrointestinal (GI) bleeding in the inpatient setting without supporting data. As is true for many screening tests, FOBT is associated with a high incidence of false-positive results, or type I errors.^8,9 False-positive FOBT results can occur from ingested blood via extra-intestinal sources (eg, epistaxis, gingival bleeding, pharyngitis, hemoptysis), or in medical conditions with intestinal mucosal inflammation (eg, esophagitis, gastritis, inflammatory bowel disease). False-positive results can also be due to clinically insignificant GI blood loss induced by medications (eg, aspirin, nonsteroidal anti-inflammatory drugs), alcohol,¹⁰ or by ingestion of meats, fruits, or vegetables containing peroxidase (eg, broccoli, cauliflower).¹¹

Outpatients using FOBTs for cancer screening are advised to hold medications and avoid foods that may lead to false-positive results. Despite institution of these restrictions, false-positive rates are still high, as 37% to 53% of CRC screening patients with a positive FOBT have a subsequent negative colonoscopy, and only 11% to 21% of these patients have a source of bleeding identified on subsequent EGD.¹² False-positive results might be even higher in the inpatient setting, where patients typically do not adhere to these restrictions. A review of FOBTs performed in 3 acute care hospitals revealed that 65% of patients tested were on at least one medication that impacted the validity of gFOBT results, and 98% had no evidence of dietary restriction prior to testing.¹³

The use of FOBTs (particularly FITs) is also subject to false-negative results, or type II errors. While FITs have increased specificity for lower GI bleeding, their ability to detect UGIB is limited, because most Hb is digested in the small intestine and not present in rectal stool.¹⁴ In a study of more than 2,700 patients, FIT results were not correlated with the presence of upper GI pathology.¹⁵ False-negative results are less common with gFOBTs, although these may occur with low volume, slow or intermittent bleeding,¹⁶ or with ingestion of substances that inhibit oxidation, such as vitamin C.¹⁷

Beyond these test limitations, studies suggest that the majority of inpatient FOBT results do not impact immediate medical decision-making or management. In one study, only 34% of hospitalized patients with a positive FOBT underwent further GI studies, with the majority of those patients (60%) receiving endoscopy before the results of the FOBT were known.¹⁸ In another study of 201 FOBTs performed on hospitalized patients, those with negative results underwent further GI evaluation at a higher rate than those with positive results (41% vs 38%).⁸ This aligns with a study that revealed the majority of patients suspected of having a GI bleed underwent endoscopic evaluation regardless of the FOBT result.⁹

FOBT currently has a role in CRC screening and may have a role in the evaluation of anemia of unknown etiology to evaluate for occult GIB, although the yield is likely low.¹³ In one retrospective analysis of inpatients with unexplained anemia, 43.6% of FOBTs were positive, but a potential GI cause was found in only 6.8% of patients.⁹ Patients with anemia from an unknown etiology should have a workup based on the history, physical, and complete blood count indices. While iron deficiency anemia warrants eventual evaluation for occult blood loss, noncritical anemia in an otherwise stable patient does not require an inpatient evaluation. When FOBT is used in the outpatient setting, patients can be counseled on proper dietary and medication modifications prior to testing.

A careful history, physical examination, and visual inspection of the stool remain the foundation of establishing UGIB as the etiology of anemia. Observed melena (either by passed stool or a rectal examination) has a likelihood ratio (LR) of 25 for UGIB; a patient’s self-report of stools that sounds melenic (black or tarry) has an LR of 5-6.¹⁹ An upper GI source may be further supported by an elevated blood urea nitrogen (BUN) to creatinine ratio, as blood is absorbed through the small bowel and patients may have concomitant decreased renal perfusion. A BUN to creatinine ratio of >30 is associated with a positive LR (LR+) of 7.5 for UGIB.¹⁹ Recall that the higher the LR+, and the lower the negative LR (LR-), the better the test is at ruling in and out the diagnosis, respectively. LR+ of 2–10 and LR– of 0.1–0.5 represent a modestly helpful diagnostic test, whereas LR+ >10 and LR- <0.1 are considered robust. These are generalizations only, as value of LR+/LR- depends on pretest probability.

Clinical decision tools, such as the Glasgow-Blatchford and Rockall scores, utilize the history, physical examination, laboratory results, and pretest probability for high-grade peptic ulcer stigmata to estimate the severity of an UGIB and risk for adverse outcomes, respectively. Notably, these scoring systems do not include FOBT results. Despite the relatively inexpensive cost per FOBT ($3.03 per unit),²⁰ this test’s poor specificity when used in the inpatient setting has the potential to lead to significant, unnecessary downstream expense (as well as the potential for procedural risk and anxiety for patients). Given that the incidence of acute UGIB is approximately 100 per 100,000 persons per year,⁷ based on the United States population in 2016,²¹ there were 323,936 patients with UGIB. If each patient underwent an FOBT, the direct expense would be nearly a million dollars. Nonetheless, the number of patients getting a FOBT in the inpatient setting for a suspected UGIB (or for other indications) is unknown, and the direct costs of the tests itself likely represent a fraction of the healthcare expenditures associated with this practice. Allowing that only a third of patients with positive FOBTs in the inpatient setting typically undergo EGD,²² overuse of this test would lead to a high number of unnecessary EGDs, and potentially colonoscopies or additional diagnostic procedures (eg, capsule endoscopy). In light of the false-positive results associated with FOBT, and lack of diagnostic utility, this brief cost analysis suggests FOBT is a low-value test for suspected UGIB in the inpatient setting, and there are potential significant cost savings if FOBTs are withheld.

Although Gastroccult²³ may be considered for the detection of occult blood in gastric juice, its package insert states: “As with any occult blood test, results with the Gastroccult test cannot be considered conclusive evidence of the presence or absence of upper gastrointestinal bleeding or pathology.” As with any diagnostic evaluation, we would only recommend this test if it would change management.

• FOBT should not be performed to diagnose UGIB.
• When there is clinical suspicion of acute GI bleeding, the best diagnostic tools are a good history, physical examination, and visual inspection of the stool by the clinician to determine the presence of hematochezia or melena.
• Deferring FOBT to the ambulatory setting may improve test performance characteristics.

Revisiting our patient, for all of the reasons discussed above, there is no indication for FOBT as it would not affect management. Based on a careful history and physical examination, our patient would likely require upper endoscopy either as an inpatient or an outpatient depending on his clinical course.

FOBT is validated as an outpatient colon cancer screening tool in asymptomatic patients, not for inpatient evaluation of acute GIB. Given the poor positive predictive value for a positive FOBT in an acute GIB scenario, the potential risk for unnecessary treatments or procedures is real. Conversely, a negative FOBT (particularly FIT) does not rule out GI bleeding and risks a false sense of security that may result in under-treatment. In most scenarios in which FOBT is performed, clinicians can make decisions based on a composite of history, physical exam, visual inspection of the stool, and laboratory investigation. Until further research substantiates the utility of FOBT for this purpose, we would recommend against the routine use of FOBT for evaluating UGIB in hospitalized patients.

Acknowledgment

The authors would like to thank and acknowledge the team from Orlando Health for their review of this manuscript.

Disclosure: The authors do not have any relevant financial disclosures to report. The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs.

Do you think this is a low-value practice? Is this truly a “Thing We Do for No Reason”? Let us know what you do in your practice and propose ideas for other “Things We Do for No Reason” topics. Please join in the conversation online at Twitter (#TWDFNR)/Facebook and don’t forget to “Like It” on Facebook or retweet it on Twitter. We invite you to propose ideas for other “Things We Do for No Reason” topics by [email protected].

The “Things We Do for No Reason” (TWDFNR) series reviews practices which have become common parts of hospital care but which may provide little value to our patients. Practices reviewed in the TWDFNR series do not represent “black and white” conclusions or clinical practice standards, but are meant as a starting place for research and active discussions among hospitalists and patients. We invite you to be part of that discussion. https://www.choosingwisely.org/

A 47-year-old man with a history of alcohol abuse, cirrhosis, and grade II esophageal varices is admitted for treatment of alcohol withdrawal. He reports having some dark-colored stools a week prior to admission, but his stools since then have been normal in color. A repeat hemoglobin is stable, but a fecal occult blood test is positive. What should be done next?

The US Preventive Services Task Force and the American College of Gastroenterology recommend fecal occult blood testing (FOBT) as one method for colorectal cancer (CRC) screening in average risk populations.^1,2 FOBTs can be divided into guaiac-based tests (gFOBTs), which measure heme, and fecal immunochemical tests (FITs), which measure the globin portion of human hemoglobin (Hb). In gFOBTs, heme present in the sample reacts with a hydrogen peroxide developer to oxidize guaiac, producing a blue color.³ Screening gFOBT was shown to decrease mortality from CRC in several landmark studies in the 1990s, but its sensitivity is poor, ranging from 30% to 57%.⁴ Because the guaiac-induced color change is determined visually, interpretation of gFOBT results are subject to error. In a survey of 173 medical providers, 12% did not accurately interpret gFOBT results.⁵ In light of these limitations, recent guidelines support the use of newer FITs for CRC screening. FITs utilize antibodies directed against the human globin moiety and demonstrate an increased sensitivity when compared with gFOBTs (by 32% to 62%) for detecting neoplasm.⁶ While evidence supports the use of FOBTs in CRC screening, providers use these tests for nonvalidated purposes, including the evaluation of suspected acute upper gastrointestinal bleeding (UGIB).

Given the incidence (up to 100 per 100,000 persons per year) and high mortality of UGIB (up to 20,000 deaths annually in the United States),⁷ there would ideally be a noninvasive test available to help guide management. In evaluating a patient with possible acute UGIB, FOBT affords several theoretical benefits. FOBT is quick, inexpensive, and can be performed by any health professional. In contrast, the primary diagnostic procedure for UGIB, esophagogastroduodenoscopy (EGD), carries procedural and sedation-related risks, can be costly and time-consuming, and requires consultation from subspecialty providers.

While FOBTs are valuable as screening tests for CRC in the outpatient setting, their use has been extended to diagnose gastrointestinal (GI) bleeding in the inpatient setting without supporting data. As is true for many screening tests, FOBT is associated with a high incidence of false-positive results, or type I errors.^8,9 False-positive FOBT results can occur from ingested blood via extra-intestinal sources (eg, epistaxis, gingival bleeding, pharyngitis, hemoptysis), or in medical conditions with intestinal mucosal inflammation (eg, esophagitis, gastritis, inflammatory bowel disease). False-positive results can also be due to clinically insignificant GI blood loss induced by medications (eg, aspirin, nonsteroidal anti-inflammatory drugs), alcohol,¹⁰ or by ingestion of meats, fruits, or vegetables containing peroxidase (eg, broccoli, cauliflower).¹¹

Outpatients using FOBTs for cancer screening are advised to hold medications and avoid foods that may lead to false-positive results. Despite institution of these restrictions, false-positive rates are still high, as 37% to 53% of CRC screening patients with a positive FOBT have a subsequent negative colonoscopy, and only 11% to 21% of these patients have a source of bleeding identified on subsequent EGD.¹² False-positive results might be even higher in the inpatient setting, where patients typically do not adhere to these restrictions. A review of FOBTs performed in 3 acute care hospitals revealed that 65% of patients tested were on at least one medication that impacted the validity of gFOBT results, and 98% had no evidence of dietary restriction prior to testing.¹³

The use of FOBTs (particularly FITs) is also subject to false-negative results, or type II errors. While FITs have increased specificity for lower GI bleeding, their ability to detect UGIB is limited, because most Hb is digested in the small intestine and not present in rectal stool.¹⁴ In a study of more than 2,700 patients, FIT results were not correlated with the presence of upper GI pathology.¹⁵ False-negative results are less common with gFOBTs, although these may occur with low volume, slow or intermittent bleeding,¹⁶ or with ingestion of substances that inhibit oxidation, such as vitamin C.¹⁷

Beyond these test limitations, studies suggest that the majority of inpatient FOBT results do not impact immediate medical decision-making or management. In one study, only 34% of hospitalized patients with a positive FOBT underwent further GI studies, with the majority of those patients (60%) receiving endoscopy before the results of the FOBT were known.¹⁸ In another study of 201 FOBTs performed on hospitalized patients, those with negative results underwent further GI evaluation at a higher rate than those with positive results (41% vs 38%).⁸ This aligns with a study that revealed the majority of patients suspected of having a GI bleed underwent endoscopic evaluation regardless of the FOBT result.⁹

FOBT currently has a role in CRC screening and may have a role in the evaluation of anemia of unknown etiology to evaluate for occult GIB, although the yield is likely low.¹³ In one retrospective analysis of inpatients with unexplained anemia, 43.6% of FOBTs were positive, but a potential GI cause was found in only 6.8% of patients.⁹ Patients with anemia from an unknown etiology should have a workup based on the history, physical, and complete blood count indices. While iron deficiency anemia warrants eventual evaluation for occult blood loss, noncritical anemia in an otherwise stable patient does not require an inpatient evaluation. When FOBT is used in the outpatient setting, patients can be counseled on proper dietary and medication modifications prior to testing.

A careful history, physical examination, and visual inspection of the stool remain the foundation of establishing UGIB as the etiology of anemia. Observed melena (either by passed stool or a rectal examination) has a likelihood ratio (LR) of 25 for UGIB; a patient’s self-report of stools that sounds melenic (black or tarry) has an LR of 5-6.¹⁹ An upper GI source may be further supported by an elevated blood urea nitrogen (BUN) to creatinine ratio, as blood is absorbed through the small bowel and patients may have concomitant decreased renal perfusion. A BUN to creatinine ratio of >30 is associated with a positive LR (LR+) of 7.5 for UGIB.¹⁹ Recall that the higher the LR+, and the lower the negative LR (LR-), the better the test is at ruling in and out the diagnosis, respectively. LR+ of 2–10 and LR– of 0.1–0.5 represent a modestly helpful diagnostic test, whereas LR+ >10 and LR- <0.1 are considered robust. These are generalizations only, as value of LR+/LR- depends on pretest probability.

Clinical decision tools, such as the Glasgow-Blatchford and Rockall scores, utilize the history, physical examination, laboratory results, and pretest probability for high-grade peptic ulcer stigmata to estimate the severity of an UGIB and risk for adverse outcomes, respectively. Notably, these scoring systems do not include FOBT results. Despite the relatively inexpensive cost per FOBT ($3.03 per unit),²⁰ this test’s poor specificity when used in the inpatient setting has the potential to lead to significant, unnecessary downstream expense (as well as the potential for procedural risk and anxiety for patients). Given that the incidence of acute UGIB is approximately 100 per 100,000 persons per year,⁷ based on the United States population in 2016,²¹ there were 323,936 patients with UGIB. If each patient underwent an FOBT, the direct expense would be nearly a million dollars. Nonetheless, the number of patients getting a FOBT in the inpatient setting for a suspected UGIB (or for other indications) is unknown, and the direct costs of the tests itself likely represent a fraction of the healthcare expenditures associated with this practice. Allowing that only a third of patients with positive FOBTs in the inpatient setting typically undergo EGD,²² overuse of this test would lead to a high number of unnecessary EGDs, and potentially colonoscopies or additional diagnostic procedures (eg, capsule endoscopy). In light of the false-positive results associated with FOBT, and lack of diagnostic utility, this brief cost analysis suggests FOBT is a low-value test for suspected UGIB in the inpatient setting, and there are potential significant cost savings if FOBTs are withheld.

Although Gastroccult²³ may be considered for the detection of occult blood in gastric juice, its package insert states: “As with any occult blood test, results with the Gastroccult test cannot be considered conclusive evidence of the presence or absence of upper gastrointestinal bleeding or pathology.” As with any diagnostic evaluation, we would only recommend this test if it would change management.

• FOBT should not be performed to diagnose UGIB.
• When there is clinical suspicion of acute GI bleeding, the best diagnostic tools are a good history, physical examination, and visual inspection of the stool by the clinician to determine the presence of hematochezia or melena.
• Deferring FOBT to the ambulatory setting may improve test performance characteristics.

Revisiting our patient, for all of the reasons discussed above, there is no indication for FOBT as it would not affect management. Based on a careful history and physical examination, our patient would likely require upper endoscopy either as an inpatient or an outpatient depending on his clinical course.

FOBT is validated as an outpatient colon cancer screening tool in asymptomatic patients, not for inpatient evaluation of acute GIB. Given the poor positive predictive value for a positive FOBT in an acute GIB scenario, the potential risk for unnecessary treatments or procedures is real. Conversely, a negative FOBT (particularly FIT) does not rule out GI bleeding and risks a false sense of security that may result in under-treatment. In most scenarios in which FOBT is performed, clinicians can make decisions based on a composite of history, physical exam, visual inspection of the stool, and laboratory investigation. Until further research substantiates the utility of FOBT for this purpose, we would recommend against the routine use of FOBT for evaluating UGIB in hospitalized patients.

Acknowledgment

The authors would like to thank and acknowledge the team from Orlando Health for their review of this manuscript.

Disclosure: The authors do not have any relevant financial disclosures to report. The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs.

Do you think this is a low-value practice? Is this truly a “Thing We Do for No Reason”? Let us know what you do in your practice and propose ideas for other “Things We Do for No Reason” topics. Please join in the conversation online at Twitter (#TWDFNR)/Facebook and don’t forget to “Like It” on Facebook or retweet it on Twitter. We invite you to propose ideas for other “Things We Do for No Reason” topics by [email protected].

We previously identified key domains of attending attributes for successful rounds¹ and adapted them to represent the trainees’ perspective: Teaching Process (eg, sharing decision-making process, physical exam skills), Learning Environment (eg, being approachable, respectful), Role Modeling (eg, teaching by example, bedside manner), and Team Management (eg, efficiency, providing autonomy). Though all domains are necessary, the relative importance may change in response to external pressures. Inpatient service demands and time constraints fluctuate daily due to patient admissions and discharges, educational conference schedules, and concurrent outpatient clinic responsibilities.^2–4 Furthermore, the 2011 Accreditation Council for Graduate Medical Education (ACGME) duty hour rules placed greater time pressure on inpatient ward attending rounds.⁵ It is plausible that these pressures affect trainees’ needs and priorities during rounds.

Therefore, we sought to refine our understanding of the learners’ needs during ward rounds. Because we were interested in the contextual influences of trainee characteristics and workload on their preferences, we used the principles of ecological momentary assessment (EMA) to design a novel system to assess daily changes in trainee priorities and associated workload.

Design, Participants, and Setting

In a prospective observational study, we assessed trainee priorities during inpatient rounds. Participants included third- and fourth-year medical students in the University of Alabama at Birmingham (UAB) School of Medicine (Birmingham campus) and residents in the Tinsley Harrison Internal Medicine Residency Training Program assigned to inpatient general medicine ward services from September 2010 to February 2011 (except from December 20, 2010, to January 3, 2011). Three training sites were included in this study: UAB Hospital (>1000-bed, university-based hospital), Birmingham Veterans Affairs Medical Center (300-bed), and Cooper Green Hospital (40-bed county hospital). Each site housed 4 or 5 general medicine ward teams, composed of 1–3 medical students (third or fourth year), 2 first-year residents, 1 upper level resident (postgraduate year 2–4), and 1 attending physician.

Assessment

Our original domains of attending rounds were derived from groupings and ratings of specific teaching characteristics by attending physicians, residents, and students.¹ However, because our goal was to understand the learners’ perspective of successful ward rounds, we revised these domains by limiting the algorithm to data on groupings and ratings by residents and students only. This process resulted in the domains used in this study (Appendix).

We used EMA principles to create a novel system to assess daily variation in trainees’ prioritization of these domains and workload.^6,7 EMA-derived assessment tools collect data frequently (several times per week, up to multiple times per day) to identify time-sensitive fluctuations. We designed a pocket-sized daily assessment card for trainees to complete each day after rounds (Figure 1). Trainees were asked to indicate the most important domain for successful ward rounds to them that day and provide individual characteristics (ie, sex and training level) and data on factors we hypothesized were related to perceived work load (ie, patient census, day of call cycle, and number of team members absent on rounds that day). We anticipated the Expectations domain would not be responsive to daily changes in workload because expectations are usually set once on the first day of the rotation, and thus we did not include this domain in our final assessment tool. Assessment cards and locked receptacles were kept in team workrooms for ease of accessibility; cards were collected twice monthly. All data were anonymous.

Analysis

Our unit of analysis was the EMA card. We examined associations between daily domain priority with respondent demographics and workload information using Pearson’s chi-squared analyses, adjusted for clustering effects by team. α was set at 0.05. All analyses were performed by using Stata 13.0 (Stata Statistical Software: Release 13; College Station, TX).

Sex and training level were associated with prioritization of teaching domains. Male trainees were more likely to choose Team Management (P = 0.01) or Teaching Process (P = 0.04) as their preferred domain. Medical students valued Teaching Process 42% of the time, compared with 23% for interns and 21% for upper level residents (P = 0.005). The opposite trend emerged for Team Management: as training level increased, the importance of Team Management increased (P < 0.001). There were no significant trends by training level for the Role Modeling and Learning Environment domains.

Domain priority was also associated with workload characteristics. On post-call days, Team Management (P < 0.001) was more likely to be selected as the most important domain, but on other days, Teaching Process (P = 0.005) was more often selected (Figure 2). Trainees also selected Team Management as the most important domain with an increasing number of team members absent (P = 0.001), and as the teams’ overall patient census increased (P < 0.001). The Learning Environment and Role Modeling domains’ importance did not vary by call-day or patient census.

We used a novel approach to assess contextual factors affecting trainee prioritization of 4 domains that contribute to successful inpatient internal medicine attending rounds. We found training level and workload demands were associated with prioritization of teaching domains. Prioritization of Teaching Process, exemplified by setting aside time to teach, demonstrating physical exam skills, and clear delineation of the attending’s thought process, was inversely associated with training level. On the days with highest workload, Team Management was most likely to be prioritized. Our findings suggest that attending physicians should consider adapting rounding style based on team members’ training levels and workload.

Prior work has described teaching and rounding styles, influences, and priorities in response to workload from the attending physician’s perspective,^8–11 and our study extends these reports by providing the complementary perspective of trainees. On days with high workload, trainees prefer the Team Management domain, characterized by organized and efficient rounds, agreement on a clear and consistent plan of care, and being allowed independence and time during rounds to meet other responsibilities.¹ These findings support an “empowerment style,” defined by Goldszmidt et al. as using integrated teaching and oversight strategies to support trainees’ progressive independence.⁹ Though some attending physicians report shifting to a more direct patient care style on days with a high patient census,⁹ our results suggest that learners instead prefer more independence, being empowered to perform more direct care. While there is an increasing pressure to heighten attending supervision due to concerns about patient safety, restricted work hours, and litigation,¹² trainees value being part of the care process and being included as integral members of the care team, which may actually mitigate patient safety risks.⁸

Our results are consistent with prior studies, reporting that learners at different levels of training have different instructional needs: medical students seek more teaching, and senior residents sought an efficient leader.^13,14 Taken together, these studies suggest that attending physicians should tailor rounds to the level of the trainee. For example, it may be beneficial for the attending physician to spend time outside of rounds with students to teach medical knowledge. During rounds, the entire group benefits most from modeling clinical reasoning, discussing new medical evidence, and demonstrating communication skills and leadership.

Our study has limitations. Though our study was performed before the 2011 ACGME duty hour restrictions,⁵ our results are likely of greater importance and relevance, as our findings ultimately highlight the competing demands of time vs duty. Also, while our study was performed at a single institution, potentially limiting generalizability, we included 3 types of training hospitals, a university, veterans and a county hospital, and found no differences between sites. Additionally, we collected over 1,000 cards over the course of 6 months, assessing rounds of over 50 different attending physicians, suggesting broader applicability. Our overall response rate was low, a typical signal for respondent bias, but because we collected daily assessments, standard interpretation of response rates referring to a one-time survey do not apply.¹⁵ We believe we achieved an adequate sample, as the majority of teams participated, the respondent demographics were proportional to the base population eligible to participate, and we received a similar number of cards on all months. Finally, although we were unable to account for clustering effects by individual respondents because response cards were anonymous, we adjusted for clustering effects by team.

Attending physicians may use our findings to adapt teaching techniques to appeal to specific training levels and to external pressures during teaching rounds. Focusing and investing time in teaching medical knowledge and clinical reasoning tailored to each level of learner is paramount on most days. However, days with a high workload may require emphasis on delegating clear, rational treatment plans, when learners are less receptive to traditional didactic methods.

Disclosure

An abstract based on the current analysis was presented at the Society of General Internal Medicine 34th Annual Meeting, April 2011, Phoenix, AZ. Dr. Brita Roy is supported by grant number K12HS023000 from the Agency for Healthcare Research and Quality. The authors have no conflicts to disclose. The opinions expressed in this article are those of the authors alone and do not reflect the views of the Department of Veterans Affairs or the Agency for Healthcare Research and Quality.

We previously identified key domains of attending attributes for successful rounds¹ and adapted them to represent the trainees’ perspective: Teaching Process (eg, sharing decision-making process, physical exam skills), Learning Environment (eg, being approachable, respectful), Role Modeling (eg, teaching by example, bedside manner), and Team Management (eg, efficiency, providing autonomy). Though all domains are necessary, the relative importance may change in response to external pressures. Inpatient service demands and time constraints fluctuate daily due to patient admissions and discharges, educational conference schedules, and concurrent outpatient clinic responsibilities.^2–4 Furthermore, the 2011 Accreditation Council for Graduate Medical Education (ACGME) duty hour rules placed greater time pressure on inpatient ward attending rounds.⁵ It is plausible that these pressures affect trainees’ needs and priorities during rounds.

Therefore, we sought to refine our understanding of the learners’ needs during ward rounds. Because we were interested in the contextual influences of trainee characteristics and workload on their preferences, we used the principles of ecological momentary assessment (EMA) to design a novel system to assess daily changes in trainee priorities and associated workload.

Design, Participants, and Setting

In a prospective observational study, we assessed trainee priorities during inpatient rounds. Participants included third- and fourth-year medical students in the University of Alabama at Birmingham (UAB) School of Medicine (Birmingham campus) and residents in the Tinsley Harrison Internal Medicine Residency Training Program assigned to inpatient general medicine ward services from September 2010 to February 2011 (except from December 20, 2010, to January 3, 2011). Three training sites were included in this study: UAB Hospital (>1000-bed, university-based hospital), Birmingham Veterans Affairs Medical Center (300-bed), and Cooper Green Hospital (40-bed county hospital). Each site housed 4 or 5 general medicine ward teams, composed of 1–3 medical students (third or fourth year), 2 first-year residents, 1 upper level resident (postgraduate year 2–4), and 1 attending physician.

Assessment

Our original domains of attending rounds were derived from groupings and ratings of specific teaching characteristics by attending physicians, residents, and students.¹ However, because our goal was to understand the learners’ perspective of successful ward rounds, we revised these domains by limiting the algorithm to data on groupings and ratings by residents and students only. This process resulted in the domains used in this study (Appendix).

We used EMA principles to create a novel system to assess daily variation in trainees’ prioritization of these domains and workload.^6,7 EMA-derived assessment tools collect data frequently (several times per week, up to multiple times per day) to identify time-sensitive fluctuations. We designed a pocket-sized daily assessment card for trainees to complete each day after rounds (Figure 1). Trainees were asked to indicate the most important domain for successful ward rounds to them that day and provide individual characteristics (ie, sex and training level) and data on factors we hypothesized were related to perceived work load (ie, patient census, day of call cycle, and number of team members absent on rounds that day). We anticipated the Expectations domain would not be responsive to daily changes in workload because expectations are usually set once on the first day of the rotation, and thus we did not include this domain in our final assessment tool. Assessment cards and locked receptacles were kept in team workrooms for ease of accessibility; cards were collected twice monthly. All data were anonymous.

Analysis

Our unit of analysis was the EMA card. We examined associations between daily domain priority with respondent demographics and workload information using Pearson’s chi-squared analyses, adjusted for clustering effects by team. α was set at 0.05. All analyses were performed by using Stata 13.0 (Stata Statistical Software: Release 13; College Station, TX).

Sex and training level were associated with prioritization of teaching domains. Male trainees were more likely to choose Team Management (P = 0.01) or Teaching Process (P = 0.04) as their preferred domain. Medical students valued Teaching Process 42% of the time, compared with 23% for interns and 21% for upper level residents (P = 0.005). The opposite trend emerged for Team Management: as training level increased, the importance of Team Management increased (P < 0.001). There were no significant trends by training level for the Role Modeling and Learning Environment domains.

Domain priority was also associated with workload characteristics. On post-call days, Team Management (P < 0.001) was more likely to be selected as the most important domain, but on other days, Teaching Process (P = 0.005) was more often selected (Figure 2). Trainees also selected Team Management as the most important domain with an increasing number of team members absent (P = 0.001), and as the teams’ overall patient census increased (P < 0.001). The Learning Environment and Role Modeling domains’ importance did not vary by call-day or patient census.

We used a novel approach to assess contextual factors affecting trainee prioritization of 4 domains that contribute to successful inpatient internal medicine attending rounds. We found training level and workload demands were associated with prioritization of teaching domains. Prioritization of Teaching Process, exemplified by setting aside time to teach, demonstrating physical exam skills, and clear delineation of the attending’s thought process, was inversely associated with training level. On the days with highest workload, Team Management was most likely to be prioritized. Our findings suggest that attending physicians should consider adapting rounding style based on team members’ training levels and workload.

Prior work has described teaching and rounding styles, influences, and priorities in response to workload from the attending physician’s perspective,^8–11 and our study extends these reports by providing the complementary perspective of trainees. On days with high workload, trainees prefer the Team Management domain, characterized by organized and efficient rounds, agreement on a clear and consistent plan of care, and being allowed independence and time during rounds to meet other responsibilities.¹ These findings support an “empowerment style,” defined by Goldszmidt et al. as using integrated teaching and oversight strategies to support trainees’ progressive independence.⁹ Though some attending physicians report shifting to a more direct patient care style on days with a high patient census,⁹ our results suggest that learners instead prefer more independence, being empowered to perform more direct care. While there is an increasing pressure to heighten attending supervision due to concerns about patient safety, restricted work hours, and litigation,¹² trainees value being part of the care process and being included as integral members of the care team, which may actually mitigate patient safety risks.⁸

Our results are consistent with prior studies, reporting that learners at different levels of training have different instructional needs: medical students seek more teaching, and senior residents sought an efficient leader.^13,14 Taken together, these studies suggest that attending physicians should tailor rounds to the level of the trainee. For example, it may be beneficial for the attending physician to spend time outside of rounds with students to teach medical knowledge. During rounds, the entire group benefits most from modeling clinical reasoning, discussing new medical evidence, and demonstrating communication skills and leadership.

Our study has limitations. Though our study was performed before the 2011 ACGME duty hour restrictions,⁵ our results are likely of greater importance and relevance, as our findings ultimately highlight the competing demands of time vs duty. Also, while our study was performed at a single institution, potentially limiting generalizability, we included 3 types of training hospitals, a university, veterans and a county hospital, and found no differences between sites. Additionally, we collected over 1,000 cards over the course of 6 months, assessing rounds of over 50 different attending physicians, suggesting broader applicability. Our overall response rate was low, a typical signal for respondent bias, but because we collected daily assessments, standard interpretation of response rates referring to a one-time survey do not apply.¹⁵ We believe we achieved an adequate sample, as the majority of teams participated, the respondent demographics were proportional to the base population eligible to participate, and we received a similar number of cards on all months. Finally, although we were unable to account for clustering effects by individual respondents because response cards were anonymous, we adjusted for clustering effects by team.

Attending physicians may use our findings to adapt teaching techniques to appeal to specific training levels and to external pressures during teaching rounds. Focusing and investing time in teaching medical knowledge and clinical reasoning tailored to each level of learner is paramount on most days. However, days with a high workload may require emphasis on delegating clear, rational treatment plans, when learners are less receptive to traditional didactic methods.

Disclosure

An abstract based on the current analysis was presented at the Society of General Internal Medicine 34th Annual Meeting, April 2011, Phoenix, AZ. Dr. Brita Roy is supported by grant number K12HS023000 from the Agency for Healthcare Research and Quality. The authors have no conflicts to disclose. The opinions expressed in this article are those of the authors alone and do not reflect the views of the Department of Veterans Affairs or the Agency for Healthcare Research and Quality.

We previously identified key domains of attending attributes for successful rounds¹ and adapted them to represent the trainees’ perspective: Teaching Process (eg, sharing decision-making process, physical exam skills), Learning Environment (eg, being approachable, respectful), Role Modeling (eg, teaching by example, bedside manner), and Team Management (eg, efficiency, providing autonomy). Though all domains are necessary, the relative importance may change in response to external pressures. Inpatient service demands and time constraints fluctuate daily due to patient admissions and discharges, educational conference schedules, and concurrent outpatient clinic responsibilities.^2–4 Furthermore, the 2011 Accreditation Council for Graduate Medical Education (ACGME) duty hour rules placed greater time pressure on inpatient ward attending rounds.⁵ It is plausible that these pressures affect trainees’ needs and priorities during rounds.

Therefore, we sought to refine our understanding of the learners’ needs during ward rounds. Because we were interested in the contextual influences of trainee characteristics and workload on their preferences, we used the principles of ecological momentary assessment (EMA) to design a novel system to assess daily changes in trainee priorities and associated workload.

Design, Participants, and Setting

In a prospective observational study, we assessed trainee priorities during inpatient rounds. Participants included third- and fourth-year medical students in the University of Alabama at Birmingham (UAB) School of Medicine (Birmingham campus) and residents in the Tinsley Harrison Internal Medicine Residency Training Program assigned to inpatient general medicine ward services from September 2010 to February 2011 (except from December 20, 2010, to January 3, 2011). Three training sites were included in this study: UAB Hospital (>1000-bed, university-based hospital), Birmingham Veterans Affairs Medical Center (300-bed), and Cooper Green Hospital (40-bed county hospital). Each site housed 4 or 5 general medicine ward teams, composed of 1–3 medical students (third or fourth year), 2 first-year residents, 1 upper level resident (postgraduate year 2–4), and 1 attending physician.

Assessment

Our original domains of attending rounds were derived from groupings and ratings of specific teaching characteristics by attending physicians, residents, and students.¹ However, because our goal was to understand the learners’ perspective of successful ward rounds, we revised these domains by limiting the algorithm to data on groupings and ratings by residents and students only. This process resulted in the domains used in this study (Appendix).

We used EMA principles to create a novel system to assess daily variation in trainees’ prioritization of these domains and workload.^6,7 EMA-derived assessment tools collect data frequently (several times per week, up to multiple times per day) to identify time-sensitive fluctuations. We designed a pocket-sized daily assessment card for trainees to complete each day after rounds (Figure 1). Trainees were asked to indicate the most important domain for successful ward rounds to them that day and provide individual characteristics (ie, sex and training level) and data on factors we hypothesized were related to perceived work load (ie, patient census, day of call cycle, and number of team members absent on rounds that day). We anticipated the Expectations domain would not be responsive to daily changes in workload because expectations are usually set once on the first day of the rotation, and thus we did not include this domain in our final assessment tool. Assessment cards and locked receptacles were kept in team workrooms for ease of accessibility; cards were collected twice monthly. All data were anonymous.

Analysis

Our unit of analysis was the EMA card. We examined associations between daily domain priority with respondent demographics and workload information using Pearson’s chi-squared analyses, adjusted for clustering effects by team. α was set at 0.05. All analyses were performed by using Stata 13.0 (Stata Statistical Software: Release 13; College Station, TX).

Sex and training level were associated with prioritization of teaching domains. Male trainees were more likely to choose Team Management (P = 0.01) or Teaching Process (P = 0.04) as their preferred domain. Medical students valued Teaching Process 42% of the time, compared with 23% for interns and 21% for upper level residents (P = 0.005). The opposite trend emerged for Team Management: as training level increased, the importance of Team Management increased (P < 0.001). There were no significant trends by training level for the Role Modeling and Learning Environment domains.

Domain priority was also associated with workload characteristics. On post-call days, Team Management (P < 0.001) was more likely to be selected as the most important domain, but on other days, Teaching Process (P = 0.005) was more often selected (Figure 2). Trainees also selected Team Management as the most important domain with an increasing number of team members absent (P = 0.001), and as the teams’ overall patient census increased (P < 0.001). The Learning Environment and Role Modeling domains’ importance did not vary by call-day or patient census.

We used a novel approach to assess contextual factors affecting trainee prioritization of 4 domains that contribute to successful inpatient internal medicine attending rounds. We found training level and workload demands were associated with prioritization of teaching domains. Prioritization of Teaching Process, exemplified by setting aside time to teach, demonstrating physical exam skills, and clear delineation of the attending’s thought process, was inversely associated with training level. On the days with highest workload, Team Management was most likely to be prioritized. Our findings suggest that attending physicians should consider adapting rounding style based on team members’ training levels and workload.

Prior work has described teaching and rounding styles, influences, and priorities in response to workload from the attending physician’s perspective,^8–11 and our study extends these reports by providing the complementary perspective of trainees. On days with high workload, trainees prefer the Team Management domain, characterized by organized and efficient rounds, agreement on a clear and consistent plan of care, and being allowed independence and time during rounds to meet other responsibilities.¹ These findings support an “empowerment style,” defined by Goldszmidt et al. as using integrated teaching and oversight strategies to support trainees’ progressive independence.⁹ Though some attending physicians report shifting to a more direct patient care style on days with a high patient census,⁹ our results suggest that learners instead prefer more independence, being empowered to perform more direct care. While there is an increasing pressure to heighten attending supervision due to concerns about patient safety, restricted work hours, and litigation,¹² trainees value being part of the care process and being included as integral members of the care team, which may actually mitigate patient safety risks.⁸

Our results are consistent with prior studies, reporting that learners at different levels of training have different instructional needs: medical students seek more teaching, and senior residents sought an efficient leader.^13,14 Taken together, these studies suggest that attending physicians should tailor rounds to the level of the trainee. For example, it may be beneficial for the attending physician to spend time outside of rounds with students to teach medical knowledge. During rounds, the entire group benefits most from modeling clinical reasoning, discussing new medical evidence, and demonstrating communication skills and leadership.

Our study has limitations. Though our study was performed before the 2011 ACGME duty hour restrictions,⁵ our results are likely of greater importance and relevance, as our findings ultimately highlight the competing demands of time vs duty. Also, while our study was performed at a single institution, potentially limiting generalizability, we included 3 types of training hospitals, a university, veterans and a county hospital, and found no differences between sites. Additionally, we collected over 1,000 cards over the course of 6 months, assessing rounds of over 50 different attending physicians, suggesting broader applicability. Our overall response rate was low, a typical signal for respondent bias, but because we collected daily assessments, standard interpretation of response rates referring to a one-time survey do not apply.¹⁵ We believe we achieved an adequate sample, as the majority of teams participated, the respondent demographics were proportional to the base population eligible to participate, and we received a similar number of cards on all months. Finally, although we were unable to account for clustering effects by individual respondents because response cards were anonymous, we adjusted for clustering effects by team.

Attending physicians may use our findings to adapt teaching techniques to appeal to specific training levels and to external pressures during teaching rounds. Focusing and investing time in teaching medical knowledge and clinical reasoning tailored to each level of learner is paramount on most days. However, days with a high workload may require emphasis on delegating clear, rational treatment plans, when learners are less receptive to traditional didactic methods.

Disclosure

An abstract based on the current analysis was presented at the Society of General Internal Medicine 34th Annual Meeting, April 2011, Phoenix, AZ. Dr. Brita Roy is supported by grant number K12HS023000 from the Agency for Healthcare Research and Quality. The authors have no conflicts to disclose. The opinions expressed in this article are those of the authors alone and do not reflect the views of the Department of Veterans Affairs or the Agency for Healthcare Research and Quality.

Clinician educators face numerous obstacles to their joint mission of facilitating learning while also ensuring high-quality and patient-centered care. Time constraints, including the institution of house officer duty hour limitations,¹ shorter lengths of stay for hospitalized patients,² and competing career responsibilities, combine to create a dynamic learning environment. Additionally, clinician educators must balance the autonomy of their learners with the safety of their patients. They must teach to multiple learning levels and work collaboratively with multiple disciplines to foster an effective team-based approach to patient care. Yet, many clinician educators have no formal training in pedagogical methods.³ Such challenges necessitate increased attention to the work of excellent clinician educators and their respective teaching approaches.

Many studies of clinical teaching rely primarily on survey data of attributes of good clinical teachers.^3-7 While some studies have incorporated direct observations of teaching^8,9 or interviews with clinician educators or learners,^10,11 few have incorporated multiple perspectives from the current team and from former learners in order to provide a comprehensive picture of team-based learning.¹²

The goal of this study was to gain a thorough understanding, through multiple perspectives, of the techniques and behaviors used by exemplary educators within actual clinical environments. We studied attitudes, behaviors, and approaches of 12 such inpatient clinician educators.

Study Design and Sampling

This was a multisite study using an exploratory qualitative approach to inquiry. This approach was used to study the techniques and behaviors of excellent attendings during inpatient general medicine rounds. A modified snowball sampling approach¹³ was used, meaning individuals known to one member of the research team (SS) were initially contacted and asked to identify clinician educators (also referred to as attendings) for potential inclusion in the study. In an effort to identify attendings from a broad range of medical schools, the “2015 U.S. News and World Report Top Medical Schools: Research” rankings¹⁴ were also reviewed, with priority given to the top 25, as these are widely used to represent the best US hospitals. In an attempt to invite attendings from diverse institutions, additional medical schools not in the top 25 as well as historically black medical schools were also included. Division chiefs and chairs of internal medicine and/or directors of internal medicine residency programs at these schools were contacted and asked for recommendations of attendings, both within and outside their institutions, who they considered to be great inpatient teachers. In addition, key experts who have won teaching awards or were known to be specialists in the field of medical education were asked to nominate one or two other outstanding attendings.

By using this sampling method, 59 potential participants were identified. An internet search was conducted to obtain information about the potential participants and their institutions. Organizational characteristics such as geographic location, hospital size and affiliation, and patient population, as well as individual characteristics such as gender, medical education and training, and educational awards received were considered so that a diversity of organizations and backgrounds was represented. The list was narrowed down to 16 attendings who were contacted via e-mail and asked to participate. Interested participants were asked for a list of their current team members and 6 to 10 former learners to contact for interviews and focus groups. Former learners were included in an effort to better understand lasting effects on learners from their exemplary teaching attendings. A total of 12 attending physicians agreed to participate (Table 1). Literature on field methods has shown that 12 interviews are found to be adequate in accomplishing data saturation.¹⁵ Although 2 attendings were located at the same institution, we decided to include them given that both are recognized as master clinician educators and were each recommended by several individuals from various institutions. Hospitals were located throughout the US and included both university-affiliated hospitals and Veterans Affairs medical centers. Despite efforts to include physicians from historically black colleges and universities, only one attending was identified, and they declined the request to participate.

Data Collection

Observations. The one-day site visits were mainly conducted by two research team members, a physician (SS) and a medical anthropologist (MH), both of whom have extensive experience in qualitative methods. Teams were not uniform but were generally comprised of 1 attending, 1 senior medical resident, 1 to 2 interns, and approximately 2 medical students. Occasionally, a pharmacist, clinical assistant, or other health professional accompanied the team on rounds. Not infrequently, the bedside nurse would explicitly be included in the discussion regarding his or her specific patient. Each site visit began with observing attendings (N = 12) and current learners (N = 57) during rounds. Each research team member recorded their own observations via handwritten field notes, paying particular attention to group interactions, teaching approach, conversations occurring within and peripheral to the team, patient-team interactions, and the physical environment. By standing outside of the medical team circle and remaining silent during rounds, research team members remained unobtrusive to the discussion and process of rounds. Materials the attendings used during their teaching rounds were also documented and collected. Rounds generally lasted 2 to 3 hours. After each site visit, the research team met to compare and combine field notes.

Interviews and Focus Groups. The research team then conducted individual, semi-structured interviews with the attendings, focus groups with their current team (N = 46), and interviews or focus groups with their former learners (N = 26; Supplement 1). Eleven of the current team members observed during rounds were unable to participate in the focus groups due to clinical duties. Because the current learners who participated in the focus groups were also observed during rounds, the research team was able to ask them open-ended questions regarding teaching rounds and their roles as learners within this environment. Former learners who were still at the hospital participated in separate focus groups or interviews. Former learners who were no longer present at the hospital were contacted by telephone and individually interviewed by one research team member (MH). All interviews and focus groups were audio-recorded and transcribed.

This study was determined to be exempt by the University of Michigan Institutional Review Board. All participants were informed that their participation was completely voluntary and that they could terminate their involvement at any time.

Data Analysis

Data were analyzed using a thematic analysis approach.¹⁶ Thematic analysis entails reading through the data to identify patterns (and create codes) that relate to behaviors, experiences, meanings, and activities. Once patterns have been identified, they are grouped according to similarity into themes, which help to further explain the findings.¹⁷

After the first site visit was completed, the research team members that participated (SS and MH) met to develop initial ideas about meanings and possible patterns. All transcripts were read by one team member (MH) and, based on review of the data, codes were developed, defined, and documented in a codebook. This process was repeated after every site visit using the codebook to expand or combine codes and refine definitions as necessary. If a new code was added, the previously coded data were reviewed to apply the new code. NVivo® 10 software (QSR International; Melbourne, Australia) was used to manage the data.

Once all field notes and transcripts were coded (MH), the code reports, which list all data described within a specific code, were run to ensure consistency and identify relationships between codes. Once coding was verified, codes were grouped based on similarities and relationships into salient themes by 3 members of the research team (NH, MH, and SM). Themes, along with their supporting codes, were then further defined to understand how these attendings worked to facilitate excellent teaching in clinical settings.

The coded interview data and field notes were categorized into broad, overlapping themes. Three of these major themes include (1) fostering positive relationships, (2) patient-centered teaching, and (3) collaboration and coaching. Table 2 lists each theme, salient behaviors, examples, and selected quotes that further elucidate its meaning.

Fostering Positive Relationships

Attending physicians took observable steps to develop positive relationships with their team members, which in turn created a safe learning environment. For instance, attendings used learners’ first names, demonstrated interest in their well-being, deployed humor, and generally displayed informal actions—uncrossed arms, “fist bump” when recognizing learners’ success, standing outside the circle of team members and leaning in to listen—during learner interactions. Attendings also made it a priority to get to know individuals on a personal level. As one current learner put it, “He asks about where we are from. He will try to find some kind of connection that he can establish with not only each of the team members but also with each of the patients.”

Additionally, attendings built positive relationships with their learners by responding thoughtfully to their input, even when learners’ evaluations of patients required modification. In turn, learners reported feeling safe to ask questions, admit uncertainty, and respectfully disagree with their attendings. As one attending reflected, “If I can get them into a place where they feel like the learning environment is someplace where they can make a mistake and know that that mistake does not necessarily mean that it’s going to cost them in their evaluation part, then I feel like that’s why it’s important.”

To build rapport and create a safe learning environment, attendings used a number of strategies to position themselves as learners alongside their team members. For instance, attendings indicated that they wanted their ideas questioned because they saw it as an opportunity to learn. Moreover, in conversations with learners, attendings demonstrated humility, admitting when they did not know something. One former learner noted, “There have been times when he has asked [a] question…nobody knows and then he admits that he doesn’t know either. So everybody goes and looks it up…The whole thing turns out to be a fun learning experience.”

Attendings demonstrated respect for their team members’ time by reading about patients before rounds, identifying learning opportunities during rounds, and integrating teaching points into the daily work of patient care. Teaching was not relegated exclusively to the conference room or confined to the traditional “chalk talk” before or after rounds but rather was assimilated into daily workflow. They appeared to be responsive to the needs of individual patients and the team, which allowed attendings to both directly oversee their patients’ care and overcome the challenges of multiple competing demands for time. The importance of this approach was made clear by one current learner who stated “…she does prepare before, especially you know on call days, she does prepare for the new patients before coming in to staff, which is really appreciated… it saves a lot of time on rounds.”

Attendings also included other health professionals in team discussions. Attendings used many of the same relationship-building techniques with these professionals as they did with learners and patients. They consistently asked these professionals to provide insight and direction in patients’ plans of care. A former learner commented, “He always asks the [nurse] what is her impression of the patient...he truly values the [nurse’s] opinion of the patient.” One attending reiterated this approach, stating “I don’t want them to think that anything I have to say is more valuable than our pharmacist or the [nurse].”

Patient-Centered Teaching

Attending physicians modeled numerous teaching techniques that focused learning around the patient. Attendings knew their patients well through review of the medical records, discussion with the patient, and personal examination. This preparation allowed attendings to focus on key teaching points in the context of the patient. One former learner noted, “He tended to bring up a variety of things that really fit well into the clinical scenario. So whether that is talking about what is the differential for a new symptom that just came up for this patient or kind of here is a new paper talking about this condition or maybe some other pearl of physical exam for a patient that has a certain physical condition.”

Attendings served as effective role models by being directly involved in examining and talking with patients as well as demonstrating excellent physical examination and communication techniques. One current learner articulated the importance of learning these skills by observing them done well: “I think he teaches by example and by doing, again, those little things: being attentive to the patients and being very careful during exams…I think those are things that you teach people by doing them, not by saying you need to do this better during the patient encounter.”

Collaboration and Coaching

Attending physicians used varied collaboration and coaching techniques to facilitate learning across the entire care team. During rounds, attendings utilized visual aids to reinforce key concepts and simplify complex topics. They also collaborated by using discussion rather than lecture to engage with team members. For instance, attendings used Socratic questioning, asking questions that lead learners through critical thinking and allow them to solve problems themselves, to guide learners’ decision-making. One former learner reported, “He never gives you the answer, and he always asks your opinion; ‘So what are your thoughts on this?’”

Coaching for success, rather than directing the various team members, was emphasized. Attendings did not wish to be seen as the “leaders” of the team. During rounds, one attending was noted to explain his role in ensuring that the team was building connections with others: “When we have a bad outcome, if it feels like your soul has been ripped out, then you’ve done something right. You’ve made that connection with the patient. My job, as your coach, was to build communication between all of us so we feel vested in each other and our patients.”

Attendings also fostered clinical reasoning skills in their learners by encouraging them to verbalize their thought processes aloud in order to clarify and check for understanding. Attendings also placed emphasis not simply on memorizing content but rather prioritization of the patient’s problems and thinking step by step through individual medical problems. One current learner applauded an attending who could “come up with schematics of how to approach problems rather than feeding us factual information of this paper or this trial.”

Additionally, attendings facilitated learning across the entire care team by differentiating their teaching to meet the needs of multiple learning levels. While the entire team was explicitly included in the learning process, attendings encouraged learners to play various roles, execute tasks, and answer questions depending on their educational level. Attendings positioned learners as leaders of the team by allowing them to talk without interruption and by encouraging them to take ownership of their patients’ care. One former learner stated, “She set expectations…we would be the ones who would be running the team, that you know it would very much be our team and that she is there to advise us and provide supervision but also safety for the patients as well.”

This study reveals the complex ways effective attendings build rapport, create a safe learning environment, utilize patient-centered teaching strategies, and engage in collaboration and coaching with all members of the team. These findings provide a framework of shared themes and their salient behaviors that may influence the success of inpatient general medicine clinician educators (Table 3).

There is a broad and voluminous literature on the subject of outstanding clinical teaching characteristics, much of which has shaped various faculty development curricula for decades. This study sought not to identify novel approaches of inpatient teaching necessarily but rather to closely examine the techniques and behaviors of clinician educators identified as exemplary. The findings affirm and reinforce the numerous, well-documented lists of personal attributes, techniques, and behaviors that resonate with learners, including creating a positive environment, demonstrating enthusiasm and interest in the learner, reading facial expressions, being student-centered, maintaining a high level of clinical knowledge, and utilizing effective communication skills.^18-24 The strengths of this study lie within the nuanced and rich observations and discussions that move beyond learners’ Likert scale evaluations and responses.^3-7,12 Input was sought from multiple perspectives on the care team, which provided detail from key stakeholders. Out of these comprehensive data arose several conclusions that extend the research literature on medical education.

In their seminal review, Sutkin et al.¹⁸ demonstrate that two thirds of characteristics of outstanding clinical teachers are “noncognitive” and that, “Perhaps what makes a clinical educator truly great depends less on the acquisition of cognitive skills such as medical knowledge and formulating learning objectives, and more on inherent, relationship-based, noncognitive attributes. Whereas cognitive abilities generally involve skills that may be taught and learned, albeit with difficulty, noncognitive abilities represent personal attributes, such as relationship skills, personality types, and emotional states, which are more difficult to develop and teach.”¹⁸ Our study, thus, adds to the literature by (1) highlighting examples of techniques and behaviors that encompass the crucial “noncognitive” arena and (2) informing best practices in teaching clinical medicine, especially those that resonate with learners, for future faculty development.

The findings highlight the role that relationships play in the teaching and learning of team-based medicine. Building rapport and sustaining successful relationships are cornerstones of effective teaching.¹⁸ For the attendings in this study, this manifested in observable, tangible behaviors such as greeting others by name, joking, using physical touch, and actively involving all team members, regardless of role or level of education. Previous literature has highlighted the importance of showing interest in learners.^7,19,25-27 This study provides multiple and varied examples of ways in which interest might be displayed.

For patients, the critical role of relationships was evidenced through rapport building and attention to patients as people outside their acute hospitalization. For instance, attendings regularly put patients’ medical issues into context and anticipated future outpatient challenges. To the authors’ knowledge, previous scholarship has not significantly emphasized this form of contextualized medicine, which involves the mindful consideration of the ongoing needs patients may experience upon transitions of care.

Several participants highlighted humility as an important characteristic of effective clinician educators. Attendings recognized that the field produces more new knowledge than can possibly be assimilated and that uncertainty is a mainstay of modern medical care. Attendings frequently utilized self-deprecation to acknowledge doubt, a technique that created a collaborative environment in which learners also felt safe to ask questions. These findings support the viewpoints by Reilly and Beckman that humility and an appreciation for questions and push-back from learners encourage lifelong learning through role modeling.^19,23 In responding to the interviewer’s question “And what happens when [the attending] is wrong?” one learner simply stated, “He makes fun of himself.”

This study has several limitations. First, it was conducted in a limited number of US based healthcare systems. The majority of institutions represented were larger, research intensive hospitals. While these hospitals were purposefully selected to provide a range in geography, size, type, and access to resources, the findings may differ in other settings. Second, it was conducted with a limited number of attendings and learners, which may limit the study’s generalizability. However, enough interviews were conducted to reach data saturation.¹⁵ Because evidence for a causal relationship between quality teaching and student and patient outcomes is lacking,¹⁸ we must rely on imperfect proxies for teaching excellence, including awards and recognition. This study attempted to identify exemplary educators through various means, but it is recognized that bias is likely. Third, because attendings provided lists of former learners, selection and recall biases may have been introduced, as attendings may have more readily identified former learners with whom they formed strong relationships. Fourth, focus was placed exclusively on teaching and learning within general medicine rounds. This was because there would be ample opportunity for teaching on this service, the structure of the teams and the types of patients would be comparable across sites, and the principal investigator was also a general medicine attending and would have a frame of reference for these types of rounds. Due to this narrow focus, the findings may not be generalizable to other subspecialties. Fifth, attendings were selected through a nonexhaustive method. However, the multisite design, the modified snowball sampling, and the inclusion of several types of institutions in the final participant pool introduced diversity to the final list. Finally, although we cannot discount the potential role of a Hawthorne effect on our data collection, the research team did attempt to mitigate this by standing apart from the care teams and remaining unobtrusive during observations.

Using a combination of interviews, focus group discussions, and direct observation, we identified consistent techniques and behaviors of excellent teaching attendings during inpatient general medicine rounds. We hope that all levels of clinician educators may use them to elevate their own teaching.

Disclosure

Dr. Saint is on a medical advisory board of Doximity, a new social networking site for physicians, and receives an honorarium. He is also on the scientific advisory board of Jvion, a healthcare technology company. Drs. Houchens, Harrod, Moody, and Ms. Fowler have no conflicts of interest.

Clinician educators face numerous obstacles to their joint mission of facilitating learning while also ensuring high-quality and patient-centered care. Time constraints, including the institution of house officer duty hour limitations,¹ shorter lengths of stay for hospitalized patients,² and competing career responsibilities, combine to create a dynamic learning environment. Additionally, clinician educators must balance the autonomy of their learners with the safety of their patients. They must teach to multiple learning levels and work collaboratively with multiple disciplines to foster an effective team-based approach to patient care. Yet, many clinician educators have no formal training in pedagogical methods.³ Such challenges necessitate increased attention to the work of excellent clinician educators and their respective teaching approaches.

Many studies of clinical teaching rely primarily on survey data of attributes of good clinical teachers.^3-7 While some studies have incorporated direct observations of teaching^8,9 or interviews with clinician educators or learners,^10,11 few have incorporated multiple perspectives from the current team and from former learners in order to provide a comprehensive picture of team-based learning.¹²

The goal of this study was to gain a thorough understanding, through multiple perspectives, of the techniques and behaviors used by exemplary educators within actual clinical environments. We studied attitudes, behaviors, and approaches of 12 such inpatient clinician educators.

Study Design and Sampling

This was a multisite study using an exploratory qualitative approach to inquiry. This approach was used to study the techniques and behaviors of excellent attendings during inpatient general medicine rounds. A modified snowball sampling approach¹³ was used, meaning individuals known to one member of the research team (SS) were initially contacted and asked to identify clinician educators (also referred to as attendings) for potential inclusion in the study. In an effort to identify attendings from a broad range of medical schools, the “2015 U.S. News and World Report Top Medical Schools: Research” rankings¹⁴ were also reviewed, with priority given to the top 25, as these are widely used to represent the best US hospitals. In an attempt to invite attendings from diverse institutions, additional medical schools not in the top 25 as well as historically black medical schools were also included. Division chiefs and chairs of internal medicine and/or directors of internal medicine residency programs at these schools were contacted and asked for recommendations of attendings, both within and outside their institutions, who they considered to be great inpatient teachers. In addition, key experts who have won teaching awards or were known to be specialists in the field of medical education were asked to nominate one or two other outstanding attendings.

By using this sampling method, 59 potential participants were identified. An internet search was conducted to obtain information about the potential participants and their institutions. Organizational characteristics such as geographic location, hospital size and affiliation, and patient population, as well as individual characteristics such as gender, medical education and training, and educational awards received were considered so that a diversity of organizations and backgrounds was represented. The list was narrowed down to 16 attendings who were contacted via e-mail and asked to participate. Interested participants were asked for a list of their current team members and 6 to 10 former learners to contact for interviews and focus groups. Former learners were included in an effort to better understand lasting effects on learners from their exemplary teaching attendings. A total of 12 attending physicians agreed to participate (Table 1). Literature on field methods has shown that 12 interviews are found to be adequate in accomplishing data saturation.¹⁵ Although 2 attendings were located at the same institution, we decided to include them given that both are recognized as master clinician educators and were each recommended by several individuals from various institutions. Hospitals were located throughout the US and included both university-affiliated hospitals and Veterans Affairs medical centers. Despite efforts to include physicians from historically black colleges and universities, only one attending was identified, and they declined the request to participate.

Data Collection

Observations. The one-day site visits were mainly conducted by two research team members, a physician (SS) and a medical anthropologist (MH), both of whom have extensive experience in qualitative methods. Teams were not uniform but were generally comprised of 1 attending, 1 senior medical resident, 1 to 2 interns, and approximately 2 medical students. Occasionally, a pharmacist, clinical assistant, or other health professional accompanied the team on rounds. Not infrequently, the bedside nurse would explicitly be included in the discussion regarding his or her specific patient. Each site visit began with observing attendings (N = 12) and current learners (N = 57) during rounds. Each research team member recorded their own observations via handwritten field notes, paying particular attention to group interactions, teaching approach, conversations occurring within and peripheral to the team, patient-team interactions, and the physical environment. By standing outside of the medical team circle and remaining silent during rounds, research team members remained unobtrusive to the discussion and process of rounds. Materials the attendings used during their teaching rounds were also documented and collected. Rounds generally lasted 2 to 3 hours. After each site visit, the research team met to compare and combine field notes.

Interviews and Focus Groups. The research team then conducted individual, semi-structured interviews with the attendings, focus groups with their current team (N = 46), and interviews or focus groups with their former learners (N = 26; Supplement 1). Eleven of the current team members observed during rounds were unable to participate in the focus groups due to clinical duties. Because the current learners who participated in the focus groups were also observed during rounds, the research team was able to ask them open-ended questions regarding teaching rounds and their roles as learners within this environment. Former learners who were still at the hospital participated in separate focus groups or interviews. Former learners who were no longer present at the hospital were contacted by telephone and individually interviewed by one research team member (MH). All interviews and focus groups were audio-recorded and transcribed.

This study was determined to be exempt by the University of Michigan Institutional Review Board. All participants were informed that their participation was completely voluntary and that they could terminate their involvement at any time.

Data Analysis

Data were analyzed using a thematic analysis approach.¹⁶ Thematic analysis entails reading through the data to identify patterns (and create codes) that relate to behaviors, experiences, meanings, and activities. Once patterns have been identified, they are grouped according to similarity into themes, which help to further explain the findings.¹⁷

After the first site visit was completed, the research team members that participated (SS and MH) met to develop initial ideas about meanings and possible patterns. All transcripts were read by one team member (MH) and, based on review of the data, codes were developed, defined, and documented in a codebook. This process was repeated after every site visit using the codebook to expand or combine codes and refine definitions as necessary. If a new code was added, the previously coded data were reviewed to apply the new code. NVivo® 10 software (QSR International; Melbourne, Australia) was used to manage the data.

Once all field notes and transcripts were coded (MH), the code reports, which list all data described within a specific code, were run to ensure consistency and identify relationships between codes. Once coding was verified, codes were grouped based on similarities and relationships into salient themes by 3 members of the research team (NH, MH, and SM). Themes, along with their supporting codes, were then further defined to understand how these attendings worked to facilitate excellent teaching in clinical settings.

The coded interview data and field notes were categorized into broad, overlapping themes. Three of these major themes include (1) fostering positive relationships, (2) patient-centered teaching, and (3) collaboration and coaching. Table 2 lists each theme, salient behaviors, examples, and selected quotes that further elucidate its meaning.

Fostering Positive Relationships

Attending physicians took observable steps to develop positive relationships with their team members, which in turn created a safe learning environment. For instance, attendings used learners’ first names, demonstrated interest in their well-being, deployed humor, and generally displayed informal actions—uncrossed arms, “fist bump” when recognizing learners’ success, standing outside the circle of team members and leaning in to listen—during learner interactions. Attendings also made it a priority to get to know individuals on a personal level. As one current learner put it, “He asks about where we are from. He will try to find some kind of connection that he can establish with not only each of the team members but also with each of the patients.”

Additionally, attendings built positive relationships with their learners by responding thoughtfully to their input, even when learners’ evaluations of patients required modification. In turn, learners reported feeling safe to ask questions, admit uncertainty, and respectfully disagree with their attendings. As one attending reflected, “If I can get them into a place where they feel like the learning environment is someplace where they can make a mistake and know that that mistake does not necessarily mean that it’s going to cost them in their evaluation part, then I feel like that’s why it’s important.”

To build rapport and create a safe learning environment, attendings used a number of strategies to position themselves as learners alongside their team members. For instance, attendings indicated that they wanted their ideas questioned because they saw it as an opportunity to learn. Moreover, in conversations with learners, attendings demonstrated humility, admitting when they did not know something. One former learner noted, “There have been times when he has asked [a] question…nobody knows and then he admits that he doesn’t know either. So everybody goes and looks it up…The whole thing turns out to be a fun learning experience.”

Attendings demonstrated respect for their team members’ time by reading about patients before rounds, identifying learning opportunities during rounds, and integrating teaching points into the daily work of patient care. Teaching was not relegated exclusively to the conference room or confined to the traditional “chalk talk” before or after rounds but rather was assimilated into daily workflow. They appeared to be responsive to the needs of individual patients and the team, which allowed attendings to both directly oversee their patients’ care and overcome the challenges of multiple competing demands for time. The importance of this approach was made clear by one current learner who stated “…she does prepare before, especially you know on call days, she does prepare for the new patients before coming in to staff, which is really appreciated… it saves a lot of time on rounds.”

Attendings also included other health professionals in team discussions. Attendings used many of the same relationship-building techniques with these professionals as they did with learners and patients. They consistently asked these professionals to provide insight and direction in patients’ plans of care. A former learner commented, “He always asks the [nurse] what is her impression of the patient...he truly values the [nurse’s] opinion of the patient.” One attending reiterated this approach, stating “I don’t want them to think that anything I have to say is more valuable than our pharmacist or the [nurse].”

Patient-Centered Teaching

Attending physicians modeled numerous teaching techniques that focused learning around the patient. Attendings knew their patients well through review of the medical records, discussion with the patient, and personal examination. This preparation allowed attendings to focus on key teaching points in the context of the patient. One former learner noted, “He tended to bring up a variety of things that really fit well into the clinical scenario. So whether that is talking about what is the differential for a new symptom that just came up for this patient or kind of here is a new paper talking about this condition or maybe some other pearl of physical exam for a patient that has a certain physical condition.”

Attendings served as effective role models by being directly involved in examining and talking with patients as well as demonstrating excellent physical examination and communication techniques. One current learner articulated the importance of learning these skills by observing them done well: “I think he teaches by example and by doing, again, those little things: being attentive to the patients and being very careful during exams…I think those are things that you teach people by doing them, not by saying you need to do this better during the patient encounter.”

Collaboration and Coaching

Attending physicians used varied collaboration and coaching techniques to facilitate learning across the entire care team. During rounds, attendings utilized visual aids to reinforce key concepts and simplify complex topics. They also collaborated by using discussion rather than lecture to engage with team members. For instance, attendings used Socratic questioning, asking questions that lead learners through critical thinking and allow them to solve problems themselves, to guide learners’ decision-making. One former learner reported, “He never gives you the answer, and he always asks your opinion; ‘So what are your thoughts on this?’”

Coaching for success, rather than directing the various team members, was emphasized. Attendings did not wish to be seen as the “leaders” of the team. During rounds, one attending was noted to explain his role in ensuring that the team was building connections with others: “When we have a bad outcome, if it feels like your soul has been ripped out, then you’ve done something right. You’ve made that connection with the patient. My job, as your coach, was to build communication between all of us so we feel vested in each other and our patients.”

Attendings also fostered clinical reasoning skills in their learners by encouraging them to verbalize their thought processes aloud in order to clarify and check for understanding. Attendings also placed emphasis not simply on memorizing content but rather prioritization of the patient’s problems and thinking step by step through individual medical problems. One current learner applauded an attending who could “come up with schematics of how to approach problems rather than feeding us factual information of this paper or this trial.”

Additionally, attendings facilitated learning across the entire care team by differentiating their teaching to meet the needs of multiple learning levels. While the entire team was explicitly included in the learning process, attendings encouraged learners to play various roles, execute tasks, and answer questions depending on their educational level. Attendings positioned learners as leaders of the team by allowing them to talk without interruption and by encouraging them to take ownership of their patients’ care. One former learner stated, “She set expectations…we would be the ones who would be running the team, that you know it would very much be our team and that she is there to advise us and provide supervision but also safety for the patients as well.”

This study reveals the complex ways effective attendings build rapport, create a safe learning environment, utilize patient-centered teaching strategies, and engage in collaboration and coaching with all members of the team. These findings provide a framework of shared themes and their salient behaviors that may influence the success of inpatient general medicine clinician educators (Table 3).

There is a broad and voluminous literature on the subject of outstanding clinical teaching characteristics, much of which has shaped various faculty development curricula for decades. This study sought not to identify novel approaches of inpatient teaching necessarily but rather to closely examine the techniques and behaviors of clinician educators identified as exemplary. The findings affirm and reinforce the numerous, well-documented lists of personal attributes, techniques, and behaviors that resonate with learners, including creating a positive environment, demonstrating enthusiasm and interest in the learner, reading facial expressions, being student-centered, maintaining a high level of clinical knowledge, and utilizing effective communication skills.^18-24 The strengths of this study lie within the nuanced and rich observations and discussions that move beyond learners’ Likert scale evaluations and responses.^3-7,12 Input was sought from multiple perspectives on the care team, which provided detail from key stakeholders. Out of these comprehensive data arose several conclusions that extend the research literature on medical education.

In their seminal review, Sutkin et al.¹⁸ demonstrate that two thirds of characteristics of outstanding clinical teachers are “noncognitive” and that, “Perhaps what makes a clinical educator truly great depends less on the acquisition of cognitive skills such as medical knowledge and formulating learning objectives, and more on inherent, relationship-based, noncognitive attributes. Whereas cognitive abilities generally involve skills that may be taught and learned, albeit with difficulty, noncognitive abilities represent personal attributes, such as relationship skills, personality types, and emotional states, which are more difficult to develop and teach.”¹⁸ Our study, thus, adds to the literature by (1) highlighting examples of techniques and behaviors that encompass the crucial “noncognitive” arena and (2) informing best practices in teaching clinical medicine, especially those that resonate with learners, for future faculty development.

The findings highlight the role that relationships play in the teaching and learning of team-based medicine. Building rapport and sustaining successful relationships are cornerstones of effective teaching.¹⁸ For the attendings in this study, this manifested in observable, tangible behaviors such as greeting others by name, joking, using physical touch, and actively involving all team members, regardless of role or level of education. Previous literature has highlighted the importance of showing interest in learners.^7,19,25-27 This study provides multiple and varied examples of ways in which interest might be displayed.

For patients, the critical role of relationships was evidenced through rapport building and attention to patients as people outside their acute hospitalization. For instance, attendings regularly put patients’ medical issues into context and anticipated future outpatient challenges. To the authors’ knowledge, previous scholarship has not significantly emphasized this form of contextualized medicine, which involves the mindful consideration of the ongoing needs patients may experience upon transitions of care.

Several participants highlighted humility as an important characteristic of effective clinician educators. Attendings recognized that the field produces more new knowledge than can possibly be assimilated and that uncertainty is a mainstay of modern medical care. Attendings frequently utilized self-deprecation to acknowledge doubt, a technique that created a collaborative environment in which learners also felt safe to ask questions. These findings support the viewpoints by Reilly and Beckman that humility and an appreciation for questions and push-back from learners encourage lifelong learning through role modeling.^19,23 In responding to the interviewer’s question “And what happens when [the attending] is wrong?” one learner simply stated, “He makes fun of himself.”

This study has several limitations. First, it was conducted in a limited number of US based healthcare systems. The majority of institutions represented were larger, research intensive hospitals. While these hospitals were purposefully selected to provide a range in geography, size, type, and access to resources, the findings may differ in other settings. Second, it was conducted with a limited number of attendings and learners, which may limit the study’s generalizability. However, enough interviews were conducted to reach data saturation.¹⁵ Because evidence for a causal relationship between quality teaching and student and patient outcomes is lacking,¹⁸ we must rely on imperfect proxies for teaching excellence, including awards and recognition. This study attempted to identify exemplary educators through various means, but it is recognized that bias is likely. Third, because attendings provided lists of former learners, selection and recall biases may have been introduced, as attendings may have more readily identified former learners with whom they formed strong relationships. Fourth, focus was placed exclusively on teaching and learning within general medicine rounds. This was because there would be ample opportunity for teaching on this service, the structure of the teams and the types of patients would be comparable across sites, and the principal investigator was also a general medicine attending and would have a frame of reference for these types of rounds. Due to this narrow focus, the findings may not be generalizable to other subspecialties. Fifth, attendings were selected through a nonexhaustive method. However, the multisite design, the modified snowball sampling, and the inclusion of several types of institutions in the final participant pool introduced diversity to the final list. Finally, although we cannot discount the potential role of a Hawthorne effect on our data collection, the research team did attempt to mitigate this by standing apart from the care teams and remaining unobtrusive during observations.

Using a combination of interviews, focus group discussions, and direct observation, we identified consistent techniques and behaviors of excellent teaching attendings during inpatient general medicine rounds. We hope that all levels of clinician educators may use them to elevate their own teaching.

Disclosure

Dr. Saint is on a medical advisory board of Doximity, a new social networking site for physicians, and receives an honorarium. He is also on the scientific advisory board of Jvion, a healthcare technology company. Drs. Houchens, Harrod, Moody, and Ms. Fowler have no conflicts of interest.

Clinician educators face numerous obstacles to their joint mission of facilitating learning while also ensuring high-quality and patient-centered care. Time constraints, including the institution of house officer duty hour limitations,¹ shorter lengths of stay for hospitalized patients,² and competing career responsibilities, combine to create a dynamic learning environment. Additionally, clinician educators must balance the autonomy of their learners with the safety of their patients. They must teach to multiple learning levels and work collaboratively with multiple disciplines to foster an effective team-based approach to patient care. Yet, many clinician educators have no formal training in pedagogical methods.³ Such challenges necessitate increased attention to the work of excellent clinician educators and their respective teaching approaches.

Many studies of clinical teaching rely primarily on survey data of attributes of good clinical teachers.^3-7 While some studies have incorporated direct observations of teaching^8,9 or interviews with clinician educators or learners,^10,11 few have incorporated multiple perspectives from the current team and from former learners in order to provide a comprehensive picture of team-based learning.¹²

The goal of this study was to gain a thorough understanding, through multiple perspectives, of the techniques and behaviors used by exemplary educators within actual clinical environments. We studied attitudes, behaviors, and approaches of 12 such inpatient clinician educators.

Study Design and Sampling

This was a multisite study using an exploratory qualitative approach to inquiry. This approach was used to study the techniques and behaviors of excellent attendings during inpatient general medicine rounds. A modified snowball sampling approach¹³ was used, meaning individuals known to one member of the research team (SS) were initially contacted and asked to identify clinician educators (also referred to as attendings) for potential inclusion in the study. In an effort to identify attendings from a broad range of medical schools, the “2015 U.S. News and World Report Top Medical Schools: Research” rankings¹⁴ were also reviewed, with priority given to the top 25, as these are widely used to represent the best US hospitals. In an attempt to invite attendings from diverse institutions, additional medical schools not in the top 25 as well as historically black medical schools were also included. Division chiefs and chairs of internal medicine and/or directors of internal medicine residency programs at these schools were contacted and asked for recommendations of attendings, both within and outside their institutions, who they considered to be great inpatient teachers. In addition, key experts who have won teaching awards or were known to be specialists in the field of medical education were asked to nominate one or two other outstanding attendings.

By using this sampling method, 59 potential participants were identified. An internet search was conducted to obtain information about the potential participants and their institutions. Organizational characteristics such as geographic location, hospital size and affiliation, and patient population, as well as individual characteristics such as gender, medical education and training, and educational awards received were considered so that a diversity of organizations and backgrounds was represented. The list was narrowed down to 16 attendings who were contacted via e-mail and asked to participate. Interested participants were asked for a list of their current team members and 6 to 10 former learners to contact for interviews and focus groups. Former learners were included in an effort to better understand lasting effects on learners from their exemplary teaching attendings. A total of 12 attending physicians agreed to participate (Table 1). Literature on field methods has shown that 12 interviews are found to be adequate in accomplishing data saturation.¹⁵ Although 2 attendings were located at the same institution, we decided to include them given that both are recognized as master clinician educators and were each recommended by several individuals from various institutions. Hospitals were located throughout the US and included both university-affiliated hospitals and Veterans Affairs medical centers. Despite efforts to include physicians from historically black colleges and universities, only one attending was identified, and they declined the request to participate.

Data Collection

Observations. The one-day site visits were mainly conducted by two research team members, a physician (SS) and a medical anthropologist (MH), both of whom have extensive experience in qualitative methods. Teams were not uniform but were generally comprised of 1 attending, 1 senior medical resident, 1 to 2 interns, and approximately 2 medical students. Occasionally, a pharmacist, clinical assistant, or other health professional accompanied the team on rounds. Not infrequently, the bedside nurse would explicitly be included in the discussion regarding his or her specific patient. Each site visit began with observing attendings (N = 12) and current learners (N = 57) during rounds. Each research team member recorded their own observations via handwritten field notes, paying particular attention to group interactions, teaching approach, conversations occurring within and peripheral to the team, patient-team interactions, and the physical environment. By standing outside of the medical team circle and remaining silent during rounds, research team members remained unobtrusive to the discussion and process of rounds. Materials the attendings used during their teaching rounds were also documented and collected. Rounds generally lasted 2 to 3 hours. After each site visit, the research team met to compare and combine field notes.

Interviews and Focus Groups. The research team then conducted individual, semi-structured interviews with the attendings, focus groups with their current team (N = 46), and interviews or focus groups with their former learners (N = 26; Supplement 1). Eleven of the current team members observed during rounds were unable to participate in the focus groups due to clinical duties. Because the current learners who participated in the focus groups were also observed during rounds, the research team was able to ask them open-ended questions regarding teaching rounds and their roles as learners within this environment. Former learners who were still at the hospital participated in separate focus groups or interviews. Former learners who were no longer present at the hospital were contacted by telephone and individually interviewed by one research team member (MH). All interviews and focus groups were audio-recorded and transcribed.

This study was determined to be exempt by the University of Michigan Institutional Review Board. All participants were informed that their participation was completely voluntary and that they could terminate their involvement at any time.

Data Analysis

Data were analyzed using a thematic analysis approach.¹⁶ Thematic analysis entails reading through the data to identify patterns (and create codes) that relate to behaviors, experiences, meanings, and activities. Once patterns have been identified, they are grouped according to similarity into themes, which help to further explain the findings.¹⁷

After the first site visit was completed, the research team members that participated (SS and MH) met to develop initial ideas about meanings and possible patterns. All transcripts were read by one team member (MH) and, based on review of the data, codes were developed, defined, and documented in a codebook. This process was repeated after every site visit using the codebook to expand or combine codes and refine definitions as necessary. If a new code was added, the previously coded data were reviewed to apply the new code. NVivo® 10 software (QSR International; Melbourne, Australia) was used to manage the data.

Once all field notes and transcripts were coded (MH), the code reports, which list all data described within a specific code, were run to ensure consistency and identify relationships between codes. Once coding was verified, codes were grouped based on similarities and relationships into salient themes by 3 members of the research team (NH, MH, and SM). Themes, along with their supporting codes, were then further defined to understand how these attendings worked to facilitate excellent teaching in clinical settings.

The coded interview data and field notes were categorized into broad, overlapping themes. Three of these major themes include (1) fostering positive relationships, (2) patient-centered teaching, and (3) collaboration and coaching. Table 2 lists each theme, salient behaviors, examples, and selected quotes that further elucidate its meaning.

Fostering Positive Relationships

Attending physicians took observable steps to develop positive relationships with their team members, which in turn created a safe learning environment. For instance, attendings used learners’ first names, demonstrated interest in their well-being, deployed humor, and generally displayed informal actions—uncrossed arms, “fist bump” when recognizing learners’ success, standing outside the circle of team members and leaning in to listen—during learner interactions. Attendings also made it a priority to get to know individuals on a personal level. As one current learner put it, “He asks about where we are from. He will try to find some kind of connection that he can establish with not only each of the team members but also with each of the patients.”

Additionally, attendings built positive relationships with their learners by responding thoughtfully to their input, even when learners’ evaluations of patients required modification. In turn, learners reported feeling safe to ask questions, admit uncertainty, and respectfully disagree with their attendings. As one attending reflected, “If I can get them into a place where they feel like the learning environment is someplace where they can make a mistake and know that that mistake does not necessarily mean that it’s going to cost them in their evaluation part, then I feel like that’s why it’s important.”

To build rapport and create a safe learning environment, attendings used a number of strategies to position themselves as learners alongside their team members. For instance, attendings indicated that they wanted their ideas questioned because they saw it as an opportunity to learn. Moreover, in conversations with learners, attendings demonstrated humility, admitting when they did not know something. One former learner noted, “There have been times when he has asked [a] question…nobody knows and then he admits that he doesn’t know either. So everybody goes and looks it up…The whole thing turns out to be a fun learning experience.”

Attendings demonstrated respect for their team members’ time by reading about patients before rounds, identifying learning opportunities during rounds, and integrating teaching points into the daily work of patient care. Teaching was not relegated exclusively to the conference room or confined to the traditional “chalk talk” before or after rounds but rather was assimilated into daily workflow. They appeared to be responsive to the needs of individual patients and the team, which allowed attendings to both directly oversee their patients’ care and overcome the challenges of multiple competing demands for time. The importance of this approach was made clear by one current learner who stated “…she does prepare before, especially you know on call days, she does prepare for the new patients before coming in to staff, which is really appreciated… it saves a lot of time on rounds.”

Attendings also included other health professionals in team discussions. Attendings used many of the same relationship-building techniques with these professionals as they did with learners and patients. They consistently asked these professionals to provide insight and direction in patients’ plans of care. A former learner commented, “He always asks the [nurse] what is her impression of the patient...he truly values the [nurse’s] opinion of the patient.” One attending reiterated this approach, stating “I don’t want them to think that anything I have to say is more valuable than our pharmacist or the [nurse].”

Patient-Centered Teaching

Attending physicians modeled numerous teaching techniques that focused learning around the patient. Attendings knew their patients well through review of the medical records, discussion with the patient, and personal examination. This preparation allowed attendings to focus on key teaching points in the context of the patient. One former learner noted, “He tended to bring up a variety of things that really fit well into the clinical scenario. So whether that is talking about what is the differential for a new symptom that just came up for this patient or kind of here is a new paper talking about this condition or maybe some other pearl of physical exam for a patient that has a certain physical condition.”

Attendings served as effective role models by being directly involved in examining and talking with patients as well as demonstrating excellent physical examination and communication techniques. One current learner articulated the importance of learning these skills by observing them done well: “I think he teaches by example and by doing, again, those little things: being attentive to the patients and being very careful during exams…I think those are things that you teach people by doing them, not by saying you need to do this better during the patient encounter.”

Collaboration and Coaching

Attending physicians used varied collaboration and coaching techniques to facilitate learning across the entire care team. During rounds, attendings utilized visual aids to reinforce key concepts and simplify complex topics. They also collaborated by using discussion rather than lecture to engage with team members. For instance, attendings used Socratic questioning, asking questions that lead learners through critical thinking and allow them to solve problems themselves, to guide learners’ decision-making. One former learner reported, “He never gives you the answer, and he always asks your opinion; ‘So what are your thoughts on this?’”

Coaching for success, rather than directing the various team members, was emphasized. Attendings did not wish to be seen as the “leaders” of the team. During rounds, one attending was noted to explain his role in ensuring that the team was building connections with others: “When we have a bad outcome, if it feels like your soul has been ripped out, then you’ve done something right. You’ve made that connection with the patient. My job, as your coach, was to build communication between all of us so we feel vested in each other and our patients.”

Attendings also fostered clinical reasoning skills in their learners by encouraging them to verbalize their thought processes aloud in order to clarify and check for understanding. Attendings also placed emphasis not simply on memorizing content but rather prioritization of the patient’s problems and thinking step by step through individual medical problems. One current learner applauded an attending who could “come up with schematics of how to approach problems rather than feeding us factual information of this paper or this trial.”

Additionally, attendings facilitated learning across the entire care team by differentiating their teaching to meet the needs of multiple learning levels. While the entire team was explicitly included in the learning process, attendings encouraged learners to play various roles, execute tasks, and answer questions depending on their educational level. Attendings positioned learners as leaders of the team by allowing them to talk without interruption and by encouraging them to take ownership of their patients’ care. One former learner stated, “She set expectations…we would be the ones who would be running the team, that you know it would very much be our team and that she is there to advise us and provide supervision but also safety for the patients as well.”

This study reveals the complex ways effective attendings build rapport, create a safe learning environment, utilize patient-centered teaching strategies, and engage in collaboration and coaching with all members of the team. These findings provide a framework of shared themes and their salient behaviors that may influence the success of inpatient general medicine clinician educators (Table 3).

There is a broad and voluminous literature on the subject of outstanding clinical teaching characteristics, much of which has shaped various faculty development curricula for decades. This study sought not to identify novel approaches of inpatient teaching necessarily but rather to closely examine the techniques and behaviors of clinician educators identified as exemplary. The findings affirm and reinforce the numerous, well-documented lists of personal attributes, techniques, and behaviors that resonate with learners, including creating a positive environment, demonstrating enthusiasm and interest in the learner, reading facial expressions, being student-centered, maintaining a high level of clinical knowledge, and utilizing effective communication skills.^18-24 The strengths of this study lie within the nuanced and rich observations and discussions that move beyond learners’ Likert scale evaluations and responses.^3-7,12 Input was sought from multiple perspectives on the care team, which provided detail from key stakeholders. Out of these comprehensive data arose several conclusions that extend the research literature on medical education.

In their seminal review, Sutkin et al.¹⁸ demonstrate that two thirds of characteristics of outstanding clinical teachers are “noncognitive” and that, “Perhaps what makes a clinical educator truly great depends less on the acquisition of cognitive skills such as medical knowledge and formulating learning objectives, and more on inherent, relationship-based, noncognitive attributes. Whereas cognitive abilities generally involve skills that may be taught and learned, albeit with difficulty, noncognitive abilities represent personal attributes, such as relationship skills, personality types, and emotional states, which are more difficult to develop and teach.”¹⁸ Our study, thus, adds to the literature by (1) highlighting examples of techniques and behaviors that encompass the crucial “noncognitive” arena and (2) informing best practices in teaching clinical medicine, especially those that resonate with learners, for future faculty development.

The findings highlight the role that relationships play in the teaching and learning of team-based medicine. Building rapport and sustaining successful relationships are cornerstones of effective teaching.¹⁸ For the attendings in this study, this manifested in observable, tangible behaviors such as greeting others by name, joking, using physical touch, and actively involving all team members, regardless of role or level of education. Previous literature has highlighted the importance of showing interest in learners.^7,19,25-27 This study provides multiple and varied examples of ways in which interest might be displayed.

For patients, the critical role of relationships was evidenced through rapport building and attention to patients as people outside their acute hospitalization. For instance, attendings regularly put patients’ medical issues into context and anticipated future outpatient challenges. To the authors’ knowledge, previous scholarship has not significantly emphasized this form of contextualized medicine, which involves the mindful consideration of the ongoing needs patients may experience upon transitions of care.

Several participants highlighted humility as an important characteristic of effective clinician educators. Attendings recognized that the field produces more new knowledge than can possibly be assimilated and that uncertainty is a mainstay of modern medical care. Attendings frequently utilized self-deprecation to acknowledge doubt, a technique that created a collaborative environment in which learners also felt safe to ask questions. These findings support the viewpoints by Reilly and Beckman that humility and an appreciation for questions and push-back from learners encourage lifelong learning through role modeling.^19,23 In responding to the interviewer’s question “And what happens when [the attending] is wrong?” one learner simply stated, “He makes fun of himself.”

This study has several limitations. First, it was conducted in a limited number of US based healthcare systems. The majority of institutions represented were larger, research intensive hospitals. While these hospitals were purposefully selected to provide a range in geography, size, type, and access to resources, the findings may differ in other settings. Second, it was conducted with a limited number of attendings and learners, which may limit the study’s generalizability. However, enough interviews were conducted to reach data saturation.¹⁵ Because evidence for a causal relationship between quality teaching and student and patient outcomes is lacking,¹⁸ we must rely on imperfect proxies for teaching excellence, including awards and recognition. This study attempted to identify exemplary educators through various means, but it is recognized that bias is likely. Third, because attendings provided lists of former learners, selection and recall biases may have been introduced, as attendings may have more readily identified former learners with whom they formed strong relationships. Fourth, focus was placed exclusively on teaching and learning within general medicine rounds. This was because there would be ample opportunity for teaching on this service, the structure of the teams and the types of patients would be comparable across sites, and the principal investigator was also a general medicine attending and would have a frame of reference for these types of rounds. Due to this narrow focus, the findings may not be generalizable to other subspecialties. Fifth, attendings were selected through a nonexhaustive method. However, the multisite design, the modified snowball sampling, and the inclusion of several types of institutions in the final participant pool introduced diversity to the final list. Finally, although we cannot discount the potential role of a Hawthorne effect on our data collection, the research team did attempt to mitigate this by standing apart from the care teams and remaining unobtrusive during observations.

Using a combination of interviews, focus group discussions, and direct observation, we identified consistent techniques and behaviors of excellent teaching attendings during inpatient general medicine rounds. We hope that all levels of clinician educators may use them to elevate their own teaching.

Disclosure

Dr. Saint is on a medical advisory board of Doximity, a new social networking site for physicians, and receives an honorarium. He is also on the scientific advisory board of Jvion, a healthcare technology company. Drs. Houchens, Harrod, Moody, and Ms. Fowler have no conflicts of interest.

When a patient presents with suspected venous thromboembolism, ie, deep vein thrombosis or pulmonary embolism, what diagnostic tests are needed to confirm the diagnosis? The clinical signs and symptoms of venous thromboembolism are nonspecific and often difficult to interpret. Therefore, it is essential for clinicians to use a standardized, structured approach to diagnosis that incorporates clinical findings and laboratory testing, as well as judicious use of diagnostic imaging. But while information is important, clinicians must also strive to avoid unnecessary testing, not only to decrease costs, but also to avoid potential harm.

If the diagnosis is confirmed, does the patient need testing for an underlying thrombophilic disorder? Such screening is often considered after a thromboembolic event occurs. However, a growing body of evidence indicates that the results of thrombophilia testing can be misinterpreted and potentially harmful.¹ We need to understand the utility of this testing as well as when and how it should be used. Patients and thrombosis specialists should be involved in deciding whether to perform these tests.

In this article, we provide practical information about how to diagnose venous thromboembolism, including strategies to optimize testing in suspected cases. We also offer guidance on how to decide whether further thrombophilia testing is warranted.

COMMON AND SERIOUS

Venous thromboembolism is a major cause of morbidity and death. Approximately 900,000 cases of pulmonary embolism and deep vein thrombosis occur in the United States each year, causing 60,000 to 300,000 deaths,² with the number of cases projected to double over the next 40 years.³

INITIAL APPROACH: PRETEST PROBABILITY

Given the morbidity and mortality associated with venous thromboembolism, prompt recognition and diagnosis are imperative. Clinical diagnosis alone is insufficient, with confirmed disease found in only 15% to 25% of patients suspected of having venous thromboembolism.^4–8 Therefore, the pretest probability should be coupled with objective testing.

The Wells score shows good discrimination in the outpatient and emergency department settings, but it has been invalidated in the inpatient setting, and thus it should not be used in inpatients.¹⁰

LABORATORY TESTS FOR SUSPECTED VENOUS THROMBOEMBOLISM

Employing an understanding of diagnostic testing is fundamental to identifying patients with venous thromboembolism.

D-dimer is a byproduct of fibrinolysis.

D-dimer testing has very high sensitivity for venous thromboembolism (> 90%) but low specificity (about 50%), and levels can be elevated in a variety of situations such as advanced age, acute inflammation, and cancer.¹⁵ The standard threshold is 500 μg/L, but because the D-dimer level increases with age, some clinicians advocate using an age-adjusted threshold for patients age 50 or older (age in years × 10 μg/L) to increase the diagnostic yield.¹⁶

Of the laboratory tests for D-dimer, the enzyme-linked immunosorbent assay has the highest sensitivity and highest negative predictive value (100%) and may be preferred over the other test methodologies.¹⁷

With its high sensitivity, D-dimer testing is clinically useful for ruling out venous thromboembolism, particularly when the pretest probability is low, but it lacks the specificity required for diagnosing and treating the disease if positive. Thus, it is not useful for ruling in venous thromboembolism. If the patient has a high pretest probability, we can omit D-dimer testing in favor of imaging studies.

Other laboratory tests such as arterial blood gas and brain natriuretic peptide levels have been proposed as markers of pulmonary embolism, but studies suggest they have limited utility in predicting the presence of disease.^18,19

DIAGNOSTIC TESTS FOR DEEP VEIN THROMBOSIS

Ultrasonography

If the pretest probability of deep vein thrombosis is high or a D-dimer test is found to be positive, the next step in evaluation is compression ultrasonography. 

While some guidelines recommend scanning only the proximal leg, many facilities in the United States scan the whole leg, which may reveal distal deep vein thrombosis.²⁰ The clinical significance of isolated distal deep vein thrombosis is unknown, and a selective anticoagulation approach may be used if this condition is discovered. The 2012 and 2016 American College of Chest Physicians (ACCP) guidelines on diagnosis and management of venous thromboembolism address this topic.^20,21

Deep vein thrombosis in the arm should be evaluated in the same manner as in the lower extremities.

Venography

Invasive and therefore no longer often used, venography is considered the gold standard for diagnosing deep vein thrombosis. Computed tomographic (CT) or magnetic resonance (MR) venography is most useful if the patient has aberrant anatomy such as a deformity of the leg, or in situations where the use of ultrasonography is difficult or unreliable, such as in the setting of severe obesity. CT or MR venography may be considered when looking for thrombosis in noncompressible veins of the thorax and abdomen (eg, the subclavian vein, iliac vein, and inferior vena cava) if ultrasonography is negative but clinical suspicion is high. Venous-phase CT angiography is particularly useful in diagnosing deep vein thrombosis in the inferior vena cava and iliac vein when deep vein thrombosis is clinically suspected but cannot be visualized on duplex ultrasonography.

Computed tomography

Imaging is warranted in patients who have a high pretest probability of pulmonary embolism, or in whom the D-dimer assay was positive but the pretest probability was low or moderate.

Once the gold standard, pulmonary angiography is no longer recommended for the initial diagnosis of pulmonary embolism because it is invasive, often unavailable, less sophisticated, and more expensive than noninvasive imaging techniques such as CT angiography. It is still used, however, in catheter-directed thrombolysis.

Thus, multiphasic CT angiography, as guided by pretest probability and the D-dimer level, is the imaging test of choice in the evaluation of pulmonary embolism. It can also offer insight into thrombotic burden and can reveal concurrent or alternative diagnoses (eg, pneumonia).

Ventilation-perfusion scanning

When CT angiography is unavailable or the patient should not be exposed to contrast medium (eg, due to concern for contrast-induced nephropathy or contrast allergy), ventilation-perfusion (V/Q) scanning remains an option for ruling out pulmonary embolism.²²

Anderson et al²³ compared CT angiography and V/Q scanning in a study in 1,417 patients considered likely to have acute pulmonary embolism. Rates of symptomatic pulmonary embolism during 3-month follow-up were similar in patients who initially had negative results on V/Q scanning compared with those who initially had negative results on CT angiography. However, this study used single-detector CT scanners for one-third of the patients. Therefore, the results may have been different if current technology had been used.

Limitations of V/Q scanning include length of time to perform (30–45 minutes), cost, inability to identify other causes of symptoms, and difficulty with interpretation  when other pulmonary pathology is present (eg, lung infiltrate). V/Q scanning is helpful when negative but is often reported based on probability (low, intermediate, or high) and may not provide adequate guidance. Therefore, CT angiography should be used whenever possible for diagnosing pulmonary embolism.

Other tests for pulmonary embolism

Electrocardiography, transthoracic echocardiography, and chest radiography may aid in the search for alternative diagnoses and assess the degree of right heart strain as a sequela of pulmonary embolism, but they do not confirm the diagnosis.

ORDER IMAGING ONLY IF NEEDED

Diagnostic imaging can be optimized by avoiding unnecessary tests that carry both costs and clinical risks.

Most patients in whom acute pulmonary embolism is discovered will not need testing for deep vein thrombosis, as they will receive anticoagulation regardless. Similarly, many patients with acute symptomatic deep vein thrombosis do not need testing for pulmonary embolism with chest CT imaging, as they too will receive anticoagulation regardless.

Therefore, clinicians are encouraged to use diagnostic reasoning while practicing high-value care (including estimating pretest probability and measuring D-dimer when appropriate), ordering additional tests judiciously and only if indicated.

THROMBOEMBOLISM IS CONFIRMED—IS FURTHER TESTING WARRANTED?

Once acute venous thromboembolism is confirmed, key considerations include whether the event was provoked or unprovoked (ie, idiopathic) and whether the patient needs indefinite anticoagulation (eg, after 2 or more unprovoked events).

Was the event provoked or unprovoked?

Even in cases of unprovoked venous thromboembolism, no clear consensus exists as to which patients should be tested for thrombophilia. Experts do advocate, however, that it be done only in highly selected patients and that it be coordinated with the patient, family members, and an expert in this testing. Patients for whom further testing may be considered include those with venous thromboembolism in unusual sites (eg, the cavernous sinus), with warfarin-induced skin necrosis, or with recurrent pregnancy loss.

While screening for malignancy may seem prudent in the case of unexplained venous thromboembolism, the use of CT imaging for this purpose has been found to be of low yield. In one study,²⁴ it was not found to detect additional neoplasms, and it can lead to additional cost and no added benefit for patients.

The American Board of Internal Medicine’s Choosing Wisely campaign strongly recommends consultation with an expert in thrombophilia (eg, a hematologist) before testing.²⁵ Ordering multiple tests in bundles (hypercoagulability panels) is unlikely to alter management, could have a negative clinical impact on patients, and is generally not recommended.

The ‘4 Ps’ approach to testing

• Patient selection
• Pretest counseling
• Proper laboratory interpretation
• Provision of education and advice.

Importantly, testing should be reserved for patients in whom the pretest probability of the thrombophilic disease is moderate to high, such as testing for antiphospholipid antibody syndrome in patients with systemic lupus erythematosus or recurrent miscarriage.

Venous thromboembolism in a patient who is known to have a malignant disease does not typically warrant further thrombophilia testing, as the event was likely a sequela of the malignancy. The evaluation and management of venous thromboembolism with concurrent neoplasm is covered elsewhere.²¹

Thrombophilia testing should be approached the same regardless of whether the venous thromboembolism was diagnosed intentionally or incidentally. First, determine whether the thrombosis was provoked or unprovoked, then order additional tests only if indicated, as recommended. Alternative approaches such as forgoing anticoagulation (but performing serial imaging, if indicated) may be reasonable if the thrombus is deemed clinically irrelevant (eg, nonocclusive, asymptomatic, subsegmental pulmonary embolism in the absence of proximal deep vein thrombosis; isolated distal deep vein thrombosis).^25,27

It is still debatable whether the increasing incidence of asymptomatic pulmonary embolism due to enhanced sensitivity of noninvasive diagnostic imaging warrants a change in diagnostic approach.²⁸

FACTORS TO CONSIDER BEFORE THROMBOPHILIA TESTING

Important factors to consider before testing for thrombophilia are²⁹:

• How will the results affect the anticoagulation plan?
• How may the patient’s clinical status and medications influence the results?
• Has the patient expressed a desire to understand why venous thromboembolism occurred?
• Will the results have a potential impact on the patient’s family members?

How will the results of thrombophilia testing affect anticoagulation management?

Because the goal of any diagnostic test is to find out what type of care the patient needs, clinicians must determine whether knowledge of an underlying thrombophilia will alter the short-term or long-term anticoagulation therapy the patient is receiving for an acute venous thromboembolic event.

As most acute episodes of venous thromboembolism require an initial 3 months of anticoagulation (with the exception of some nonclinically relevant events such as isolated distal deep vein thrombosis without extension on reimaging), testing in the acute setting does not change the short-term management of anticoagulation. Many hospitals have advocated for outpatient-only thrombophilia testing (if testing does occur), as testing in the acute setting may render test results uninterpretable (see What factors can influence thrombophilia testing? below) and can inappropriately affect the long-term management of anticoagulation. We recommend against testing in the inpatient setting.

To determine the duration of anticoagulation, clinicians must balance the risk of recurrent venous thromboembolism and the risk of bleeding. If a patient is at significant risk of bleeding or does not tolerate anticoagulation, clinicians may consider stopping therapy instead of evaluating for thrombophilia. For patients with provoked venous thromboembolism, anticoagulation should generally be limited to 3 months, as the risk of recurrence does not outweigh the risk of bleeding with continued anticoagulation therapy.

Patients with unprovoked venous thromboembolism have a risk of recurrence twice as high as those with provoked venous thromboembolism and generally need a longer duration of anticoagulation.^30,31 Once a patient with an unprovoked venous thromboembolic event has completed the initial 3 months of anticoagulation, a formal risk-benefit evaluation should be performed to determine whether to continue it.

Up to 42% of patients with unprovoked venous thromboembolism may have 1 or more thrombotic disorders, and some clinicians believe that detecting an underlying thrombophilia will aid in decisions regarding duration of therapy.³² However, the risk of recurrent venous thromboembolism in these patients does not differ significantly from that in patients without an underlying thrombophilia.^33–35 As such, it has been suggested that the unprovoked character of the thrombotic event, rather than an underlying thrombophilia, determines the risk of future recurrence and should be used instead of testing to guide the duration of anticoagulation therapy.³²

For more information, see the 2016 ACCP guideline update on antithrombotic therapy for venous thromboembolism.²⁷

What factors can influence the results of thrombophilia testing?

For example, antithrombin is consumed during thrombus formation; therefore, antithrombin levels may be transiently suppressed in acute venous thromboembolism. Moreover, since antithrombin binds to unfractionated heparin, low-molecular-weight heparin, and fondaparinux and mediates their activity as anticoagulants, antithrombin levels may be decreased by heparin therapy.

Similarly, vitamin K antagonists (eg, warfarin) suppress protein C and S activity levels by inhibiting vitamin K epoxide reductase and may falsely indicate a protein C or S deficiency.

Direct oral anticoagulants can cause false-positive results on lupus anticoagulant assays (dilute Russell viper venom time, augmented partial thromboplastin time), raise protein C, protein S, and antithrombin activity levels, and normalize activated protein C resistance assays, leading to missed diagnoses.⁴¹

Since estrogen therapy and pregnancy lead to increases in C4b binding protein, resulting in decreased free protein S, these situations can result in clinicians falsely labeling patients as having congenital protein S deficiency when in fact the patient had a transient reduction in protein S levels.³³

Therefore, to optimize accuracy and interpretation of results, thrombophilia testing should ideally be performed when the patient:

• Is past the acute event and out of the hospital
• Is not pregnant
• Has received the required 3 months of anticoagulation and is off this therapy.

For warfarin, most recommendations say that testing should be performed after the patient has been off therapy for 2 to 6 weeks.⁴² Low-molecular-weight heparins and direct oral anticoagulants should be discontinued for at least 48 to 72 hours, or longer if the patient has kidney impairment, as these medications are renally eliminated.

Genetic tests such as factor V Leiden and prothrombin gene mutation are not affected by these factors and do not require repeat or confirmatory testing.

What if the patient or family wants to understand why an event occurred?

Some experts advocate thrombophilia testing of asymptomatic family members to identify carriers who may need prophylaxis against venous thromboembolism in high-risk situations such as pregnancy, oral contraceptive use, hospitalization, and surgery.²⁹ Asymptomatic family members of a first-degree relative with a history of venous thromboembolism have a 2 times higher risk of an index event.⁴³ Thus, it may be argued that these asymptomatic individuals should receive prophylactic measures in any high-risk situation, based on the family history itself rather than results of thrombophilia testing.

Occasionally, patients and family members want to know the cause of the thrombotic event and want to be tested. In these instances, pretest counseling for the patient and family about the potential implications of testing and shared decision-making between the provider and patient are of utmost importance.²⁹

What is the impact on family members if thrombophilia is diagnosed?

While positive test results can give patients some satisfaction, this knowledge may also cause unnecessary worry, as the patient knows he or she has a hematologic disorder and could possible die of venous thromboembolism.

Thrombophilia testing can have other adverse consequences. For example, while the Genetic Information Nondiscrimination Act of 2008 protects against denial of health insurance benefits based on genetic information, known carriers of thrombophilia may have trouble obtaining life or disability insurance.⁴⁴

Unfortunately, it is not uncommon for thrombophilia testing to be inappropriately performed, interpreted, or followed up. These suboptimal approaches can lead to unnecessary exposure to high-risk therapeutic anticoagulation, excessive durations of therapy, and labeling with an unconfirmed or incorrect diagnosis. Additionally, there are significant costs associated with thrombophilia testing, including the cost of the tests and anticoagulant medications and management of adverse events such as bleeding.

WHAT ARE THE ALTERNATIVES TO THROMBOPHILIA TESTING?

Because discovered thrombophilias (eg, factor V Leiden mutation, prothrombin gene mutation) have not consistently shown a strong correlation with increased recurrence of venous thromboembolism, alternative approaches are emerging to determine the duration of therapy for unprovoked events.

Clinical prediction tools based on patient characteristics and laboratory markers that are more consistently associated with recurrent venous thromboembolism (eg, male sex, persistently elevated D-dimer) have been developed to aid clinicians dealing with this challenging question. Several prediction tools are available:

The “Men Continue and HERDOO2” rule (HERDOO2 = hyperpigmentation, edema, or redness in either leg; D-dimer level ≥ 250 μg/L; obesity with body mass index ≥ 30 kg/m²; or older age, ≥ 65)⁴⁵

The DASH score (D-dimer, age, sex, and hormonal therapy)⁴⁶

The Vienna score,^47,48 at http://cemsiis.meduniwien.ac.at/en/kb/science-research/software/clinical-software/recurrent-vte/.

SUMMARY OF THROMBOPHILIA TESTING RECOMMENDATIONS

Test for thrombophilia only when…

• Discussing with a specialist (eg, hematologist) who has an understanding of thrombophilia
• Using the 4 Ps approach
• A patient requests testing to understand why a thrombotic event occurred, and the patient understands the implications of testing (ie, received counseling) for self and for family
• An expert deems identification of asymptomatic family members important for those who may be carriers of a detected thrombophilia
• The patient with a venous thromboembolic event has completed 3 months of anticoagulation and has been off anticoagulation for the appropriate length of time
• The results will change management.

Forgo thrombophilia testing when…

• A patient has a provoked venous thromboembolic event
• You do not intend to discontinue anticoagulation (ie, anticoagulation is indefinite)
• The patient is in the acute (eg, inpatient) setting
• The patient is on anticoagulants that may render test results uninterpretable
• The patient is pregnant or on oral contraceptives
• Use of alternative patient characteristics and laboratory markers to predict venous thromboembolism recurrence may be an option.

OPTIMIZING THE DIAGNOSIS

With the incidence of venous thromboembolism rapidly increasing, optimizing its diagnosis from both a financial and clinical perspective is becoming increasingly important. Clinicians should be familiar with the use of pretest probability scoring for venous thromboembolism, as well as which diagnostic tests are preferred if further workup is indicated. They should strive to minimize or avoid indiscriminate thrombophilia testing, which may lead to increased healthcare costs and patient exposure to potentially harmful anticoagulation.

Testing for thrombophilia should be based on whether a venous thromboembolic event was provoked or unprovoked. Patients with provoked venous thromboembolism or those receiving indefinite anticoagulation therapy should not be tested for thrombophilia. If testing is being considered in a patient with unprovoked venous thromboembolism, a specialist who is able to implement the 4 Ps approach should be consulted to ensure well-informed, shared decision-making with patients and family members.

When a patient presents with suspected venous thromboembolism, ie, deep vein thrombosis or pulmonary embolism, what diagnostic tests are needed to confirm the diagnosis? The clinical signs and symptoms of venous thromboembolism are nonspecific and often difficult to interpret. Therefore, it is essential for clinicians to use a standardized, structured approach to diagnosis that incorporates clinical findings and laboratory testing, as well as judicious use of diagnostic imaging. But while information is important, clinicians must also strive to avoid unnecessary testing, not only to decrease costs, but also to avoid potential harm.

If the diagnosis is confirmed, does the patient need testing for an underlying thrombophilic disorder? Such screening is often considered after a thromboembolic event occurs. However, a growing body of evidence indicates that the results of thrombophilia testing can be misinterpreted and potentially harmful.¹ We need to understand the utility of this testing as well as when and how it should be used. Patients and thrombosis specialists should be involved in deciding whether to perform these tests.

In this article, we provide practical information about how to diagnose venous thromboembolism, including strategies to optimize testing in suspected cases. We also offer guidance on how to decide whether further thrombophilia testing is warranted.

COMMON AND SERIOUS

Venous thromboembolism is a major cause of morbidity and death. Approximately 900,000 cases of pulmonary embolism and deep vein thrombosis occur in the United States each year, causing 60,000 to 300,000 deaths,² with the number of cases projected to double over the next 40 years.³

INITIAL APPROACH: PRETEST PROBABILITY

Given the morbidity and mortality associated with venous thromboembolism, prompt recognition and diagnosis are imperative. Clinical diagnosis alone is insufficient, with confirmed disease found in only 15% to 25% of patients suspected of having venous thromboembolism.^4–8 Therefore, the pretest probability should be coupled with objective testing.

The Wells score shows good discrimination in the outpatient and emergency department settings, but it has been invalidated in the inpatient setting, and thus it should not be used in inpatients.¹⁰

LABORATORY TESTS FOR SUSPECTED VENOUS THROMBOEMBOLISM

Employing an understanding of diagnostic testing is fundamental to identifying patients with venous thromboembolism.

D-dimer is a byproduct of fibrinolysis.

D-dimer testing has very high sensitivity for venous thromboembolism (> 90%) but low specificity (about 50%), and levels can be elevated in a variety of situations such as advanced age, acute inflammation, and cancer.¹⁵ The standard threshold is 500 μg/L, but because the D-dimer level increases with age, some clinicians advocate using an age-adjusted threshold for patients age 50 or older (age in years × 10 μg/L) to increase the diagnostic yield.¹⁶

Of the laboratory tests for D-dimer, the enzyme-linked immunosorbent assay has the highest sensitivity and highest negative predictive value (100%) and may be preferred over the other test methodologies.¹⁷

With its high sensitivity, D-dimer testing is clinically useful for ruling out venous thromboembolism, particularly when the pretest probability is low, but it lacks the specificity required for diagnosing and treating the disease if positive. Thus, it is not useful for ruling in venous thromboembolism. If the patient has a high pretest probability, we can omit D-dimer testing in favor of imaging studies.

Other laboratory tests such as arterial blood gas and brain natriuretic peptide levels have been proposed as markers of pulmonary embolism, but studies suggest they have limited utility in predicting the presence of disease.^18,19

DIAGNOSTIC TESTS FOR DEEP VEIN THROMBOSIS

Ultrasonography

If the pretest probability of deep vein thrombosis is high or a D-dimer test is found to be positive, the next step in evaluation is compression ultrasonography. 

While some guidelines recommend scanning only the proximal leg, many facilities in the United States scan the whole leg, which may reveal distal deep vein thrombosis.²⁰ The clinical significance of isolated distal deep vein thrombosis is unknown, and a selective anticoagulation approach may be used if this condition is discovered. The 2012 and 2016 American College of Chest Physicians (ACCP) guidelines on diagnosis and management of venous thromboembolism address this topic.^20,21

Deep vein thrombosis in the arm should be evaluated in the same manner as in the lower extremities.

Venography

Invasive and therefore no longer often used, venography is considered the gold standard for diagnosing deep vein thrombosis. Computed tomographic (CT) or magnetic resonance (MR) venography is most useful if the patient has aberrant anatomy such as a deformity of the leg, or in situations where the use of ultrasonography is difficult or unreliable, such as in the setting of severe obesity. CT or MR venography may be considered when looking for thrombosis in noncompressible veins of the thorax and abdomen (eg, the subclavian vein, iliac vein, and inferior vena cava) if ultrasonography is negative but clinical suspicion is high. Venous-phase CT angiography is particularly useful in diagnosing deep vein thrombosis in the inferior vena cava and iliac vein when deep vein thrombosis is clinically suspected but cannot be visualized on duplex ultrasonography.

Computed tomography

Imaging is warranted in patients who have a high pretest probability of pulmonary embolism, or in whom the D-dimer assay was positive but the pretest probability was low or moderate.

Once the gold standard, pulmonary angiography is no longer recommended for the initial diagnosis of pulmonary embolism because it is invasive, often unavailable, less sophisticated, and more expensive than noninvasive imaging techniques such as CT angiography. It is still used, however, in catheter-directed thrombolysis.

Thus, multiphasic CT angiography, as guided by pretest probability and the D-dimer level, is the imaging test of choice in the evaluation of pulmonary embolism. It can also offer insight into thrombotic burden and can reveal concurrent or alternative diagnoses (eg, pneumonia).

Ventilation-perfusion scanning

When CT angiography is unavailable or the patient should not be exposed to contrast medium (eg, due to concern for contrast-induced nephropathy or contrast allergy), ventilation-perfusion (V/Q) scanning remains an option for ruling out pulmonary embolism.²²

Anderson et al²³ compared CT angiography and V/Q scanning in a study in 1,417 patients considered likely to have acute pulmonary embolism. Rates of symptomatic pulmonary embolism during 3-month follow-up were similar in patients who initially had negative results on V/Q scanning compared with those who initially had negative results on CT angiography. However, this study used single-detector CT scanners for one-third of the patients. Therefore, the results may have been different if current technology had been used.

Limitations of V/Q scanning include length of time to perform (30–45 minutes), cost, inability to identify other causes of symptoms, and difficulty with interpretation  when other pulmonary pathology is present (eg, lung infiltrate). V/Q scanning is helpful when negative but is often reported based on probability (low, intermediate, or high) and may not provide adequate guidance. Therefore, CT angiography should be used whenever possible for diagnosing pulmonary embolism.

Other tests for pulmonary embolism

Electrocardiography, transthoracic echocardiography, and chest radiography may aid in the search for alternative diagnoses and assess the degree of right heart strain as a sequela of pulmonary embolism, but they do not confirm the diagnosis.

ORDER IMAGING ONLY IF NEEDED

Diagnostic imaging can be optimized by avoiding unnecessary tests that carry both costs and clinical risks.

Most patients in whom acute pulmonary embolism is discovered will not need testing for deep vein thrombosis, as they will receive anticoagulation regardless. Similarly, many patients with acute symptomatic deep vein thrombosis do not need testing for pulmonary embolism with chest CT imaging, as they too will receive anticoagulation regardless.

Therefore, clinicians are encouraged to use diagnostic reasoning while practicing high-value care (including estimating pretest probability and measuring D-dimer when appropriate), ordering additional tests judiciously and only if indicated.

THROMBOEMBOLISM IS CONFIRMED—IS FURTHER TESTING WARRANTED?

Once acute venous thromboembolism is confirmed, key considerations include whether the event was provoked or unprovoked (ie, idiopathic) and whether the patient needs indefinite anticoagulation (eg, after 2 or more unprovoked events).

Was the event provoked or unprovoked?

Even in cases of unprovoked venous thromboembolism, no clear consensus exists as to which patients should be tested for thrombophilia. Experts do advocate, however, that it be done only in highly selected patients and that it be coordinated with the patient, family members, and an expert in this testing. Patients for whom further testing may be considered include those with venous thromboembolism in unusual sites (eg, the cavernous sinus), with warfarin-induced skin necrosis, or with recurrent pregnancy loss.

While screening for malignancy may seem prudent in the case of unexplained venous thromboembolism, the use of CT imaging for this purpose has been found to be of low yield. In one study,²⁴ it was not found to detect additional neoplasms, and it can lead to additional cost and no added benefit for patients.

The American Board of Internal Medicine’s Choosing Wisely campaign strongly recommends consultation with an expert in thrombophilia (eg, a hematologist) before testing.²⁵ Ordering multiple tests in bundles (hypercoagulability panels) is unlikely to alter management, could have a negative clinical impact on patients, and is generally not recommended.

The ‘4 Ps’ approach to testing

• Patient selection
• Pretest counseling
• Proper laboratory interpretation
• Provision of education and advice.

Importantly, testing should be reserved for patients in whom the pretest probability of the thrombophilic disease is moderate to high, such as testing for antiphospholipid antibody syndrome in patients with systemic lupus erythematosus or recurrent miscarriage.

Venous thromboembolism in a patient who is known to have a malignant disease does not typically warrant further thrombophilia testing, as the event was likely a sequela of the malignancy. The evaluation and management of venous thromboembolism with concurrent neoplasm is covered elsewhere.²¹

Thrombophilia testing should be approached the same regardless of whether the venous thromboembolism was diagnosed intentionally or incidentally. First, determine whether the thrombosis was provoked or unprovoked, then order additional tests only if indicated, as recommended. Alternative approaches such as forgoing anticoagulation (but performing serial imaging, if indicated) may be reasonable if the thrombus is deemed clinically irrelevant (eg, nonocclusive, asymptomatic, subsegmental pulmonary embolism in the absence of proximal deep vein thrombosis; isolated distal deep vein thrombosis).^25,27

It is still debatable whether the increasing incidence of asymptomatic pulmonary embolism due to enhanced sensitivity of noninvasive diagnostic imaging warrants a change in diagnostic approach.²⁸

FACTORS TO CONSIDER BEFORE THROMBOPHILIA TESTING

Important factors to consider before testing for thrombophilia are²⁹:

• How will the results affect the anticoagulation plan?
• How may the patient’s clinical status and medications influence the results?
• Has the patient expressed a desire to understand why venous thromboembolism occurred?
• Will the results have a potential impact on the patient’s family members?

How will the results of thrombophilia testing affect anticoagulation management?

Because the goal of any diagnostic test is to find out what type of care the patient needs, clinicians must determine whether knowledge of an underlying thrombophilia will alter the short-term or long-term anticoagulation therapy the patient is receiving for an acute venous thromboembolic event.

As most acute episodes of venous thromboembolism require an initial 3 months of anticoagulation (with the exception of some nonclinically relevant events such as isolated distal deep vein thrombosis without extension on reimaging), testing in the acute setting does not change the short-term management of anticoagulation. Many hospitals have advocated for outpatient-only thrombophilia testing (if testing does occur), as testing in the acute setting may render test results uninterpretable (see What factors can influence thrombophilia testing? below) and can inappropriately affect the long-term management of anticoagulation. We recommend against testing in the inpatient setting.

To determine the duration of anticoagulation, clinicians must balance the risk of recurrent venous thromboembolism and the risk of bleeding. If a patient is at significant risk of bleeding or does not tolerate anticoagulation, clinicians may consider stopping therapy instead of evaluating for thrombophilia. For patients with provoked venous thromboembolism, anticoagulation should generally be limited to 3 months, as the risk of recurrence does not outweigh the risk of bleeding with continued anticoagulation therapy.

Patients with unprovoked venous thromboembolism have a risk of recurrence twice as high as those with provoked venous thromboembolism and generally need a longer duration of anticoagulation.^30,31 Once a patient with an unprovoked venous thromboembolic event has completed the initial 3 months of anticoagulation, a formal risk-benefit evaluation should be performed to determine whether to continue it.

Up to 42% of patients with unprovoked venous thromboembolism may have 1 or more thrombotic disorders, and some clinicians believe that detecting an underlying thrombophilia will aid in decisions regarding duration of therapy.³² However, the risk of recurrent venous thromboembolism in these patients does not differ significantly from that in patients without an underlying thrombophilia.^33–35 As such, it has been suggested that the unprovoked character of the thrombotic event, rather than an underlying thrombophilia, determines the risk of future recurrence and should be used instead of testing to guide the duration of anticoagulation therapy.³²

For more information, see the 2016 ACCP guideline update on antithrombotic therapy for venous thromboembolism.²⁷

What factors can influence the results of thrombophilia testing?

For example, antithrombin is consumed during thrombus formation; therefore, antithrombin levels may be transiently suppressed in acute venous thromboembolism. Moreover, since antithrombin binds to unfractionated heparin, low-molecular-weight heparin, and fondaparinux and mediates their activity as anticoagulants, antithrombin levels may be decreased by heparin therapy.

Similarly, vitamin K antagonists (eg, warfarin) suppress protein C and S activity levels by inhibiting vitamin K epoxide reductase and may falsely indicate a protein C or S deficiency.

Direct oral anticoagulants can cause false-positive results on lupus anticoagulant assays (dilute Russell viper venom time, augmented partial thromboplastin time), raise protein C, protein S, and antithrombin activity levels, and normalize activated protein C resistance assays, leading to missed diagnoses.⁴¹

Since estrogen therapy and pregnancy lead to increases in C4b binding protein, resulting in decreased free protein S, these situations can result in clinicians falsely labeling patients as having congenital protein S deficiency when in fact the patient had a transient reduction in protein S levels.³³

Therefore, to optimize accuracy and interpretation of results, thrombophilia testing should ideally be performed when the patient:

• Is past the acute event and out of the hospital
• Is not pregnant
• Has received the required 3 months of anticoagulation and is off this therapy.

For warfarin, most recommendations say that testing should be performed after the patient has been off therapy for 2 to 6 weeks.⁴² Low-molecular-weight heparins and direct oral anticoagulants should be discontinued for at least 48 to 72 hours, or longer if the patient has kidney impairment, as these medications are renally eliminated.

Genetic tests such as factor V Leiden and prothrombin gene mutation are not affected by these factors and do not require repeat or confirmatory testing.

What if the patient or family wants to understand why an event occurred?

Some experts advocate thrombophilia testing of asymptomatic family members to identify carriers who may need prophylaxis against venous thromboembolism in high-risk situations such as pregnancy, oral contraceptive use, hospitalization, and surgery.²⁹ Asymptomatic family members of a first-degree relative with a history of venous thromboembolism have a 2 times higher risk of an index event.⁴³ Thus, it may be argued that these asymptomatic individuals should receive prophylactic measures in any high-risk situation, based on the family history itself rather than results of thrombophilia testing.

Occasionally, patients and family members want to know the cause of the thrombotic event and want to be tested. In these instances, pretest counseling for the patient and family about the potential implications of testing and shared decision-making between the provider and patient are of utmost importance.²⁹

What is the impact on family members if thrombophilia is diagnosed?

While positive test results can give patients some satisfaction, this knowledge may also cause unnecessary worry, as the patient knows he or she has a hematologic disorder and could possible die of venous thromboembolism.

Thrombophilia testing can have other adverse consequences. For example, while the Genetic Information Nondiscrimination Act of 2008 protects against denial of health insurance benefits based on genetic information, known carriers of thrombophilia may have trouble obtaining life or disability insurance.⁴⁴

Unfortunately, it is not uncommon for thrombophilia testing to be inappropriately performed, interpreted, or followed up. These suboptimal approaches can lead to unnecessary exposure to high-risk therapeutic anticoagulation, excessive durations of therapy, and labeling with an unconfirmed or incorrect diagnosis. Additionally, there are significant costs associated with thrombophilia testing, including the cost of the tests and anticoagulant medications and management of adverse events such as bleeding.

WHAT ARE THE ALTERNATIVES TO THROMBOPHILIA TESTING?

Because discovered thrombophilias (eg, factor V Leiden mutation, prothrombin gene mutation) have not consistently shown a strong correlation with increased recurrence of venous thromboembolism, alternative approaches are emerging to determine the duration of therapy for unprovoked events.

Clinical prediction tools based on patient characteristics and laboratory markers that are more consistently associated with recurrent venous thromboembolism (eg, male sex, persistently elevated D-dimer) have been developed to aid clinicians dealing with this challenging question. Several prediction tools are available:

The “Men Continue and HERDOO2” rule (HERDOO2 = hyperpigmentation, edema, or redness in either leg; D-dimer level ≥ 250 μg/L; obesity with body mass index ≥ 30 kg/m²; or older age, ≥ 65)⁴⁵

The DASH score (D-dimer, age, sex, and hormonal therapy)⁴⁶

The Vienna score,^47,48 at http://cemsiis.meduniwien.ac.at/en/kb/science-research/software/clinical-software/recurrent-vte/.

SUMMARY OF THROMBOPHILIA TESTING RECOMMENDATIONS

Test for thrombophilia only when…

• Discussing with a specialist (eg, hematologist) who has an understanding of thrombophilia
• Using the 4 Ps approach
• A patient requests testing to understand why a thrombotic event occurred, and the patient understands the implications of testing (ie, received counseling) for self and for family
• An expert deems identification of asymptomatic family members important for those who may be carriers of a detected thrombophilia
• The patient with a venous thromboembolic event has completed 3 months of anticoagulation and has been off anticoagulation for the appropriate length of time
• The results will change management.

Forgo thrombophilia testing when…

• A patient has a provoked venous thromboembolic event
• You do not intend to discontinue anticoagulation (ie, anticoagulation is indefinite)
• The patient is in the acute (eg, inpatient) setting
• The patient is on anticoagulants that may render test results uninterpretable
• The patient is pregnant or on oral contraceptives
• Use of alternative patient characteristics and laboratory markers to predict venous thromboembolism recurrence may be an option.

OPTIMIZING THE DIAGNOSIS

With the incidence of venous thromboembolism rapidly increasing, optimizing its diagnosis from both a financial and clinical perspective is becoming increasingly important. Clinicians should be familiar with the use of pretest probability scoring for venous thromboembolism, as well as which diagnostic tests are preferred if further workup is indicated. They should strive to minimize or avoid indiscriminate thrombophilia testing, which may lead to increased healthcare costs and patient exposure to potentially harmful anticoagulation.

Testing for thrombophilia should be based on whether a venous thromboembolic event was provoked or unprovoked. Patients with provoked venous thromboembolism or those receiving indefinite anticoagulation therapy should not be tested for thrombophilia. If testing is being considered in a patient with unprovoked venous thromboembolism, a specialist who is able to implement the 4 Ps approach should be consulted to ensure well-informed, shared decision-making with patients and family members.

When a patient presents with suspected venous thromboembolism, ie, deep vein thrombosis or pulmonary embolism, what diagnostic tests are needed to confirm the diagnosis? The clinical signs and symptoms of venous thromboembolism are nonspecific and often difficult to interpret. Therefore, it is essential for clinicians to use a standardized, structured approach to diagnosis that incorporates clinical findings and laboratory testing, as well as judicious use of diagnostic imaging. But while information is important, clinicians must also strive to avoid unnecessary testing, not only to decrease costs, but also to avoid potential harm.

If the diagnosis is confirmed, does the patient need testing for an underlying thrombophilic disorder? Such screening is often considered after a thromboembolic event occurs. However, a growing body of evidence indicates that the results of thrombophilia testing can be misinterpreted and potentially harmful.¹ We need to understand the utility of this testing as well as when and how it should be used. Patients and thrombosis specialists should be involved in deciding whether to perform these tests.

In this article, we provide practical information about how to diagnose venous thromboembolism, including strategies to optimize testing in suspected cases. We also offer guidance on how to decide whether further thrombophilia testing is warranted.

COMMON AND SERIOUS

Venous thromboembolism is a major cause of morbidity and death. Approximately 900,000 cases of pulmonary embolism and deep vein thrombosis occur in the United States each year, causing 60,000 to 300,000 deaths,² with the number of cases projected to double over the next 40 years.³

INITIAL APPROACH: PRETEST PROBABILITY

Given the morbidity and mortality associated with venous thromboembolism, prompt recognition and diagnosis are imperative. Clinical diagnosis alone is insufficient, with confirmed disease found in only 15% to 25% of patients suspected of having venous thromboembolism.^4–8 Therefore, the pretest probability should be coupled with objective testing.

The Wells score shows good discrimination in the outpatient and emergency department settings, but it has been invalidated in the inpatient setting, and thus it should not be used in inpatients.¹⁰

LABORATORY TESTS FOR SUSPECTED VENOUS THROMBOEMBOLISM

Employing an understanding of diagnostic testing is fundamental to identifying patients with venous thromboembolism.

D-dimer is a byproduct of fibrinolysis.

D-dimer testing has very high sensitivity for venous thromboembolism (> 90%) but low specificity (about 50%), and levels can be elevated in a variety of situations such as advanced age, acute inflammation, and cancer.¹⁵ The standard threshold is 500 μg/L, but because the D-dimer level increases with age, some clinicians advocate using an age-adjusted threshold for patients age 50 or older (age in years × 10 μg/L) to increase the diagnostic yield.¹⁶

Of the laboratory tests for D-dimer, the enzyme-linked immunosorbent assay has the highest sensitivity and highest negative predictive value (100%) and may be preferred over the other test methodologies.¹⁷

With its high sensitivity, D-dimer testing is clinically useful for ruling out venous thromboembolism, particularly when the pretest probability is low, but it lacks the specificity required for diagnosing and treating the disease if positive. Thus, it is not useful for ruling in venous thromboembolism. If the patient has a high pretest probability, we can omit D-dimer testing in favor of imaging studies.

Other laboratory tests such as arterial blood gas and brain natriuretic peptide levels have been proposed as markers of pulmonary embolism, but studies suggest they have limited utility in predicting the presence of disease.^18,19

DIAGNOSTIC TESTS FOR DEEP VEIN THROMBOSIS

Ultrasonography

If the pretest probability of deep vein thrombosis is high or a D-dimer test is found to be positive, the next step in evaluation is compression ultrasonography. 

While some guidelines recommend scanning only the proximal leg, many facilities in the United States scan the whole leg, which may reveal distal deep vein thrombosis.²⁰ The clinical significance of isolated distal deep vein thrombosis is unknown, and a selective anticoagulation approach may be used if this condition is discovered. The 2012 and 2016 American College of Chest Physicians (ACCP) guidelines on diagnosis and management of venous thromboembolism address this topic.^20,21

Deep vein thrombosis in the arm should be evaluated in the same manner as in the lower extremities.

Venography

Invasive and therefore no longer often used, venography is considered the gold standard for diagnosing deep vein thrombosis. Computed tomographic (CT) or magnetic resonance (MR) venography is most useful if the patient has aberrant anatomy such as a deformity of the leg, or in situations where the use of ultrasonography is difficult or unreliable, such as in the setting of severe obesity. CT or MR venography may be considered when looking for thrombosis in noncompressible veins of the thorax and abdomen (eg, the subclavian vein, iliac vein, and inferior vena cava) if ultrasonography is negative but clinical suspicion is high. Venous-phase CT angiography is particularly useful in diagnosing deep vein thrombosis in the inferior vena cava and iliac vein when deep vein thrombosis is clinically suspected but cannot be visualized on duplex ultrasonography.

Computed tomography

Imaging is warranted in patients who have a high pretest probability of pulmonary embolism, or in whom the D-dimer assay was positive but the pretest probability was low or moderate.

Once the gold standard, pulmonary angiography is no longer recommended for the initial diagnosis of pulmonary embolism because it is invasive, often unavailable, less sophisticated, and more expensive than noninvasive imaging techniques such as CT angiography. It is still used, however, in catheter-directed thrombolysis.

Thus, multiphasic CT angiography, as guided by pretest probability and the D-dimer level, is the imaging test of choice in the evaluation of pulmonary embolism. It can also offer insight into thrombotic burden and can reveal concurrent or alternative diagnoses (eg, pneumonia).

Ventilation-perfusion scanning

When CT angiography is unavailable or the patient should not be exposed to contrast medium (eg, due to concern for contrast-induced nephropathy or contrast allergy), ventilation-perfusion (V/Q) scanning remains an option for ruling out pulmonary embolism.²²

Anderson et al²³ compared CT angiography and V/Q scanning in a study in 1,417 patients considered likely to have acute pulmonary embolism. Rates of symptomatic pulmonary embolism during 3-month follow-up were similar in patients who initially had negative results on V/Q scanning compared with those who initially had negative results on CT angiography. However, this study used single-detector CT scanners for one-third of the patients. Therefore, the results may have been different if current technology had been used.

Limitations of V/Q scanning include length of time to perform (30–45 minutes), cost, inability to identify other causes of symptoms, and difficulty with interpretation  when other pulmonary pathology is present (eg, lung infiltrate). V/Q scanning is helpful when negative but is often reported based on probability (low, intermediate, or high) and may not provide adequate guidance. Therefore, CT angiography should be used whenever possible for diagnosing pulmonary embolism.

Other tests for pulmonary embolism

Electrocardiography, transthoracic echocardiography, and chest radiography may aid in the search for alternative diagnoses and assess the degree of right heart strain as a sequela of pulmonary embolism, but they do not confirm the diagnosis.

ORDER IMAGING ONLY IF NEEDED

Diagnostic imaging can be optimized by avoiding unnecessary tests that carry both costs and clinical risks.

Most patients in whom acute pulmonary embolism is discovered will not need testing for deep vein thrombosis, as they will receive anticoagulation regardless. Similarly, many patients with acute symptomatic deep vein thrombosis do not need testing for pulmonary embolism with chest CT imaging, as they too will receive anticoagulation regardless.

Therefore, clinicians are encouraged to use diagnostic reasoning while practicing high-value care (including estimating pretest probability and measuring D-dimer when appropriate), ordering additional tests judiciously and only if indicated.

THROMBOEMBOLISM IS CONFIRMED—IS FURTHER TESTING WARRANTED?

Once acute venous thromboembolism is confirmed, key considerations include whether the event was provoked or unprovoked (ie, idiopathic) and whether the patient needs indefinite anticoagulation (eg, after 2 or more unprovoked events).

Was the event provoked or unprovoked?

Even in cases of unprovoked venous thromboembolism, no clear consensus exists as to which patients should be tested for thrombophilia. Experts do advocate, however, that it be done only in highly selected patients and that it be coordinated with the patient, family members, and an expert in this testing. Patients for whom further testing may be considered include those with venous thromboembolism in unusual sites (eg, the cavernous sinus), with warfarin-induced skin necrosis, or with recurrent pregnancy loss.

While screening for malignancy may seem prudent in the case of unexplained venous thromboembolism, the use of CT imaging for this purpose has been found to be of low yield. In one study,²⁴ it was not found to detect additional neoplasms, and it can lead to additional cost and no added benefit for patients.

The American Board of Internal Medicine’s Choosing Wisely campaign strongly recommends consultation with an expert in thrombophilia (eg, a hematologist) before testing.²⁵ Ordering multiple tests in bundles (hypercoagulability panels) is unlikely to alter management, could have a negative clinical impact on patients, and is generally not recommended.

The ‘4 Ps’ approach to testing

• Patient selection
• Pretest counseling
• Proper laboratory interpretation
• Provision of education and advice.

Importantly, testing should be reserved for patients in whom the pretest probability of the thrombophilic disease is moderate to high, such as testing for antiphospholipid antibody syndrome in patients with systemic lupus erythematosus or recurrent miscarriage.

Venous thromboembolism in a patient who is known to have a malignant disease does not typically warrant further thrombophilia testing, as the event was likely a sequela of the malignancy. The evaluation and management of venous thromboembolism with concurrent neoplasm is covered elsewhere.²¹

Thrombophilia testing should be approached the same regardless of whether the venous thromboembolism was diagnosed intentionally or incidentally. First, determine whether the thrombosis was provoked or unprovoked, then order additional tests only if indicated, as recommended. Alternative approaches such as forgoing anticoagulation (but performing serial imaging, if indicated) may be reasonable if the thrombus is deemed clinically irrelevant (eg, nonocclusive, asymptomatic, subsegmental pulmonary embolism in the absence of proximal deep vein thrombosis; isolated distal deep vein thrombosis).^25,27

It is still debatable whether the increasing incidence of asymptomatic pulmonary embolism due to enhanced sensitivity of noninvasive diagnostic imaging warrants a change in diagnostic approach.²⁸

FACTORS TO CONSIDER BEFORE THROMBOPHILIA TESTING

Important factors to consider before testing for thrombophilia are²⁹:

• How will the results affect the anticoagulation plan?
• How may the patient’s clinical status and medications influence the results?
• Has the patient expressed a desire to understand why venous thromboembolism occurred?
• Will the results have a potential impact on the patient’s family members?

How will the results of thrombophilia testing affect anticoagulation management?

Because the goal of any diagnostic test is to find out what type of care the patient needs, clinicians must determine whether knowledge of an underlying thrombophilia will alter the short-term or long-term anticoagulation therapy the patient is receiving for an acute venous thromboembolic event.

As most acute episodes of venous thromboembolism require an initial 3 months of anticoagulation (with the exception of some nonclinically relevant events such as isolated distal deep vein thrombosis without extension on reimaging), testing in the acute setting does not change the short-term management of anticoagulation. Many hospitals have advocated for outpatient-only thrombophilia testing (if testing does occur), as testing in the acute setting may render test results uninterpretable (see What factors can influence thrombophilia testing? below) and can inappropriately affect the long-term management of anticoagulation. We recommend against testing in the inpatient setting.

To determine the duration of anticoagulation, clinicians must balance the risk of recurrent venous thromboembolism and the risk of bleeding. If a patient is at significant risk of bleeding or does not tolerate anticoagulation, clinicians may consider stopping therapy instead of evaluating for thrombophilia. For patients with provoked venous thromboembolism, anticoagulation should generally be limited to 3 months, as the risk of recurrence does not outweigh the risk of bleeding with continued anticoagulation therapy.

Patients with unprovoked venous thromboembolism have a risk of recurrence twice as high as those with provoked venous thromboembolism and generally need a longer duration of anticoagulation.^30,31 Once a patient with an unprovoked venous thromboembolic event has completed the initial 3 months of anticoagulation, a formal risk-benefit evaluation should be performed to determine whether to continue it.

Up to 42% of patients with unprovoked venous thromboembolism may have 1 or more thrombotic disorders, and some clinicians believe that detecting an underlying thrombophilia will aid in decisions regarding duration of therapy.³² However, the risk of recurrent venous thromboembolism in these patients does not differ significantly from that in patients without an underlying thrombophilia.^33–35 As such, it has been suggested that the unprovoked character of the thrombotic event, rather than an underlying thrombophilia, determines the risk of future recurrence and should be used instead of testing to guide the duration of anticoagulation therapy.³²

For more information, see the 2016 ACCP guideline update on antithrombotic therapy for venous thromboembolism.²⁷

What factors can influence the results of thrombophilia testing?

For example, antithrombin is consumed during thrombus formation; therefore, antithrombin levels may be transiently suppressed in acute venous thromboembolism. Moreover, since antithrombin binds to unfractionated heparin, low-molecular-weight heparin, and fondaparinux and mediates their activity as anticoagulants, antithrombin levels may be decreased by heparin therapy.

Similarly, vitamin K antagonists (eg, warfarin) suppress protein C and S activity levels by inhibiting vitamin K epoxide reductase and may falsely indicate a protein C or S deficiency.

Direct oral anticoagulants can cause false-positive results on lupus anticoagulant assays (dilute Russell viper venom time, augmented partial thromboplastin time), raise protein C, protein S, and antithrombin activity levels, and normalize activated protein C resistance assays, leading to missed diagnoses.⁴¹

Since estrogen therapy and pregnancy lead to increases in C4b binding protein, resulting in decreased free protein S, these situations can result in clinicians falsely labeling patients as having congenital protein S deficiency when in fact the patient had a transient reduction in protein S levels.³³

Therefore, to optimize accuracy and interpretation of results, thrombophilia testing should ideally be performed when the patient:

• Is past the acute event and out of the hospital
• Is not pregnant
• Has received the required 3 months of anticoagulation and is off this therapy.

For warfarin, most recommendations say that testing should be performed after the patient has been off therapy for 2 to 6 weeks.⁴² Low-molecular-weight heparins and direct oral anticoagulants should be discontinued for at least 48 to 72 hours, or longer if the patient has kidney impairment, as these medications are renally eliminated.

Genetic tests such as factor V Leiden and prothrombin gene mutation are not affected by these factors and do not require repeat or confirmatory testing.

What if the patient or family wants to understand why an event occurred?

Some experts advocate thrombophilia testing of asymptomatic family members to identify carriers who may need prophylaxis against venous thromboembolism in high-risk situations such as pregnancy, oral contraceptive use, hospitalization, and surgery.²⁹ Asymptomatic family members of a first-degree relative with a history of venous thromboembolism have a 2 times higher risk of an index event.⁴³ Thus, it may be argued that these asymptomatic individuals should receive prophylactic measures in any high-risk situation, based on the family history itself rather than results of thrombophilia testing.

Occasionally, patients and family members want to know the cause of the thrombotic event and want to be tested. In these instances, pretest counseling for the patient and family about the potential implications of testing and shared decision-making between the provider and patient are of utmost importance.²⁹

What is the impact on family members if thrombophilia is diagnosed?

While positive test results can give patients some satisfaction, this knowledge may also cause unnecessary worry, as the patient knows he or she has a hematologic disorder and could possible die of venous thromboembolism.

Thrombophilia testing can have other adverse consequences. For example, while the Genetic Information Nondiscrimination Act of 2008 protects against denial of health insurance benefits based on genetic information, known carriers of thrombophilia may have trouble obtaining life or disability insurance.⁴⁴

Unfortunately, it is not uncommon for thrombophilia testing to be inappropriately performed, interpreted, or followed up. These suboptimal approaches can lead to unnecessary exposure to high-risk therapeutic anticoagulation, excessive durations of therapy, and labeling with an unconfirmed or incorrect diagnosis. Additionally, there are significant costs associated with thrombophilia testing, including the cost of the tests and anticoagulant medications and management of adverse events such as bleeding.

WHAT ARE THE ALTERNATIVES TO THROMBOPHILIA TESTING?

Because discovered thrombophilias (eg, factor V Leiden mutation, prothrombin gene mutation) have not consistently shown a strong correlation with increased recurrence of venous thromboembolism, alternative approaches are emerging to determine the duration of therapy for unprovoked events.

Clinical prediction tools based on patient characteristics and laboratory markers that are more consistently associated with recurrent venous thromboembolism (eg, male sex, persistently elevated D-dimer) have been developed to aid clinicians dealing with this challenging question. Several prediction tools are available:

The “Men Continue and HERDOO2” rule (HERDOO2 = hyperpigmentation, edema, or redness in either leg; D-dimer level ≥ 250 μg/L; obesity with body mass index ≥ 30 kg/m²; or older age, ≥ 65)⁴⁵

The DASH score (D-dimer, age, sex, and hormonal therapy)⁴⁶

The Vienna score,^47,48 at http://cemsiis.meduniwien.ac.at/en/kb/science-research/software/clinical-software/recurrent-vte/.

SUMMARY OF THROMBOPHILIA TESTING RECOMMENDATIONS

Test for thrombophilia only when…

• Discussing with a specialist (eg, hematologist) who has an understanding of thrombophilia
• Using the 4 Ps approach
• A patient requests testing to understand why a thrombotic event occurred, and the patient understands the implications of testing (ie, received counseling) for self and for family
• An expert deems identification of asymptomatic family members important for those who may be carriers of a detected thrombophilia
• The patient with a venous thromboembolic event has completed 3 months of anticoagulation and has been off anticoagulation for the appropriate length of time
• The results will change management.

Forgo thrombophilia testing when…

• A patient has a provoked venous thromboembolic event
• You do not intend to discontinue anticoagulation (ie, anticoagulation is indefinite)
• The patient is in the acute (eg, inpatient) setting
• The patient is on anticoagulants that may render test results uninterpretable
• The patient is pregnant or on oral contraceptives
• Use of alternative patient characteristics and laboratory markers to predict venous thromboembolism recurrence may be an option.

OPTIMIZING THE DIAGNOSIS

With the incidence of venous thromboembolism rapidly increasing, optimizing its diagnosis from both a financial and clinical perspective is becoming increasingly important. Clinicians should be familiar with the use of pretest probability scoring for venous thromboembolism, as well as which diagnostic tests are preferred if further workup is indicated. They should strive to minimize or avoid indiscriminate thrombophilia testing, which may lead to increased healthcare costs and patient exposure to potentially harmful anticoagulation.

Testing for thrombophilia should be based on whether a venous thromboembolic event was provoked or unprovoked. Patients with provoked venous thromboembolism or those receiving indefinite anticoagulation therapy should not be tested for thrombophilia. If testing is being considered in a patient with unprovoked venous thromboembolism, a specialist who is able to implement the 4 Ps approach should be consulted to ensure well-informed, shared decision-making with patients and family members.

The transition from hospital to home can be overwhelming for caregivers.¹ Stress of hospitalization coupled with the expectation of families to execute postdischarge care plans make understandable discharge communication critical. Communication failures, inadequate education, absence of caregiver confidence, and lack of clarity regarding care plans may prohibit smooth transitions and lead to adverse postdischarge outcomes.^2-4

Health literacy plays a pivotal role in caregivers’ capacity to navigate the healthcare system, comprehend, and execute care plans. An estimated 90 million Americans have limited health literacy that may negatively impact the provision of safe and quality care^5,6 and be a risk factor for poor outcomes, including increased emergency department (ED) utilization and readmission rates.^7-9 Readability strongly influences the effectiveness of written materials.¹⁰ However, written medical information for patients and families are frequently between the 10^th and 12^th grade reading levels; more than 75% of all pediatric health information is written at or above 10^th grade reading level.¹¹ Government agencies recommend between a 6^th and 8^th grade reading level, for written material;^5,12,13 written discharge instructions have been identified as an important quality metric for hospital-to-home transitions.^14-16

At our center, we found that discharge instructions were commonly written at high reading levels and often incomplete.¹⁷ Poor discharge instructions may contribute to increased readmission rates and unnecessary ED visits.^9,18 Our global aim targeted improved health-literate written information, including understandability and completeness.

Our specific aim was to increase the percentage of discharge instructions written at or below the 7th grade level for hospital medicine (HM) patients on a community hospital pediatric unit from 13% to 80% in 6 months.

Context

The improvement work took place at a 42-bed inpatient pediatric unit at a community satellite of our large, urban, academic hospital. The unit is staffed by medical providers including attendings, fellows, nurse practitioners (NPs), and senior pediatric residents, and had more than 1000 HM discharges in fiscal year 2016. Children with common general pediatric diagnoses are admitted to this service; postsurgical patients are not admitted primarily to the HM service. In Cincinnati, the neighborhood-level high school drop-out rates are as high as 64%.¹⁹ Discharge instructions are written by medical providers in the electronic health record (EHR). A printed copy is given to families and verbally reviewed by a bedside nurse prior to discharge. Quality improvement (QI) efforts focused on discharge instructions were ignited by a prior review of 200 discharge instructions that showed they were difficult to read (median reading level of 10^th grade), poorly understandable (36% of instructions met the threshold of understandability as measured by the Patient Education Materials Assessment Tool²⁰) and were missing key elements of information.¹⁷

Improvement Team

The improvement team consisted of 4 pediatric hospitalists, 2 NPs, 1 nurse educator with health literacy expertise, 1 pediatric resident, 1 fourth-year medical student, 1 QI consultant, and 2 parents who had first-hand experience on the HM service. The improvement team observed the discharge process, including roles of the provider, nurse and family, outlined a process map, and created a modified failure mode and effect analysis.²¹ Prior to our work, discharge instructions written by providers often occurred as a last step, and the content was created as free text or from nonstandardized templates. Key drivers that informed interventions were determined and revised over time (Figure 1). The study was reviewed by our institutional review board and deemed not human subjects research.

Improvement Activities

Key drivers were identified, and interventions were executed using Plan-Do Study-Act cycles.²² The key drivers thought to be critical for the success of the QI efforts were family engagement; standardization of discharge instructions; medical staff engagement; and audit and feedback of data. The corresponding interventions were as follows:

Family Engagement

Understanding the discharge information families desired. Prior to testing, 10 families admitted to the HM service were asked about the discharge experience. We asked families about information they wanted in written discharge instructions: 1) reasons to call your primary doctor or return to the hospital; 2) when to see your primary doctor for a follow-up visit; 3) the phone number to reach your child’s doctor; 4) more information about why your child was admitted; 5) information about new medications; and 6) what to do to help your child continue to recover at home.

Development of templates. We engaged families throughout the process of creating general and disease-specific discharge templates. After a specific template was created and reviewed by the parents on our team, it was sent to members of the institutional Patient Education Committee, which includes parents and local health literacy experts, to review and critique. Feedback from the reviewers was incorporated into the templates prior to use in the EHR.

Postdischarge phone calls.A convenience sample of families discharged from the satellite campus was called 24 to 48 hours after discharge over a 2-week period in January, 2016. A member of our improvement team solicited feedback from families about the quality of the discharge instructions. Families were asked if discharge instructions were reviewed with them prior to going home, if they were given a copy of the instructions, how they would rate the ability to read and use the information, and if there were additional pieces of information that would have improved the instructions.

Standardization of Instructions

Education. A presentation was created and shared with medical providers; it was re-disseminated monthly to new residents rotating onto the service and to the attendings, fellows, and NPs scheduled for shifts during the month. This education continued for the duration of the study. The presentation included the definition of health literacy, scope of the problem, examples of poorly written discharge instructions, and tips on how to write readable and understandable instructions. Laminated cards that included tips on how to write instructions were also placed on work stations.

While the general template was an important first step, the content relied heavily on free text by providers, which could still lead to instructions written at a high reading level. Thus, disease-specific discharge instruction templates were created with prepopulated information that was written at a reading level at or below 7^th grade level (Figure 2). The diseases were prioritized based on the most common diagnoses on our HM service. Each template included information under each of the subheadings noted in the general template. Twelve disease-specific templates were tested and ultimately embedded in the EHR; the general template remained for use when the discharge diagnosis was not covered by a disease-specific template.

Medical Staff Engagement

Previously described tests of change also aimed to enhance staff engagement. These included frequent e-mails, discussion of the QI efforts at specific team meetings, and the creation of visual cues posted at computer work stations, which prompted staff to begin to work on discharge instructions soon after admission.

Audit and Feedback of Data

Weekly phone calls. One team updated clinicians through a regularly scheduled bi-weekly phone conference. The phone conference was established prior to our work and was designed to relay pertinent information to attendings and NPs who work at the satellite hospital. During the phone conferences, clinicians were notified of current performance on discharge instruction readability and specific tests of change for the week. Additionally, providers gave feedback about the improvement efforts. These updates continued for the first 6 months of the project until sustained improvements were observed.

E-mails. Weekly e-mails were sent to all providers scheduled for clinical time at the satellite campus. The e-mail contained information on current tests of change, a list of discharge instruction templates that were available in the EHR, and the annotated run chart illustrating readability levels over time.

Additionally, individual e-mails were sent to each provider after review of the written discharge instructions for the week. Providers were given information on the number of discharge instructions they personally composed, the percentage of those instructions that were written at or below 7^th grade level, and specific feedback on how their written instructions could be improved. We also encouraged feedback from each provider to better identify barriers to achieving our goal.

Study of the Interventions

Baseline data included a review of all instructions for patients discharged from the satellite campus from the end of April 2015 through mid-September 2015. The time period for testing of interventions during the fall and winter months allowed for rapid cycle learning due to higher patient census and predictability of admissions for specific diagnosis (ie, asthma and bronchiolitis). An automated report was generated from the EHR weekly with specific demographics and identifiers for patient discharged over the past 7 days, including patient age, gender, length of stay, discharge diagnosis, and insurance classification. Data was collected during the intervention period via structured review of the discharge instructions in the EHR by the principal investigator or a trained research coordinator. Discharge instructions for medically cleared mental health patients admitted to hospital medicine while awaiting psychiatric bed availability and patients and parents who were non-English speaking were excluded from review. All other instructions for patients discharged from the HM service at our Liberty Campus were included for review.

Measures

Readability, our primary measure of interest, was calculated using the mean score from the following formulas: Flesch Kincaid Grade Level,²³ Simple Measure of Gobbledygook Index,²⁴ Coleman-Liau Index,²⁵ Gunning-Fog Index,²⁶ and Automated Readability Index²⁷ by means of an online platform (https://readability-score.com).²⁸ This platform was chosen because it incorporated a variety of formulas, was user-friendly, and required minimal data cleaning. Each of the readability formulas have been used to assesses readability of health information given to patients and families.^29,30 The threshold of 7th grade is in alignment with our institutional policy for educational materials and with recommendations from several government agencies.^5,12

Analysis

A statistical process control p-chart was used to analyze our primary measure of readability, dichotomized as percent discharge instructions written at or below 7th grade level. Run charts were used to follow mean reading level of discharge instructions and our process measure of percent of discharge instruction written with a general or disease-specific standardized template. Run chart and control chart rules for identifying special cause were used for midline shifts.³¹

The Table includes the demographic and clinical information of patients included in our analyses. Through sequential interventions, the percentage of discharge instructions written at or below 7th grade readability level increased from a mean of 13% to more than 80% in 3 months (Figure 3). Furthermore, the mean was sustained above 90% for 10 months and at 98% for the last 4 months. The use of 1 of the 13 EHR templates increased from 0% to 96% and was associated with the largest impact on the overall improvements (Supplemental Figure 1). Additionally, the average reading level of the discharge instructions decreased from 10th grade to 6th grade level (Supplemental Figure 2).

Qualitative comments from providers about the discharge instructions included:

“Are these [discharge instructions] available at base??  Great resource for interns.”
“These [discharge] instructions make the [discharge] process so easy!!! Love these...”
“Also feel like they have helped my discharge teaching in the room!”

Qualitative comments from families postdischarge included:
“I thought the instructions were very clear and easy to read. I especially thought that highlighting the important areas really helped.”
“I think this form looks great, and I really like the idea of having your child’s name on it.”

Through sequential Plan-Do Study-Act cycles, we increased the percentage of discharge instructions written at or below 7^th grade reading level from 13% to 98%. Our most impactful intervention was the creation and dissemination of standardized disease-specific discharge instruction templates. Our findings complement evidence in the adult and pediatric literature that the use of standardized, disease-specific discharge instruction templates may improve readability of instructions.^32,33 And, while quality improvement efforts have been employed to improve the discharge process for patients,^34-36 this is the first study in the inpatient setting that, to our knowledge, specifically addresses discharge instructions using quality improvement methods.

Our work targeted the critical intersection between individual health literacy, an individual’s capacity to acquire, interpret, and use health information, and the necessary changes needed within our healthcare system to ensure that appropriately written instructions are given to patients and families.^17,37 Our efforts focused on improving discharge instructions answer the call to consider health literacy a modifiable clinical risk factor.³⁷ Furthermore, we address the 6 aims for quality healthcare delivery: 1) safe, timely, efficient and equitable delivery of care through the creation and dissemination of standardized instructions that are written at the appropriate reading level for families to ease hospital-to-home transitions and streamline the workflow of medical providers; 2) effective education of medical providers on health literacy concepts; and 3) family-centeredness through the involvement of families in our QI efforts. While previous QI efforts to improve hospital-to-home transitions have focused on medication reconciliation, communication with primary care physicians, follow-up appointments, and timely discharges of patients, none have specifically focused on the quality of discharge instructions.^34-36

Most physicians do not receive education about how to write information that is readable and understandable; more than half of providers desired more education in this area.³⁸ Furthermore, pediatric providers may overestimate parental health literacy levels,³⁹ which may contribute to variability in the readability of written health materials. While education alone can contribute to a provider’s ability to create readable instructions, we note the improvement after the introduction of disease templates to demonstrate the importance of workflow-integrated higher reliability interventions to sustain improvements.

Our baseline poor readability rates were due to limited knowledge by frontline providers composing the instructions and a system in which an important element for successful hospital-to-home transitions was not tackled until patients were ready for discharge. Streamlining of the discharge process, including the creation of discharge instructions, may lead to improved efficiency, fewer discrepancies, more effective communication, and an enhanced family experience. Moreover, the success of our improvement work was due to key stakeholders, including parents, being a part of the team and the notable buy-in from providers.

Our work was not without limitations. We excluded non-English speaking families from the study. We were unable to measure reading level of our population directly and instead based our goals on national estimates. Our primary measure was readability, which is only 1 piece contributing to quality discharge instructions. Understandability and actionability are also important considerations; ^17,20,29,40 however, improvements in these areas were limited by our design options within the EHR. Our efforts focused on children with common general pediatric diagnoses, and it is unclear how our interventions would generalize to medically complex patients with more volume of information to communicate at discharge and with uncommon diagnoses that are less readily incorporated into standardized templates. Relatedly, our work occurred at the satellite campus of our tertiary care center and may not represent generalizable material or methods to implement templates at our main campus location or at other hospitals. To begin to better understand this, we have spread to HM patients at our main campus, including medically complex patients with technology dependence and/or neurological impairments. Standardized, disease-specific templates most relevant to this population as well as several patient specific templates, for those with frequent readmissions due to medical complexity, have been created and are actively being tested.

In conclusion, in using interventions targeted at standardization of discharge instructions and timely feedback to staff, we saw rapid, dramatic, and sustained improvement in the readability of discharge instructions. Next steps include adaptation and spread to other patient populations and care teams, collaborations with other centers, and assessing the impact of effectively written discharge instructions on patient outcomes, such as adverse drug events, readmission rates, and family experience.

Disclosure

No external funding was secured for this study. Dr. Brady is supported by a Patient-Centered Outcomes Research Mentored Clinical Investigator Award from the Agency for Healthcare Research and Quality, Award Number K08HS023827. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding organizations. The funding organization had no role in the design, preparation, review, or approval of this paper; nor the decision to submit the manuscript for publication. The authors have no financial relationships relevant to this article to disclose.

The transition from hospital to home can be overwhelming for caregivers.¹ Stress of hospitalization coupled with the expectation of families to execute postdischarge care plans make understandable discharge communication critical. Communication failures, inadequate education, absence of caregiver confidence, and lack of clarity regarding care plans may prohibit smooth transitions and lead to adverse postdischarge outcomes.^2-4

Health literacy plays a pivotal role in caregivers’ capacity to navigate the healthcare system, comprehend, and execute care plans. An estimated 90 million Americans have limited health literacy that may negatively impact the provision of safe and quality care^5,6 and be a risk factor for poor outcomes, including increased emergency department (ED) utilization and readmission rates.^7-9 Readability strongly influences the effectiveness of written materials.¹⁰ However, written medical information for patients and families are frequently between the 10^th and 12^th grade reading levels; more than 75% of all pediatric health information is written at or above 10^th grade reading level.¹¹ Government agencies recommend between a 6^th and 8^th grade reading level, for written material;^5,12,13 written discharge instructions have been identified as an important quality metric for hospital-to-home transitions.^14-16

At our center, we found that discharge instructions were commonly written at high reading levels and often incomplete.¹⁷ Poor discharge instructions may contribute to increased readmission rates and unnecessary ED visits.^9,18 Our global aim targeted improved health-literate written information, including understandability and completeness.

Our specific aim was to increase the percentage of discharge instructions written at or below the 7th grade level for hospital medicine (HM) patients on a community hospital pediatric unit from 13% to 80% in 6 months.

Context

The improvement work took place at a 42-bed inpatient pediatric unit at a community satellite of our large, urban, academic hospital. The unit is staffed by medical providers including attendings, fellows, nurse practitioners (NPs), and senior pediatric residents, and had more than 1000 HM discharges in fiscal year 2016. Children with common general pediatric diagnoses are admitted to this service; postsurgical patients are not admitted primarily to the HM service. In Cincinnati, the neighborhood-level high school drop-out rates are as high as 64%.¹⁹ Discharge instructions are written by medical providers in the electronic health record (EHR). A printed copy is given to families and verbally reviewed by a bedside nurse prior to discharge. Quality improvement (QI) efforts focused on discharge instructions were ignited by a prior review of 200 discharge instructions that showed they were difficult to read (median reading level of 10^th grade), poorly understandable (36% of instructions met the threshold of understandability as measured by the Patient Education Materials Assessment Tool²⁰) and were missing key elements of information.¹⁷

Improvement Team

The improvement team consisted of 4 pediatric hospitalists, 2 NPs, 1 nurse educator with health literacy expertise, 1 pediatric resident, 1 fourth-year medical student, 1 QI consultant, and 2 parents who had first-hand experience on the HM service. The improvement team observed the discharge process, including roles of the provider, nurse and family, outlined a process map, and created a modified failure mode and effect analysis.²¹ Prior to our work, discharge instructions written by providers often occurred as a last step, and the content was created as free text or from nonstandardized templates. Key drivers that informed interventions were determined and revised over time (Figure 1). The study was reviewed by our institutional review board and deemed not human subjects research.

Improvement Activities

Key drivers were identified, and interventions were executed using Plan-Do Study-Act cycles.²² The key drivers thought to be critical for the success of the QI efforts were family engagement; standardization of discharge instructions; medical staff engagement; and audit and feedback of data. The corresponding interventions were as follows:

Family Engagement

Understanding the discharge information families desired. Prior to testing, 10 families admitted to the HM service were asked about the discharge experience. We asked families about information they wanted in written discharge instructions: 1) reasons to call your primary doctor or return to the hospital; 2) when to see your primary doctor for a follow-up visit; 3) the phone number to reach your child’s doctor; 4) more information about why your child was admitted; 5) information about new medications; and 6) what to do to help your child continue to recover at home.

Development of templates. We engaged families throughout the process of creating general and disease-specific discharge templates. After a specific template was created and reviewed by the parents on our team, it was sent to members of the institutional Patient Education Committee, which includes parents and local health literacy experts, to review and critique. Feedback from the reviewers was incorporated into the templates prior to use in the EHR.

Postdischarge phone calls.A convenience sample of families discharged from the satellite campus was called 24 to 48 hours after discharge over a 2-week period in January, 2016. A member of our improvement team solicited feedback from families about the quality of the discharge instructions. Families were asked if discharge instructions were reviewed with them prior to going home, if they were given a copy of the instructions, how they would rate the ability to read and use the information, and if there were additional pieces of information that would have improved the instructions.

Standardization of Instructions

Education. A presentation was created and shared with medical providers; it was re-disseminated monthly to new residents rotating onto the service and to the attendings, fellows, and NPs scheduled for shifts during the month. This education continued for the duration of the study. The presentation included the definition of health literacy, scope of the problem, examples of poorly written discharge instructions, and tips on how to write readable and understandable instructions. Laminated cards that included tips on how to write instructions were also placed on work stations.

While the general template was an important first step, the content relied heavily on free text by providers, which could still lead to instructions written at a high reading level. Thus, disease-specific discharge instruction templates were created with prepopulated information that was written at a reading level at or below 7^th grade level (Figure 2). The diseases were prioritized based on the most common diagnoses on our HM service. Each template included information under each of the subheadings noted in the general template. Twelve disease-specific templates were tested and ultimately embedded in the EHR; the general template remained for use when the discharge diagnosis was not covered by a disease-specific template.

Medical Staff Engagement

Previously described tests of change also aimed to enhance staff engagement. These included frequent e-mails, discussion of the QI efforts at specific team meetings, and the creation of visual cues posted at computer work stations, which prompted staff to begin to work on discharge instructions soon after admission.

Audit and Feedback of Data

Weekly phone calls. One team updated clinicians through a regularly scheduled bi-weekly phone conference. The phone conference was established prior to our work and was designed to relay pertinent information to attendings and NPs who work at the satellite hospital. During the phone conferences, clinicians were notified of current performance on discharge instruction readability and specific tests of change for the week. Additionally, providers gave feedback about the improvement efforts. These updates continued for the first 6 months of the project until sustained improvements were observed.

E-mails. Weekly e-mails were sent to all providers scheduled for clinical time at the satellite campus. The e-mail contained information on current tests of change, a list of discharge instruction templates that were available in the EHR, and the annotated run chart illustrating readability levels over time.

Additionally, individual e-mails were sent to each provider after review of the written discharge instructions for the week. Providers were given information on the number of discharge instructions they personally composed, the percentage of those instructions that were written at or below 7^th grade level, and specific feedback on how their written instructions could be improved. We also encouraged feedback from each provider to better identify barriers to achieving our goal.

Study of the Interventions

Baseline data included a review of all instructions for patients discharged from the satellite campus from the end of April 2015 through mid-September 2015. The time period for testing of interventions during the fall and winter months allowed for rapid cycle learning due to higher patient census and predictability of admissions for specific diagnosis (ie, asthma and bronchiolitis). An automated report was generated from the EHR weekly with specific demographics and identifiers for patient discharged over the past 7 days, including patient age, gender, length of stay, discharge diagnosis, and insurance classification. Data was collected during the intervention period via structured review of the discharge instructions in the EHR by the principal investigator or a trained research coordinator. Discharge instructions for medically cleared mental health patients admitted to hospital medicine while awaiting psychiatric bed availability and patients and parents who were non-English speaking were excluded from review. All other instructions for patients discharged from the HM service at our Liberty Campus were included for review.

Measures

Readability, our primary measure of interest, was calculated using the mean score from the following formulas: Flesch Kincaid Grade Level,²³ Simple Measure of Gobbledygook Index,²⁴ Coleman-Liau Index,²⁵ Gunning-Fog Index,²⁶ and Automated Readability Index²⁷ by means of an online platform (https://readability-score.com).²⁸ This platform was chosen because it incorporated a variety of formulas, was user-friendly, and required minimal data cleaning. Each of the readability formulas have been used to assesses readability of health information given to patients and families.^29,30 The threshold of 7th grade is in alignment with our institutional policy for educational materials and with recommendations from several government agencies.^5,12

Analysis

A statistical process control p-chart was used to analyze our primary measure of readability, dichotomized as percent discharge instructions written at or below 7th grade level. Run charts were used to follow mean reading level of discharge instructions and our process measure of percent of discharge instruction written with a general or disease-specific standardized template. Run chart and control chart rules for identifying special cause were used for midline shifts.³¹

The Table includes the demographic and clinical information of patients included in our analyses. Through sequential interventions, the percentage of discharge instructions written at or below 7th grade readability level increased from a mean of 13% to more than 80% in 3 months (Figure 3). Furthermore, the mean was sustained above 90% for 10 months and at 98% for the last 4 months. The use of 1 of the 13 EHR templates increased from 0% to 96% and was associated with the largest impact on the overall improvements (Supplemental Figure 1). Additionally, the average reading level of the discharge instructions decreased from 10th grade to 6th grade level (Supplemental Figure 2).

Qualitative comments from providers about the discharge instructions included:

“Are these [discharge instructions] available at base??  Great resource for interns.”
“These [discharge] instructions make the [discharge] process so easy!!! Love these...”
“Also feel like they have helped my discharge teaching in the room!”

Qualitative comments from families postdischarge included:
“I thought the instructions were very clear and easy to read. I especially thought that highlighting the important areas really helped.”
“I think this form looks great, and I really like the idea of having your child’s name on it.”

Through sequential Plan-Do Study-Act cycles, we increased the percentage of discharge instructions written at or below 7^th grade reading level from 13% to 98%. Our most impactful intervention was the creation and dissemination of standardized disease-specific discharge instruction templates. Our findings complement evidence in the adult and pediatric literature that the use of standardized, disease-specific discharge instruction templates may improve readability of instructions.^32,33 And, while quality improvement efforts have been employed to improve the discharge process for patients,^34-36 this is the first study in the inpatient setting that, to our knowledge, specifically addresses discharge instructions using quality improvement methods.

Our work targeted the critical intersection between individual health literacy, an individual’s capacity to acquire, interpret, and use health information, and the necessary changes needed within our healthcare system to ensure that appropriately written instructions are given to patients and families.^17,37 Our efforts focused on improving discharge instructions answer the call to consider health literacy a modifiable clinical risk factor.³⁷ Furthermore, we address the 6 aims for quality healthcare delivery: 1) safe, timely, efficient and equitable delivery of care through the creation and dissemination of standardized instructions that are written at the appropriate reading level for families to ease hospital-to-home transitions and streamline the workflow of medical providers; 2) effective education of medical providers on health literacy concepts; and 3) family-centeredness through the involvement of families in our QI efforts. While previous QI efforts to improve hospital-to-home transitions have focused on medication reconciliation, communication with primary care physicians, follow-up appointments, and timely discharges of patients, none have specifically focused on the quality of discharge instructions.^34-36

Most physicians do not receive education about how to write information that is readable and understandable; more than half of providers desired more education in this area.³⁸ Furthermore, pediatric providers may overestimate parental health literacy levels,³⁹ which may contribute to variability in the readability of written health materials. While education alone can contribute to a provider’s ability to create readable instructions, we note the improvement after the introduction of disease templates to demonstrate the importance of workflow-integrated higher reliability interventions to sustain improvements.

Our baseline poor readability rates were due to limited knowledge by frontline providers composing the instructions and a system in which an important element for successful hospital-to-home transitions was not tackled until patients were ready for discharge. Streamlining of the discharge process, including the creation of discharge instructions, may lead to improved efficiency, fewer discrepancies, more effective communication, and an enhanced family experience. Moreover, the success of our improvement work was due to key stakeholders, including parents, being a part of the team and the notable buy-in from providers.

Our work was not without limitations. We excluded non-English speaking families from the study. We were unable to measure reading level of our population directly and instead based our goals on national estimates. Our primary measure was readability, which is only 1 piece contributing to quality discharge instructions. Understandability and actionability are also important considerations; ^17,20,29,40 however, improvements in these areas were limited by our design options within the EHR. Our efforts focused on children with common general pediatric diagnoses, and it is unclear how our interventions would generalize to medically complex patients with more volume of information to communicate at discharge and with uncommon diagnoses that are less readily incorporated into standardized templates. Relatedly, our work occurred at the satellite campus of our tertiary care center and may not represent generalizable material or methods to implement templates at our main campus location or at other hospitals. To begin to better understand this, we have spread to HM patients at our main campus, including medically complex patients with technology dependence and/or neurological impairments. Standardized, disease-specific templates most relevant to this population as well as several patient specific templates, for those with frequent readmissions due to medical complexity, have been created and are actively being tested.

In conclusion, in using interventions targeted at standardization of discharge instructions and timely feedback to staff, we saw rapid, dramatic, and sustained improvement in the readability of discharge instructions. Next steps include adaptation and spread to other patient populations and care teams, collaborations with other centers, and assessing the impact of effectively written discharge instructions on patient outcomes, such as adverse drug events, readmission rates, and family experience.

Disclosure

No external funding was secured for this study. Dr. Brady is supported by a Patient-Centered Outcomes Research Mentored Clinical Investigator Award from the Agency for Healthcare Research and Quality, Award Number K08HS023827. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding organizations. The funding organization had no role in the design, preparation, review, or approval of this paper; nor the decision to submit the manuscript for publication. The authors have no financial relationships relevant to this article to disclose.

The transition from hospital to home can be overwhelming for caregivers.¹ Stress of hospitalization coupled with the expectation of families to execute postdischarge care plans make understandable discharge communication critical. Communication failures, inadequate education, absence of caregiver confidence, and lack of clarity regarding care plans may prohibit smooth transitions and lead to adverse postdischarge outcomes.^2-4

Health literacy plays a pivotal role in caregivers’ capacity to navigate the healthcare system, comprehend, and execute care plans. An estimated 90 million Americans have limited health literacy that may negatively impact the provision of safe and quality care^5,6 and be a risk factor for poor outcomes, including increased emergency department (ED) utilization and readmission rates.^7-9 Readability strongly influences the effectiveness of written materials.¹⁰ However, written medical information for patients and families are frequently between the 10^th and 12^th grade reading levels; more than 75% of all pediatric health information is written at or above 10^th grade reading level.¹¹ Government agencies recommend between a 6^th and 8^th grade reading level, for written material;^5,12,13 written discharge instructions have been identified as an important quality metric for hospital-to-home transitions.^14-16

At our center, we found that discharge instructions were commonly written at high reading levels and often incomplete.¹⁷ Poor discharge instructions may contribute to increased readmission rates and unnecessary ED visits.^9,18 Our global aim targeted improved health-literate written information, including understandability and completeness.

Our specific aim was to increase the percentage of discharge instructions written at or below the 7th grade level for hospital medicine (HM) patients on a community hospital pediatric unit from 13% to 80% in 6 months.

Context

The improvement work took place at a 42-bed inpatient pediatric unit at a community satellite of our large, urban, academic hospital. The unit is staffed by medical providers including attendings, fellows, nurse practitioners (NPs), and senior pediatric residents, and had more than 1000 HM discharges in fiscal year 2016. Children with common general pediatric diagnoses are admitted to this service; postsurgical patients are not admitted primarily to the HM service. In Cincinnati, the neighborhood-level high school drop-out rates are as high as 64%.¹⁹ Discharge instructions are written by medical providers in the electronic health record (EHR). A printed copy is given to families and verbally reviewed by a bedside nurse prior to discharge. Quality improvement (QI) efforts focused on discharge instructions were ignited by a prior review of 200 discharge instructions that showed they were difficult to read (median reading level of 10^th grade), poorly understandable (36% of instructions met the threshold of understandability as measured by the Patient Education Materials Assessment Tool²⁰) and were missing key elements of information.¹⁷

Improvement Team

The improvement team consisted of 4 pediatric hospitalists, 2 NPs, 1 nurse educator with health literacy expertise, 1 pediatric resident, 1 fourth-year medical student, 1 QI consultant, and 2 parents who had first-hand experience on the HM service. The improvement team observed the discharge process, including roles of the provider, nurse and family, outlined a process map, and created a modified failure mode and effect analysis.²¹ Prior to our work, discharge instructions written by providers often occurred as a last step, and the content was created as free text or from nonstandardized templates. Key drivers that informed interventions were determined and revised over time (Figure 1). The study was reviewed by our institutional review board and deemed not human subjects research.

Improvement Activities

Key drivers were identified, and interventions were executed using Plan-Do Study-Act cycles.²² The key drivers thought to be critical for the success of the QI efforts were family engagement; standardization of discharge instructions; medical staff engagement; and audit and feedback of data. The corresponding interventions were as follows:

Family Engagement

Understanding the discharge information families desired. Prior to testing, 10 families admitted to the HM service were asked about the discharge experience. We asked families about information they wanted in written discharge instructions: 1) reasons to call your primary doctor or return to the hospital; 2) when to see your primary doctor for a follow-up visit; 3) the phone number to reach your child’s doctor; 4) more information about why your child was admitted; 5) information about new medications; and 6) what to do to help your child continue to recover at home.

Development of templates. We engaged families throughout the process of creating general and disease-specific discharge templates. After a specific template was created and reviewed by the parents on our team, it was sent to members of the institutional Patient Education Committee, which includes parents and local health literacy experts, to review and critique. Feedback from the reviewers was incorporated into the templates prior to use in the EHR.

Postdischarge phone calls.A convenience sample of families discharged from the satellite campus was called 24 to 48 hours after discharge over a 2-week period in January, 2016. A member of our improvement team solicited feedback from families about the quality of the discharge instructions. Families were asked if discharge instructions were reviewed with them prior to going home, if they were given a copy of the instructions, how they would rate the ability to read and use the information, and if there were additional pieces of information that would have improved the instructions.

Standardization of Instructions

Education. A presentation was created and shared with medical providers; it was re-disseminated monthly to new residents rotating onto the service and to the attendings, fellows, and NPs scheduled for shifts during the month. This education continued for the duration of the study. The presentation included the definition of health literacy, scope of the problem, examples of poorly written discharge instructions, and tips on how to write readable and understandable instructions. Laminated cards that included tips on how to write instructions were also placed on work stations.

While the general template was an important first step, the content relied heavily on free text by providers, which could still lead to instructions written at a high reading level. Thus, disease-specific discharge instruction templates were created with prepopulated information that was written at a reading level at or below 7^th grade level (Figure 2). The diseases were prioritized based on the most common diagnoses on our HM service. Each template included information under each of the subheadings noted in the general template. Twelve disease-specific templates were tested and ultimately embedded in the EHR; the general template remained for use when the discharge diagnosis was not covered by a disease-specific template.

Medical Staff Engagement

Previously described tests of change also aimed to enhance staff engagement. These included frequent e-mails, discussion of the QI efforts at specific team meetings, and the creation of visual cues posted at computer work stations, which prompted staff to begin to work on discharge instructions soon after admission.

Audit and Feedback of Data

Weekly phone calls. One team updated clinicians through a regularly scheduled bi-weekly phone conference. The phone conference was established prior to our work and was designed to relay pertinent information to attendings and NPs who work at the satellite hospital. During the phone conferences, clinicians were notified of current performance on discharge instruction readability and specific tests of change for the week. Additionally, providers gave feedback about the improvement efforts. These updates continued for the first 6 months of the project until sustained improvements were observed.

E-mails. Weekly e-mails were sent to all providers scheduled for clinical time at the satellite campus. The e-mail contained information on current tests of change, a list of discharge instruction templates that were available in the EHR, and the annotated run chart illustrating readability levels over time.

Additionally, individual e-mails were sent to each provider after review of the written discharge instructions for the week. Providers were given information on the number of discharge instructions they personally composed, the percentage of those instructions that were written at or below 7^th grade level, and specific feedback on how their written instructions could be improved. We also encouraged feedback from each provider to better identify barriers to achieving our goal.

Study of the Interventions

Baseline data included a review of all instructions for patients discharged from the satellite campus from the end of April 2015 through mid-September 2015. The time period for testing of interventions during the fall and winter months allowed for rapid cycle learning due to higher patient census and predictability of admissions for specific diagnosis (ie, asthma and bronchiolitis). An automated report was generated from the EHR weekly with specific demographics and identifiers for patient discharged over the past 7 days, including patient age, gender, length of stay, discharge diagnosis, and insurance classification. Data was collected during the intervention period via structured review of the discharge instructions in the EHR by the principal investigator or a trained research coordinator. Discharge instructions for medically cleared mental health patients admitted to hospital medicine while awaiting psychiatric bed availability and patients and parents who were non-English speaking were excluded from review. All other instructions for patients discharged from the HM service at our Liberty Campus were included for review.

Measures

Readability, our primary measure of interest, was calculated using the mean score from the following formulas: Flesch Kincaid Grade Level,²³ Simple Measure of Gobbledygook Index,²⁴ Coleman-Liau Index,²⁵ Gunning-Fog Index,²⁶ and Automated Readability Index²⁷ by means of an online platform (https://readability-score.com).²⁸ This platform was chosen because it incorporated a variety of formulas, was user-friendly, and required minimal data cleaning. Each of the readability formulas have been used to assesses readability of health information given to patients and families.^29,30 The threshold of 7th grade is in alignment with our institutional policy for educational materials and with recommendations from several government agencies.^5,12

Analysis

A statistical process control p-chart was used to analyze our primary measure of readability, dichotomized as percent discharge instructions written at or below 7th grade level. Run charts were used to follow mean reading level of discharge instructions and our process measure of percent of discharge instruction written with a general or disease-specific standardized template. Run chart and control chart rules for identifying special cause were used for midline shifts.³¹

The Table includes the demographic and clinical information of patients included in our analyses. Through sequential interventions, the percentage of discharge instructions written at or below 7th grade readability level increased from a mean of 13% to more than 80% in 3 months (Figure 3). Furthermore, the mean was sustained above 90% for 10 months and at 98% for the last 4 months. The use of 1 of the 13 EHR templates increased from 0% to 96% and was associated with the largest impact on the overall improvements (Supplemental Figure 1). Additionally, the average reading level of the discharge instructions decreased from 10th grade to 6th grade level (Supplemental Figure 2).

Qualitative comments from providers about the discharge instructions included:

“Are these [discharge instructions] available at base??  Great resource for interns.”
“These [discharge] instructions make the [discharge] process so easy!!! Love these...”
“Also feel like they have helped my discharge teaching in the room!”

Qualitative comments from families postdischarge included:
“I thought the instructions were very clear and easy to read. I especially thought that highlighting the important areas really helped.”
“I think this form looks great, and I really like the idea of having your child’s name on it.”

Through sequential Plan-Do Study-Act cycles, we increased the percentage of discharge instructions written at or below 7^th grade reading level from 13% to 98%. Our most impactful intervention was the creation and dissemination of standardized disease-specific discharge instruction templates. Our findings complement evidence in the adult and pediatric literature that the use of standardized, disease-specific discharge instruction templates may improve readability of instructions.^32,33 And, while quality improvement efforts have been employed to improve the discharge process for patients,^34-36 this is the first study in the inpatient setting that, to our knowledge, specifically addresses discharge instructions using quality improvement methods.

Our work targeted the critical intersection between individual health literacy, an individual’s capacity to acquire, interpret, and use health information, and the necessary changes needed within our healthcare system to ensure that appropriately written instructions are given to patients and families.^17,37 Our efforts focused on improving discharge instructions answer the call to consider health literacy a modifiable clinical risk factor.³⁷ Furthermore, we address the 6 aims for quality healthcare delivery: 1) safe, timely, efficient and equitable delivery of care through the creation and dissemination of standardized instructions that are written at the appropriate reading level for families to ease hospital-to-home transitions and streamline the workflow of medical providers; 2) effective education of medical providers on health literacy concepts; and 3) family-centeredness through the involvement of families in our QI efforts. While previous QI efforts to improve hospital-to-home transitions have focused on medication reconciliation, communication with primary care physicians, follow-up appointments, and timely discharges of patients, none have specifically focused on the quality of discharge instructions.^34-36

Most physicians do not receive education about how to write information that is readable and understandable; more than half of providers desired more education in this area.³⁸ Furthermore, pediatric providers may overestimate parental health literacy levels,³⁹ which may contribute to variability in the readability of written health materials. While education alone can contribute to a provider’s ability to create readable instructions, we note the improvement after the introduction of disease templates to demonstrate the importance of workflow-integrated higher reliability interventions to sustain improvements.

Our baseline poor readability rates were due to limited knowledge by frontline providers composing the instructions and a system in which an important element for successful hospital-to-home transitions was not tackled until patients were ready for discharge. Streamlining of the discharge process, including the creation of discharge instructions, may lead to improved efficiency, fewer discrepancies, more effective communication, and an enhanced family experience. Moreover, the success of our improvement work was due to key stakeholders, including parents, being a part of the team and the notable buy-in from providers.

Our work was not without limitations. We excluded non-English speaking families from the study. We were unable to measure reading level of our population directly and instead based our goals on national estimates. Our primary measure was readability, which is only 1 piece contributing to quality discharge instructions. Understandability and actionability are also important considerations; ^17,20,29,40 however, improvements in these areas were limited by our design options within the EHR. Our efforts focused on children with common general pediatric diagnoses, and it is unclear how our interventions would generalize to medically complex patients with more volume of information to communicate at discharge and with uncommon diagnoses that are less readily incorporated into standardized templates. Relatedly, our work occurred at the satellite campus of our tertiary care center and may not represent generalizable material or methods to implement templates at our main campus location or at other hospitals. To begin to better understand this, we have spread to HM patients at our main campus, including medically complex patients with technology dependence and/or neurological impairments. Standardized, disease-specific templates most relevant to this population as well as several patient specific templates, for those with frequent readmissions due to medical complexity, have been created and are actively being tested.

In conclusion, in using interventions targeted at standardization of discharge instructions and timely feedback to staff, we saw rapid, dramatic, and sustained improvement in the readability of discharge instructions. Next steps include adaptation and spread to other patient populations and care teams, collaborations with other centers, and assessing the impact of effectively written discharge instructions on patient outcomes, such as adverse drug events, readmission rates, and family experience.

Disclosure

No external funding was secured for this study. Dr. Brady is supported by a Patient-Centered Outcomes Research Mentored Clinical Investigator Award from the Agency for Healthcare Research and Quality, Award Number K08HS023827. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding organizations. The funding organization had no role in the design, preparation, review, or approval of this paper; nor the decision to submit the manuscript for publication. The authors have no financial relationships relevant to this article to disclose.

Direct-acting oral anticoagulants (DOACs) have been introduced into clinical use for stroke prevention in patients with nonvalvular atrial fibrillation (NVAF), prevention of venous thrombosis after hip or knee surgery, and treatment of deep vein thrombosis (DVT) and pulmonary embolism (PE).^1-7 Advantages of DOACs over warfarin are often stated as fixed dosing, minor drug and food interactions, wider therapeutic index, and no need for laboratory test monitoring.^1,8 Yet, recommended DOAC dosages vary by renal function and therapeutic indications. Dosing recommendations for prevention of stroke in patients with NVAF are based on estimated creatinine clearance (dabigatran, rivaroxaban, edoxaban), age (apixaban), weight (apixaban, edoxaban), serum creatinine level (apixaban, edoxaban), and presence of cirrhosis by Child-Pugh class^9,10 (apixaban, edoxaban).^4-6,11,12 Dosing recommendations based on coadministration of strong CYP34A and P-glycoprotein inhibitors or inducers vary by DOAC. In addition, dabigatran cannot be crushed and must be stored in its original packaging, and rivaroxaban should be taken with food when the dose is over 10 mg.

We studied DOAC prescribing in adults admitted to a large academic medical center by comparing initial prescribed dosing with FDA-approved prescribing information. We hypothesized that the complexity of DOAC dosing may not be recognized by prescribers.

Our study protocol was approved by the Committee on Human Research (Institutional Review Board) of the University of California San Francisco.

Data Collection

We used electronic medical records (EMRs) to identify adult inpatients who were prescribed a DOAC (apixaban, dabigatran, edoxaban, or rivaroxaban) at the University of California San Francisco Medical Center, a large academic hospital, between July 1, 2014 and June 30, 2015. Demographic and medical information related to therapeutic indications, contraindications, and indications for dose adjustments were collected and included diagnoses classified by International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) for venous thromboses; phlebitis or thrombophlebitis; PE or venous embolism; atrial arrhythmias; surgical procedures; cirrhosis and/or ascites or liver disease; coagulopathies; artificial heart valves or implanted devices; prior use of medications including parenteral anticoagulants; and laboratory data obtained before the first DOAC order (serum creatinine level, estimated glomerular filtration rate [eGFR] determined by Chronic Kidney Disease Epidemiology Collaboration,¹³ international normalized ratio, or, if available, activated partial thromboplastin time and bilirubin level). Creatinine clearance was calculated with the Cockcroft-Gault method¹⁴ using total body weight, per drug label recommendation. Child-Pugh class was calculated if cirrhosis was diagnosed.¹⁰ DOAC dose, frequency, dosing directions, and prescriber medical specialty were determined.

Accuracy of search results was confirmed by review of the first 200 patients’ records. Records were manually reviewed for encounters lacking ICD-9/10 codes and approved DOAC indications (30%) and encounters having multiple coded diagnostic indications (to identify the indication). ICD-9 codes for venous thrombosis were reviewed to differentiate acute from chronic events.

Data Analysis

The main outcome was concordance or discordance between the first DOAC prescribing order and the FDA-approved prescribing information at the time. Initial classification, performed by 2 independent reviewers (a pharmacist and a physician, or 2 pharmacists), was followed by adjudication and individual record review (by 2 independent reviewers) of all initial prescribing orders classified as discordant. A third reviewer adjudicated any disagreement. Records and notes were reviewed to identify stated or potential reasons for dosing variation and pre-admission prescriptions. Data are presented as means and standard deviations (SDs) and as raw numbers and percentages. Differences in patient characteristics by DOAC or therapeutic indication were determined by analysis of variance (ANOVA) with Bonferroni correction for post hoc comparisons. Dosing information was categorized as the same as recommended, lower than recommended, higher than recommended, or avoid drug use (drug–drug or drug–disease interaction), per FDA-approved prescribing information, and χ² tests were used to determine whether variation in dosing occurred by individual DOAC, therapeutic indication, or prescriber specialty. Relationships between dosing variation and age or renal function were tested by ANOVA with Bonferroni correction for post hoc comparisons.

Patients with AF were older with lower creatinine clearance compared to patients with other diagnoses. Mean (SD) patient age was 72.1 (12.7) years for AF, 53.1 (10.9) years for chronic PE, 55.5 (14) years for acute PE, 56.4 (15.9) years for chronic DVT, 57.9 (18.4) years for acute DVT, and 61.4 (11.6) years for DVT prevention after hip or knee surgery (P < 0.0001 for all comparisons). Mean (SD) estimated creatinine clearance was 76.8 (43.5) mL/min for AF, 92.4 (44.4) mL/min for DVT prevention after hip or knee surgery, 111 (53) mL/min for chronic DVT, 118 (55) mL/min for acute DVT, 126 (60) mL/min for chronic PE, and 127 (54) mL/min for acute PE (P < 0.0001 for all comparisons). Differences between patient groups by therapeutic indication were not detected for weight, body mass index, or serum creatinine level.

The most frequent deviation from prescribing recommendations was omission of directions to administer rivaroxaban with food—93% (248/268) of orders—but not for DVT prevention after hip or knee surgery, for which the 10-mg dose is appropriately administered without food. Doses were the same as recommended for 82% of apixaban orders, 84% of rivaroxaban orders, and 93% of initial dabigatran orders (P < 0.05 for differences; Table 3). Dosages not concordant with FDA recommendations were prescribed in 44 (18.1%) of 243 apixaban orders, 41 (14.3%) of 286 rivaroxaban orders, and 7 (7.2%) of 89 initial dabigatran orders. Lower than recommended doses were more common than higher than recommended doses (Table 3, Figure 1): 15.2% versus 2.1% of apixaban orders, 9.4% versus 3.5% of rivaroxaban orders, and 4.2% versus 1.0% of initial dabigatran orders (P < 0.05). Failure to avoid drug use (for potential drug–drug or drug–disease interactions) was uncommon (1%-2%). There were more deviations from recommended doses for patients with AF or DVT prevention after hip or knee surgery than for patients with acute or chronic PE or acute DVT (Table 3). No significant differences were detected between prescribed and recommended doses by prescriber specialty.

For apixaban, patients who were prescribed lower than recommended doses were older than those prescribed recommended doses: mean (SD), 78.1 (12.2) years versus 71 (13.6) years (P = 0.003). Seventy-six percent of those prescribed lower than recommended doses were older than 75. Prescriptions for apixaban at lower than recommended doses were continuations of prior outpatient prescriptions in 20 of 37 cases (almost half), and in 12 cases (one-fourth) antiplatelet drugs were coprescribed (aspirin in 10 cases, clopidogrel in 1, prasugrel in 1). For rivaroxaban, older age was associated with both lower than recommended dosing (P = 0.003) and higher than recommended dosing (P < 0.001). Variations from prescribing recommendations were continuations of outpatient rivaroxaban doses in about two-thirds (26 of 41; 63.4 %) with 13 receiving antiplatelet drugs. For dabigatran, 6 of 7 orders not in agreement with recommendations were continuations of outpatient dosing.

The specific equation used to estimate renal function also had the potential to lead to dosing errors. Among the 41 rivaroxaban patients categorized as receiving doses discordant with recommendations, 8 would have had an inappropriate DOAC dose if eGFR were used instead of eCrCL as recommended. No relationships were detected for other patient variables/measures and dosing deviations from recommendations.

We examined initial hospital orders for DOACs in adults admitted to a single academic medical center during 2014-2015. Dabigatran, apixaban and rivaroxaban were prescribed for prevention of stroke in patients with atrial fibrillation/flutter (AF) in three quarters of the encounters similar to national patterns. (15) Prescribing departures from FDA-approved recommendations ranged from failure to prescribe rivaroxaban with food to failure to recognize drug-drug interactions in 1% to 2%. Unexpectedly, lower than recommended dosing was more common than higher than recommended dosing of the three DOACs.

Rivaroxaban bioavailability is dose dependent with the presence of food required to enhance absorption for doses over 10 mg that are used for prevention of stroke in patients with non-valvular AF or treatment of DVT or PE.^5,16 Peak rivaroxaban concentrations are 75% higher and the total area under the concentration vs. time curve after dosing is 40% higher when rivaroxaban is administered with high fat high calorie meals compared to the fasting state.¹⁶ If rivaroxaban is not administered with food, drug concentrations and pharmacologic effects may be less than in clinical trials that specified co-administration with food.^17-19 A small survey of outpatients receiving rivaroxaban found that 23% reported taking it without food.²⁰ With electronic pharmacy systems in almost all hospitals and electronic prescriber order entry in most, automated addition of directions for rivaroxaban administration with food for doses over 10 mg to labels or dispensing instructions could easily correct this deviation from recommended practice.

Lower than recommended doses were prescribed in 9.4% of orders for rivaroxaban and 15.2% of orders for apixaban, with dose-deviations often appearing to be a continuation of outpatient doses. Patients 75 years or older were more likely to receive lower than recommended dosing of apixaban. Reductions in apixaban doses from 5 mg twice daily to 2.5 mg twice daily are recommended in patients with non-valvular AF with two of the following criteria: age ≥80 y, weight ≤60 kg, serum creatinine ≥1.5 mg/dL or co-administration of a strong PgP inhibitor to a patient without 2 of the 3 dose reduction criteria. Our study was not designed to determine reasons for under-dosing, but we speculate that clinicians may have considered patients aged 75-79 years to be similar to those 80 years of age or older, or, older and not as healthy as those enrolled in randomized trials.^21-25 The median age of our patients with AF receiving apixaban was 75y (interquartile range of 16) vs 70y ( interquartile range 63-76) in the pivotal trial comparing warfarin to apixaban.²¹ Renal function was also lower with 37% having eCrCL below 50 mL/min compared to 17% in ARISTOTLE. (21). Twenty-six percent of our apixaban-treated AF patients qualified for the lower 2.5 mg twice daily compared to only 5% of ARISTOTLE participants,²¹ further suggesting differences between patients in our sample compared to randomized trial participants.

Concerns regarding bleeding or falls in older patients, may also have contributed to lower than recommended doses. Recent analyses of patients at risk for falls confirmed that increased risk of falling was associated with more bone fractures, bleeding and all-cause death but not stroke or systemic emboli, and with less severe bleeding with the DOAC edoxaban compared to warfarin.²⁶ While a rationale for personalized or lower than recommended dosing of apixaban may exist in very old patients and those at risk of falls and bleeding, more data are needed to determine outcomes of lower than recommended doses of DOACs before such an approach can be endorsed. Monitoring of anticoagulant effect in patients who receive doses lower than those investigated in clinical trials could provide important information. The assays that measure DOAC effects are likely to be more available because of the use of reversal agents in the setting of bleeding with DOACs.²⁷

We had anticipated higher than recommended dosing for rivaroxaban as recommendations are based on creatinine clearance while laboratories routinely report estimated glomerular filtration rate (eGFR) that can provide higher estimates of renal clearance and estimated DOAC doses in older and smaller individuals.²⁸ Higher than recommended dosing was found in only 3.5% of our sample. In half, eGFR estimates were higher than creatinine clearance estimates. An international postmarketing registry of rivaroxaban use for the prevention of stroke in patients with NVAF, which included outpatients, found that 36% of those with creatinine clearances below 50 mL/min received a dose higher than recommended, and 15% received a dose lower than expected.²⁹ A more recent outpatient registry report on patients with NVAF, in which apixaban, dabigatran, or rivaroxaban was administered, found that overall 9.4% received a dose lower than recommended, and 3.4% were overdosed, with a similar percentage (34%) of rivaroxaban patients with creatinine clearance of 15 to 50 mL/min receiving higher than recommended dosing.³⁰ The lower rate of higher-than-recommended doses that we observed may have been related to the routine measurement of serum creatinine and attention to dosing adjustments for renal function in the inpatient setting compared to the outpatient setting. In addition, renal function data may not be available to outpatient pharmacies, limiting potential input on dosing recommendations. At least one cardiac society recommends monitoring of renal function in patients treated with DOACs, annually in patients with normal estimated creatinine clearance and more frequently (at intervals in months equal to the creatinine clearance divided by 10) in patients with abnormal creatinine clearance.¹¹ A hospital encounter provides an opportunity to assess or reassess renal status to optimize DOAC dosing.

Dabigatran was the first DOAC introduced into use in the United States with the same dose recommended for prevention of stroke in patients with AF or venous thromboembolic disease with reductions for creatinine clearance below 30 mL/min or creatinine clearance between 30 and 50 mL/min and concomitant use of the potent P-glycoprotein inhibitor dronedarone or systemic ketoconazole. The relative simplicity of dosing may have been responsible for the lowest rate of prescribing outside of recommendations observed in this study, but the low dabigatran use limits analyses of contributing factors.

Failure to avoid drug use in combination with use of strong P-glycoprotein inducers or inhibitors was infrequent but should be preventable. Current prescribing recommendations refer to “strong” P-glycoprotein inhibitors and list different specific agents that interact with each DOAC without a standardized definition or classification. Standardized classifications or reference sources would be helpful.

Our primary goal in this study was to compare initial prescribed dosing of DOACs with FDA-approved prescribing directions. However, therapeutic indication data warrant discussion. In our sample, 7.5% of patients with AF had bioprosthetic valves or recent mitral valve repair or replacement. Using the NVAF definition found in the 2014 AHA/ACC/HRS (American Heart Association, American College of Cardiology, Heart Rhythm Society) AF guidelines¹—“absence of rheumatic mitral valve disease, a prosthetic heart valve, or mitral valve repair”—these patients would not appear to be candidates for DOACs. However, arguments have been made that a bioprosthetic heart valve or native valve after valve repair does not have a risk profile for thromboembolism that differs from other forms of NVAF and would be equally responsive to DOAC therapy.³¹ Data are sparse, but retrospective subanalyses of limited numbers of patients with valvular disease (including bioprosthesis and mitral repair patients but excluding mechanical valve patients) enrolled in the pivotal DOAC studies support this conclusion.³² For the first months after biological valve replacement (including catheter-based valve replacement), recent European guidelines recommend vitamin K antagonists but also state, “NOACs probably deliver the same protection.”⁸ DOACs were also used for management of venous thromboembolic disease (both acute and chronic) in patients with active cancer. Our data predate the most recent American College of Chest Physician guidelines on treatment of venous thromboembolism in patients with cancer, which provide grade 2B recommendations for use of low-molecular-weight heparin (LMWH) over vitamin K antagonists and grade 2C recommendations for use of LMWH over dabigatran, rivaroxaban, apixaban, or edoxaban.³³

Our study had several limitations. First, data were from a single US academic medical center, though similar rates of prescribing deviation from recommendations have been reported for rivaroxaban and dabigatran in NVAF patients in other countries.^29,34 Second, therapeutic indications may have been misclassified because of errors, incomplete EMR data, or multiple indications. Third, we analyzed the first DOAC order and not dispensing information or subsequent corrections. Therefore, deviations from recommendations should not be interpreted as errors that reached patients. We evaluated dosing based on the measures used at the time of hospital admission, noting that, in a significant fraction of deviations from recommended doses, they represented continuations of outpatient doses when renal function or weight may have differed, and it is unknown whether patients were counseled to take rivaroxaban with food in the outpatient setting. Fourth, the number of patients with acute DVT was small, so firm conclusions cannot be drawn for this specific population. Fifth, our estimates of off-label dosing may have been underestimates, as data on cancer and cancer activity or cardiac valvular disease may not have been complete.

Healthcare professionals are prescribing DOACs in ways that differ from recommendations. These differences may reflect the older ages and reduced renal function of clinical populations relative to randomized clinical trial groups, but they could also potentially alter clinical efficacy. Our findings support the need to evaluate the appropriateness and dosing of DOACs at each encounter and to determine the outcomes of patients treated with lower than recommended doses of DOACs and the outcomes of DOAC-treated patients with bioprostheses or active malignancies.

Acknowledgment

The authors thank Tobias Schmelzinger for electronic data extraction and compilation and University of California San Francisco students Eduardo De La Torre Cruz (School of Pharmacy) and Carlos Mikell (School of Medicine) for assistance with data review.

Disclosure

Dr. Schwartz reports receiving personal fees from Bristol-Myers Squibb and Amgen and grants from Bristol-Myers Squibb and Pfizer, outside the submitted work. The other authors have nothing to report.

Direct-acting oral anticoagulants (DOACs) have been introduced into clinical use for stroke prevention in patients with nonvalvular atrial fibrillation (NVAF), prevention of venous thrombosis after hip or knee surgery, and treatment of deep vein thrombosis (DVT) and pulmonary embolism (PE).^1-7 Advantages of DOACs over warfarin are often stated as fixed dosing, minor drug and food interactions, wider therapeutic index, and no need for laboratory test monitoring.^1,8 Yet, recommended DOAC dosages vary by renal function and therapeutic indications. Dosing recommendations for prevention of stroke in patients with NVAF are based on estimated creatinine clearance (dabigatran, rivaroxaban, edoxaban), age (apixaban), weight (apixaban, edoxaban), serum creatinine level (apixaban, edoxaban), and presence of cirrhosis by Child-Pugh class^9,10 (apixaban, edoxaban).^4-6,11,12 Dosing recommendations based on coadministration of strong CYP34A and P-glycoprotein inhibitors or inducers vary by DOAC. In addition, dabigatran cannot be crushed and must be stored in its original packaging, and rivaroxaban should be taken with food when the dose is over 10 mg.

We studied DOAC prescribing in adults admitted to a large academic medical center by comparing initial prescribed dosing with FDA-approved prescribing information. We hypothesized that the complexity of DOAC dosing may not be recognized by prescribers.

Our study protocol was approved by the Committee on Human Research (Institutional Review Board) of the University of California San Francisco.

Data Collection

We used electronic medical records (EMRs) to identify adult inpatients who were prescribed a DOAC (apixaban, dabigatran, edoxaban, or rivaroxaban) at the University of California San Francisco Medical Center, a large academic hospital, between July 1, 2014 and June 30, 2015. Demographic and medical information related to therapeutic indications, contraindications, and indications for dose adjustments were collected and included diagnoses classified by International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) for venous thromboses; phlebitis or thrombophlebitis; PE or venous embolism; atrial arrhythmias; surgical procedures; cirrhosis and/or ascites or liver disease; coagulopathies; artificial heart valves or implanted devices; prior use of medications including parenteral anticoagulants; and laboratory data obtained before the first DOAC order (serum creatinine level, estimated glomerular filtration rate [eGFR] determined by Chronic Kidney Disease Epidemiology Collaboration,¹³ international normalized ratio, or, if available, activated partial thromboplastin time and bilirubin level). Creatinine clearance was calculated with the Cockcroft-Gault method¹⁴ using total body weight, per drug label recommendation. Child-Pugh class was calculated if cirrhosis was diagnosed.¹⁰ DOAC dose, frequency, dosing directions, and prescriber medical specialty were determined.

Accuracy of search results was confirmed by review of the first 200 patients’ records. Records were manually reviewed for encounters lacking ICD-9/10 codes and approved DOAC indications (30%) and encounters having multiple coded diagnostic indications (to identify the indication). ICD-9 codes for venous thrombosis were reviewed to differentiate acute from chronic events.

Data Analysis

The main outcome was concordance or discordance between the first DOAC prescribing order and the FDA-approved prescribing information at the time. Initial classification, performed by 2 independent reviewers (a pharmacist and a physician, or 2 pharmacists), was followed by adjudication and individual record review (by 2 independent reviewers) of all initial prescribing orders classified as discordant. A third reviewer adjudicated any disagreement. Records and notes were reviewed to identify stated or potential reasons for dosing variation and pre-admission prescriptions. Data are presented as means and standard deviations (SDs) and as raw numbers and percentages. Differences in patient characteristics by DOAC or therapeutic indication were determined by analysis of variance (ANOVA) with Bonferroni correction for post hoc comparisons. Dosing information was categorized as the same as recommended, lower than recommended, higher than recommended, or avoid drug use (drug–drug or drug–disease interaction), per FDA-approved prescribing information, and χ² tests were used to determine whether variation in dosing occurred by individual DOAC, therapeutic indication, or prescriber specialty. Relationships between dosing variation and age or renal function were tested by ANOVA with Bonferroni correction for post hoc comparisons.

Patients with AF were older with lower creatinine clearance compared to patients with other diagnoses. Mean (SD) patient age was 72.1 (12.7) years for AF, 53.1 (10.9) years for chronic PE, 55.5 (14) years for acute PE, 56.4 (15.9) years for chronic DVT, 57.9 (18.4) years for acute DVT, and 61.4 (11.6) years for DVT prevention after hip or knee surgery (P < 0.0001 for all comparisons). Mean (SD) estimated creatinine clearance was 76.8 (43.5) mL/min for AF, 92.4 (44.4) mL/min for DVT prevention after hip or knee surgery, 111 (53) mL/min for chronic DVT, 118 (55) mL/min for acute DVT, 126 (60) mL/min for chronic PE, and 127 (54) mL/min for acute PE (P < 0.0001 for all comparisons). Differences between patient groups by therapeutic indication were not detected for weight, body mass index, or serum creatinine level.

The most frequent deviation from prescribing recommendations was omission of directions to administer rivaroxaban with food—93% (248/268) of orders—but not for DVT prevention after hip or knee surgery, for which the 10-mg dose is appropriately administered without food. Doses were the same as recommended for 82% of apixaban orders, 84% of rivaroxaban orders, and 93% of initial dabigatran orders (P < 0.05 for differences; Table 3). Dosages not concordant with FDA recommendations were prescribed in 44 (18.1%) of 243 apixaban orders, 41 (14.3%) of 286 rivaroxaban orders, and 7 (7.2%) of 89 initial dabigatran orders. Lower than recommended doses were more common than higher than recommended doses (Table 3, Figure 1): 15.2% versus 2.1% of apixaban orders, 9.4% versus 3.5% of rivaroxaban orders, and 4.2% versus 1.0% of initial dabigatran orders (P < 0.05). Failure to avoid drug use (for potential drug–drug or drug–disease interactions) was uncommon (1%-2%). There were more deviations from recommended doses for patients with AF or DVT prevention after hip or knee surgery than for patients with acute or chronic PE or acute DVT (Table 3). No significant differences were detected between prescribed and recommended doses by prescriber specialty.

For apixaban, patients who were prescribed lower than recommended doses were older than those prescribed recommended doses: mean (SD), 78.1 (12.2) years versus 71 (13.6) years (P = 0.003). Seventy-six percent of those prescribed lower than recommended doses were older than 75. Prescriptions for apixaban at lower than recommended doses were continuations of prior outpatient prescriptions in 20 of 37 cases (almost half), and in 12 cases (one-fourth) antiplatelet drugs were coprescribed (aspirin in 10 cases, clopidogrel in 1, prasugrel in 1). For rivaroxaban, older age was associated with both lower than recommended dosing (P = 0.003) and higher than recommended dosing (P < 0.001). Variations from prescribing recommendations were continuations of outpatient rivaroxaban doses in about two-thirds (26 of 41; 63.4 %) with 13 receiving antiplatelet drugs. For dabigatran, 6 of 7 orders not in agreement with recommendations were continuations of outpatient dosing.

The specific equation used to estimate renal function also had the potential to lead to dosing errors. Among the 41 rivaroxaban patients categorized as receiving doses discordant with recommendations, 8 would have had an inappropriate DOAC dose if eGFR were used instead of eCrCL as recommended. No relationships were detected for other patient variables/measures and dosing deviations from recommendations.

We examined initial hospital orders for DOACs in adults admitted to a single academic medical center during 2014-2015. Dabigatran, apixaban and rivaroxaban were prescribed for prevention of stroke in patients with atrial fibrillation/flutter (AF) in three quarters of the encounters similar to national patterns. (15) Prescribing departures from FDA-approved recommendations ranged from failure to prescribe rivaroxaban with food to failure to recognize drug-drug interactions in 1% to 2%. Unexpectedly, lower than recommended dosing was more common than higher than recommended dosing of the three DOACs.

Rivaroxaban bioavailability is dose dependent with the presence of food required to enhance absorption for doses over 10 mg that are used for prevention of stroke in patients with non-valvular AF or treatment of DVT or PE.^5,16 Peak rivaroxaban concentrations are 75% higher and the total area under the concentration vs. time curve after dosing is 40% higher when rivaroxaban is administered with high fat high calorie meals compared to the fasting state.¹⁶ If rivaroxaban is not administered with food, drug concentrations and pharmacologic effects may be less than in clinical trials that specified co-administration with food.^17-19 A small survey of outpatients receiving rivaroxaban found that 23% reported taking it without food.²⁰ With electronic pharmacy systems in almost all hospitals and electronic prescriber order entry in most, automated addition of directions for rivaroxaban administration with food for doses over 10 mg to labels or dispensing instructions could easily correct this deviation from recommended practice.

Lower than recommended doses were prescribed in 9.4% of orders for rivaroxaban and 15.2% of orders for apixaban, with dose-deviations often appearing to be a continuation of outpatient doses. Patients 75 years or older were more likely to receive lower than recommended dosing of apixaban. Reductions in apixaban doses from 5 mg twice daily to 2.5 mg twice daily are recommended in patients with non-valvular AF with two of the following criteria: age ≥80 y, weight ≤60 kg, serum creatinine ≥1.5 mg/dL or co-administration of a strong PgP inhibitor to a patient without 2 of the 3 dose reduction criteria. Our study was not designed to determine reasons for under-dosing, but we speculate that clinicians may have considered patients aged 75-79 years to be similar to those 80 years of age or older, or, older and not as healthy as those enrolled in randomized trials.^21-25 The median age of our patients with AF receiving apixaban was 75y (interquartile range of 16) vs 70y ( interquartile range 63-76) in the pivotal trial comparing warfarin to apixaban.²¹ Renal function was also lower with 37% having eCrCL below 50 mL/min compared to 17% in ARISTOTLE. (21). Twenty-six percent of our apixaban-treated AF patients qualified for the lower 2.5 mg twice daily compared to only 5% of ARISTOTLE participants,²¹ further suggesting differences between patients in our sample compared to randomized trial participants.

Concerns regarding bleeding or falls in older patients, may also have contributed to lower than recommended doses. Recent analyses of patients at risk for falls confirmed that increased risk of falling was associated with more bone fractures, bleeding and all-cause death but not stroke or systemic emboli, and with less severe bleeding with the DOAC edoxaban compared to warfarin.²⁶ While a rationale for personalized or lower than recommended dosing of apixaban may exist in very old patients and those at risk of falls and bleeding, more data are needed to determine outcomes of lower than recommended doses of DOACs before such an approach can be endorsed. Monitoring of anticoagulant effect in patients who receive doses lower than those investigated in clinical trials could provide important information. The assays that measure DOAC effects are likely to be more available because of the use of reversal agents in the setting of bleeding with DOACs.²⁷

We had anticipated higher than recommended dosing for rivaroxaban as recommendations are based on creatinine clearance while laboratories routinely report estimated glomerular filtration rate (eGFR) that can provide higher estimates of renal clearance and estimated DOAC doses in older and smaller individuals.²⁸ Higher than recommended dosing was found in only 3.5% of our sample. In half, eGFR estimates were higher than creatinine clearance estimates. An international postmarketing registry of rivaroxaban use for the prevention of stroke in patients with NVAF, which included outpatients, found that 36% of those with creatinine clearances below 50 mL/min received a dose higher than recommended, and 15% received a dose lower than expected.²⁹ A more recent outpatient registry report on patients with NVAF, in which apixaban, dabigatran, or rivaroxaban was administered, found that overall 9.4% received a dose lower than recommended, and 3.4% were overdosed, with a similar percentage (34%) of rivaroxaban patients with creatinine clearance of 15 to 50 mL/min receiving higher than recommended dosing.³⁰ The lower rate of higher-than-recommended doses that we observed may have been related to the routine measurement of serum creatinine and attention to dosing adjustments for renal function in the inpatient setting compared to the outpatient setting. In addition, renal function data may not be available to outpatient pharmacies, limiting potential input on dosing recommendations. At least one cardiac society recommends monitoring of renal function in patients treated with DOACs, annually in patients with normal estimated creatinine clearance and more frequently (at intervals in months equal to the creatinine clearance divided by 10) in patients with abnormal creatinine clearance.¹¹ A hospital encounter provides an opportunity to assess or reassess renal status to optimize DOAC dosing.

Dabigatran was the first DOAC introduced into use in the United States with the same dose recommended for prevention of stroke in patients with AF or venous thromboembolic disease with reductions for creatinine clearance below 30 mL/min or creatinine clearance between 30 and 50 mL/min and concomitant use of the potent P-glycoprotein inhibitor dronedarone or systemic ketoconazole. The relative simplicity of dosing may have been responsible for the lowest rate of prescribing outside of recommendations observed in this study, but the low dabigatran use limits analyses of contributing factors.

Failure to avoid drug use in combination with use of strong P-glycoprotein inducers or inhibitors was infrequent but should be preventable. Current prescribing recommendations refer to “strong” P-glycoprotein inhibitors and list different specific agents that interact with each DOAC without a standardized definition or classification. Standardized classifications or reference sources would be helpful.

Our primary goal in this study was to compare initial prescribed dosing of DOACs with FDA-approved prescribing directions. However, therapeutic indication data warrant discussion. In our sample, 7.5% of patients with AF had bioprosthetic valves or recent mitral valve repair or replacement. Using the NVAF definition found in the 2014 AHA/ACC/HRS (American Heart Association, American College of Cardiology, Heart Rhythm Society) AF guidelines¹—“absence of rheumatic mitral valve disease, a prosthetic heart valve, or mitral valve repair”—these patients would not appear to be candidates for DOACs. However, arguments have been made that a bioprosthetic heart valve or native valve after valve repair does not have a risk profile for thromboembolism that differs from other forms of NVAF and would be equally responsive to DOAC therapy.³¹ Data are sparse, but retrospective subanalyses of limited numbers of patients with valvular disease (including bioprosthesis and mitral repair patients but excluding mechanical valve patients) enrolled in the pivotal DOAC studies support this conclusion.³² For the first months after biological valve replacement (including catheter-based valve replacement), recent European guidelines recommend vitamin K antagonists but also state, “NOACs probably deliver the same protection.”⁸ DOACs were also used for management of venous thromboembolic disease (both acute and chronic) in patients with active cancer. Our data predate the most recent American College of Chest Physician guidelines on treatment of venous thromboembolism in patients with cancer, which provide grade 2B recommendations for use of low-molecular-weight heparin (LMWH) over vitamin K antagonists and grade 2C recommendations for use of LMWH over dabigatran, rivaroxaban, apixaban, or edoxaban.³³

Our study had several limitations. First, data were from a single US academic medical center, though similar rates of prescribing deviation from recommendations have been reported for rivaroxaban and dabigatran in NVAF patients in other countries.^29,34 Second, therapeutic indications may have been misclassified because of errors, incomplete EMR data, or multiple indications. Third, we analyzed the first DOAC order and not dispensing information or subsequent corrections. Therefore, deviations from recommendations should not be interpreted as errors that reached patients. We evaluated dosing based on the measures used at the time of hospital admission, noting that, in a significant fraction of deviations from recommended doses, they represented continuations of outpatient doses when renal function or weight may have differed, and it is unknown whether patients were counseled to take rivaroxaban with food in the outpatient setting. Fourth, the number of patients with acute DVT was small, so firm conclusions cannot be drawn for this specific population. Fifth, our estimates of off-label dosing may have been underestimates, as data on cancer and cancer activity or cardiac valvular disease may not have been complete.

Healthcare professionals are prescribing DOACs in ways that differ from recommendations. These differences may reflect the older ages and reduced renal function of clinical populations relative to randomized clinical trial groups, but they could also potentially alter clinical efficacy. Our findings support the need to evaluate the appropriateness and dosing of DOACs at each encounter and to determine the outcomes of patients treated with lower than recommended doses of DOACs and the outcomes of DOAC-treated patients with bioprostheses or active malignancies.

Acknowledgment

The authors thank Tobias Schmelzinger for electronic data extraction and compilation and University of California San Francisco students Eduardo De La Torre Cruz (School of Pharmacy) and Carlos Mikell (School of Medicine) for assistance with data review.

Disclosure

Dr. Schwartz reports receiving personal fees from Bristol-Myers Squibb and Amgen and grants from Bristol-Myers Squibb and Pfizer, outside the submitted work. The other authors have nothing to report.

Direct-acting oral anticoagulants (DOACs) have been introduced into clinical use for stroke prevention in patients with nonvalvular atrial fibrillation (NVAF), prevention of venous thrombosis after hip or knee surgery, and treatment of deep vein thrombosis (DVT) and pulmonary embolism (PE).^1-7 Advantages of DOACs over warfarin are often stated as fixed dosing, minor drug and food interactions, wider therapeutic index, and no need for laboratory test monitoring.^1,8 Yet, recommended DOAC dosages vary by renal function and therapeutic indications. Dosing recommendations for prevention of stroke in patients with NVAF are based on estimated creatinine clearance (dabigatran, rivaroxaban, edoxaban), age (apixaban), weight (apixaban, edoxaban), serum creatinine level (apixaban, edoxaban), and presence of cirrhosis by Child-Pugh class^9,10 (apixaban, edoxaban).^4-6,11,12 Dosing recommendations based on coadministration of strong CYP34A and P-glycoprotein inhibitors or inducers vary by DOAC. In addition, dabigatran cannot be crushed and must be stored in its original packaging, and rivaroxaban should be taken with food when the dose is over 10 mg.

We studied DOAC prescribing in adults admitted to a large academic medical center by comparing initial prescribed dosing with FDA-approved prescribing information. We hypothesized that the complexity of DOAC dosing may not be recognized by prescribers.

Our study protocol was approved by the Committee on Human Research (Institutional Review Board) of the University of California San Francisco.

Data Collection

We used electronic medical records (EMRs) to identify adult inpatients who were prescribed a DOAC (apixaban, dabigatran, edoxaban, or rivaroxaban) at the University of California San Francisco Medical Center, a large academic hospital, between July 1, 2014 and June 30, 2015. Demographic and medical information related to therapeutic indications, contraindications, and indications for dose adjustments were collected and included diagnoses classified by International Classification of Diseases, Ninth Revision (ICD-9) and Tenth Revision (ICD-10) for venous thromboses; phlebitis or thrombophlebitis; PE or venous embolism; atrial arrhythmias; surgical procedures; cirrhosis and/or ascites or liver disease; coagulopathies; artificial heart valves or implanted devices; prior use of medications including parenteral anticoagulants; and laboratory data obtained before the first DOAC order (serum creatinine level, estimated glomerular filtration rate [eGFR] determined by Chronic Kidney Disease Epidemiology Collaboration,¹³ international normalized ratio, or, if available, activated partial thromboplastin time and bilirubin level). Creatinine clearance was calculated with the Cockcroft-Gault method¹⁴ using total body weight, per drug label recommendation. Child-Pugh class was calculated if cirrhosis was diagnosed.¹⁰ DOAC dose, frequency, dosing directions, and prescriber medical specialty were determined.

Accuracy of search results was confirmed by review of the first 200 patients’ records. Records were manually reviewed for encounters lacking ICD-9/10 codes and approved DOAC indications (30%) and encounters having multiple coded diagnostic indications (to identify the indication). ICD-9 codes for venous thrombosis were reviewed to differentiate acute from chronic events.

Data Analysis

The main outcome was concordance or discordance between the first DOAC prescribing order and the FDA-approved prescribing information at the time. Initial classification, performed by 2 independent reviewers (a pharmacist and a physician, or 2 pharmacists), was followed by adjudication and individual record review (by 2 independent reviewers) of all initial prescribing orders classified as discordant. A third reviewer adjudicated any disagreement. Records and notes were reviewed to identify stated or potential reasons for dosing variation and pre-admission prescriptions. Data are presented as means and standard deviations (SDs) and as raw numbers and percentages. Differences in patient characteristics by DOAC or therapeutic indication were determined by analysis of variance (ANOVA) with Bonferroni correction for post hoc comparisons. Dosing information was categorized as the same as recommended, lower than recommended, higher than recommended, or avoid drug use (drug–drug or drug–disease interaction), per FDA-approved prescribing information, and χ² tests were used to determine whether variation in dosing occurred by individual DOAC, therapeutic indication, or prescriber specialty. Relationships between dosing variation and age or renal function were tested by ANOVA with Bonferroni correction for post hoc comparisons.

Patients with AF were older with lower creatinine clearance compared to patients with other diagnoses. Mean (SD) patient age was 72.1 (12.7) years for AF, 53.1 (10.9) years for chronic PE, 55.5 (14) years for acute PE, 56.4 (15.9) years for chronic DVT, 57.9 (18.4) years for acute DVT, and 61.4 (11.6) years for DVT prevention after hip or knee surgery (P < 0.0001 for all comparisons). Mean (SD) estimated creatinine clearance was 76.8 (43.5) mL/min for AF, 92.4 (44.4) mL/min for DVT prevention after hip or knee surgery, 111 (53) mL/min for chronic DVT, 118 (55) mL/min for acute DVT, 126 (60) mL/min for chronic PE, and 127 (54) mL/min for acute PE (P < 0.0001 for all comparisons). Differences between patient groups by therapeutic indication were not detected for weight, body mass index, or serum creatinine level.

The most frequent deviation from prescribing recommendations was omission of directions to administer rivaroxaban with food—93% (248/268) of orders—but not for DVT prevention after hip or knee surgery, for which the 10-mg dose is appropriately administered without food. Doses were the same as recommended for 82% of apixaban orders, 84% of rivaroxaban orders, and 93% of initial dabigatran orders (P < 0.05 for differences; Table 3). Dosages not concordant with FDA recommendations were prescribed in 44 (18.1%) of 243 apixaban orders, 41 (14.3%) of 286 rivaroxaban orders, and 7 (7.2%) of 89 initial dabigatran orders. Lower than recommended doses were more common than higher than recommended doses (Table 3, Figure 1): 15.2% versus 2.1% of apixaban orders, 9.4% versus 3.5% of rivaroxaban orders, and 4.2% versus 1.0% of initial dabigatran orders (P < 0.05). Failure to avoid drug use (for potential drug–drug or drug–disease interactions) was uncommon (1%-2%). There were more deviations from recommended doses for patients with AF or DVT prevention after hip or knee surgery than for patients with acute or chronic PE or acute DVT (Table 3). No significant differences were detected between prescribed and recommended doses by prescriber specialty.

For apixaban, patients who were prescribed lower than recommended doses were older than those prescribed recommended doses: mean (SD), 78.1 (12.2) years versus 71 (13.6) years (P = 0.003). Seventy-six percent of those prescribed lower than recommended doses were older than 75. Prescriptions for apixaban at lower than recommended doses were continuations of prior outpatient prescriptions in 20 of 37 cases (almost half), and in 12 cases (one-fourth) antiplatelet drugs were coprescribed (aspirin in 10 cases, clopidogrel in 1, prasugrel in 1). For rivaroxaban, older age was associated with both lower than recommended dosing (P = 0.003) and higher than recommended dosing (P < 0.001). Variations from prescribing recommendations were continuations of outpatient rivaroxaban doses in about two-thirds (26 of 41; 63.4 %) with 13 receiving antiplatelet drugs. For dabigatran, 6 of 7 orders not in agreement with recommendations were continuations of outpatient dosing.

The specific equation used to estimate renal function also had the potential to lead to dosing errors. Among the 41 rivaroxaban patients categorized as receiving doses discordant with recommendations, 8 would have had an inappropriate DOAC dose if eGFR were used instead of eCrCL as recommended. No relationships were detected for other patient variables/measures and dosing deviations from recommendations.

We examined initial hospital orders for DOACs in adults admitted to a single academic medical center during 2014-2015. Dabigatran, apixaban and rivaroxaban were prescribed for prevention of stroke in patients with atrial fibrillation/flutter (AF) in three quarters of the encounters similar to national patterns. (15) Prescribing departures from FDA-approved recommendations ranged from failure to prescribe rivaroxaban with food to failure to recognize drug-drug interactions in 1% to 2%. Unexpectedly, lower than recommended dosing was more common than higher than recommended dosing of the three DOACs.

Rivaroxaban bioavailability is dose dependent with the presence of food required to enhance absorption for doses over 10 mg that are used for prevention of stroke in patients with non-valvular AF or treatment of DVT or PE.^5,16 Peak rivaroxaban concentrations are 75% higher and the total area under the concentration vs. time curve after dosing is 40% higher when rivaroxaban is administered with high fat high calorie meals compared to the fasting state.¹⁶ If rivaroxaban is not administered with food, drug concentrations and pharmacologic effects may be less than in clinical trials that specified co-administration with food.^17-19 A small survey of outpatients receiving rivaroxaban found that 23% reported taking it without food.²⁰ With electronic pharmacy systems in almost all hospitals and electronic prescriber order entry in most, automated addition of directions for rivaroxaban administration with food for doses over 10 mg to labels or dispensing instructions could easily correct this deviation from recommended practice.

Lower than recommended doses were prescribed in 9.4% of orders for rivaroxaban and 15.2% of orders for apixaban, with dose-deviations often appearing to be a continuation of outpatient doses. Patients 75 years or older were more likely to receive lower than recommended dosing of apixaban. Reductions in apixaban doses from 5 mg twice daily to 2.5 mg twice daily are recommended in patients with non-valvular AF with two of the following criteria: age ≥80 y, weight ≤60 kg, serum creatinine ≥1.5 mg/dL or co-administration of a strong PgP inhibitor to a patient without 2 of the 3 dose reduction criteria. Our study was not designed to determine reasons for under-dosing, but we speculate that clinicians may have considered patients aged 75-79 years to be similar to those 80 years of age or older, or, older and not as healthy as those enrolled in randomized trials.^21-25 The median age of our patients with AF receiving apixaban was 75y (interquartile range of 16) vs 70y ( interquartile range 63-76) in the pivotal trial comparing warfarin to apixaban.²¹ Renal function was also lower with 37% having eCrCL below 50 mL/min compared to 17% in ARISTOTLE. (21). Twenty-six percent of our apixaban-treated AF patients qualified for the lower 2.5 mg twice daily compared to only 5% of ARISTOTLE participants,²¹ further suggesting differences between patients in our sample compared to randomized trial participants.

Concerns regarding bleeding or falls in older patients, may also have contributed to lower than recommended doses. Recent analyses of patients at risk for falls confirmed that increased risk of falling was associated with more bone fractures, bleeding and all-cause death but not stroke or systemic emboli, and with less severe bleeding with the DOAC edoxaban compared to warfarin.²⁶ While a rationale for personalized or lower than recommended dosing of apixaban may exist in very old patients and those at risk of falls and bleeding, more data are needed to determine outcomes of lower than recommended doses of DOACs before such an approach can be endorsed. Monitoring of anticoagulant effect in patients who receive doses lower than those investigated in clinical trials could provide important information. The assays that measure DOAC effects are likely to be more available because of the use of reversal agents in the setting of bleeding with DOACs.²⁷

We had anticipated higher than recommended dosing for rivaroxaban as recommendations are based on creatinine clearance while laboratories routinely report estimated glomerular filtration rate (eGFR) that can provide higher estimates of renal clearance and estimated DOAC doses in older and smaller individuals.²⁸ Higher than recommended dosing was found in only 3.5% of our sample. In half, eGFR estimates were higher than creatinine clearance estimates. An international postmarketing registry of rivaroxaban use for the prevention of stroke in patients with NVAF, which included outpatients, found that 36% of those with creatinine clearances below 50 mL/min received a dose higher than recommended, and 15% received a dose lower than expected.²⁹ A more recent outpatient registry report on patients with NVAF, in which apixaban, dabigatran, or rivaroxaban was administered, found that overall 9.4% received a dose lower than recommended, and 3.4% were overdosed, with a similar percentage (34%) of rivaroxaban patients with creatinine clearance of 15 to 50 mL/min receiving higher than recommended dosing.³⁰ The lower rate of higher-than-recommended doses that we observed may have been related to the routine measurement of serum creatinine and attention to dosing adjustments for renal function in the inpatient setting compared to the outpatient setting. In addition, renal function data may not be available to outpatient pharmacies, limiting potential input on dosing recommendations. At least one cardiac society recommends monitoring of renal function in patients treated with DOACs, annually in patients with normal estimated creatinine clearance and more frequently (at intervals in months equal to the creatinine clearance divided by 10) in patients with abnormal creatinine clearance.¹¹ A hospital encounter provides an opportunity to assess or reassess renal status to optimize DOAC dosing.

Dabigatran was the first DOAC introduced into use in the United States with the same dose recommended for prevention of stroke in patients with AF or venous thromboembolic disease with reductions for creatinine clearance below 30 mL/min or creatinine clearance between 30 and 50 mL/min and concomitant use of the potent P-glycoprotein inhibitor dronedarone or systemic ketoconazole. The relative simplicity of dosing may have been responsible for the lowest rate of prescribing outside of recommendations observed in this study, but the low dabigatran use limits analyses of contributing factors.

Failure to avoid drug use in combination with use of strong P-glycoprotein inducers or inhibitors was infrequent but should be preventable. Current prescribing recommendations refer to “strong” P-glycoprotein inhibitors and list different specific agents that interact with each DOAC without a standardized definition or classification. Standardized classifications or reference sources would be helpful.

Our primary goal in this study was to compare initial prescribed dosing of DOACs with FDA-approved prescribing directions. However, therapeutic indication data warrant discussion. In our sample, 7.5% of patients with AF had bioprosthetic valves or recent mitral valve repair or replacement. Using the NVAF definition found in the 2014 AHA/ACC/HRS (American Heart Association, American College of Cardiology, Heart Rhythm Society) AF guidelines¹—“absence of rheumatic mitral valve disease, a prosthetic heart valve, or mitral valve repair”—these patients would not appear to be candidates for DOACs. However, arguments have been made that a bioprosthetic heart valve or native valve after valve repair does not have a risk profile for thromboembolism that differs from other forms of NVAF and would be equally responsive to DOAC therapy.³¹ Data are sparse, but retrospective subanalyses of limited numbers of patients with valvular disease (including bioprosthesis and mitral repair patients but excluding mechanical valve patients) enrolled in the pivotal DOAC studies support this conclusion.³² For the first months after biological valve replacement (including catheter-based valve replacement), recent European guidelines recommend vitamin K antagonists but also state, “NOACs probably deliver the same protection.”⁸ DOACs were also used for management of venous thromboembolic disease (both acute and chronic) in patients with active cancer. Our data predate the most recent American College of Chest Physician guidelines on treatment of venous thromboembolism in patients with cancer, which provide grade 2B recommendations for use of low-molecular-weight heparin (LMWH) over vitamin K antagonists and grade 2C recommendations for use of LMWH over dabigatran, rivaroxaban, apixaban, or edoxaban.³³

Our study had several limitations. First, data were from a single US academic medical center, though similar rates of prescribing deviation from recommendations have been reported for rivaroxaban and dabigatran in NVAF patients in other countries.^29,34 Second, therapeutic indications may have been misclassified because of errors, incomplete EMR data, or multiple indications. Third, we analyzed the first DOAC order and not dispensing information or subsequent corrections. Therefore, deviations from recommendations should not be interpreted as errors that reached patients. We evaluated dosing based on the measures used at the time of hospital admission, noting that, in a significant fraction of deviations from recommended doses, they represented continuations of outpatient doses when renal function or weight may have differed, and it is unknown whether patients were counseled to take rivaroxaban with food in the outpatient setting. Fourth, the number of patients with acute DVT was small, so firm conclusions cannot be drawn for this specific population. Fifth, our estimates of off-label dosing may have been underestimates, as data on cancer and cancer activity or cardiac valvular disease may not have been complete.