Each day, hospitalists interact with a variety of specialists and sub‐specialists to provide consultative or procedural assistance in care of their patients. Physicians have a duty to practice beneficently and to simultaneously respect patients' autonomy.1 Whether to offer a treatment is a function of many variables, but when benefits approach zero, or when risks substantially outweigh benefits, physicians may justifiably withhold therapies without assent or consent of patients.2 The purpose of this article is to explore why it is accepted practice in the United States to permit unilateral withholding of some potentially life‐prolonging treatments (eg, surgery as the paradigm), while it is not common practice for other critical care procedures (eg, cardiopulmonary resuscitation [CPR]). We offer that these examples demonstrate the tension of 2 pillars of medical ethical conduct, namely beneficence and respect of autonomy.1

Consider 2 real cases that demonstrated a juxtaposition of diametrically opposing views of thoughtful, capable surgeons asked to provide life‐saving surgery to critically ill patients.

CASE 1

A 33‐year‐old man, with a history of obesity, presents with mild epigastric pain and hematemesis of a day's duration. Endoscopic evaluation demonstrates a deep gastric ulcer with visible vessel that is injected with epinephrine. He is transferred to the medical intensive care unit (ICU) for monitoring and has an uneventful first 24 hours. On his second hospital day, he develops severe epigastric pain of sudden onset, accompanied by light‐headedness. He is diaphoretic and dyspneic, sitting bolt upright. His body mass index (BMI) is 40 kg/m², and his vital signs are: 130/min, 140/80 mmHg, 30/min, 99.0F. Examination is normal except for severe upper abdominal tenderness, absent bowel sounds, and voluntary guarding. Abdominal computed tomography demonstrates a perforation, free air, and a loculated collection adjacent to the original ulcer. He is treated with 4 liters of crystalloids, oxygen, and an emergent surgical consultation is performed. The examining surgeon confirms the diagnosis of acute perforation, but asserts that his operative risk of mortality, due to obesity, is excessive. He will never get off the ventilator. He advises watchful waiting. The medical intensivist believes the patient will die without surgery; he asks for a second opinion. A more senior colleague assesses the patient and reiterates the first surgeon's opinion. The intensivist begins preparations to transfer the patient to the nearest tertiary care center for a third opinion, when the surgeons reverse themselves. The patient is taken to surgery where the collection is removed, with partial gastrectomy. He is extubated in the recovery room, spends 12 hours in the ICU, and is transferred to the wards where he undergoes an uneventful recovery.

CASE 2

A 50‐year‐old man, with a history of end‐stage alcoholic cirrhosis, presents to the intensive care unit with respiratory embarrassment associated with tense ascites, complicated by pneumococcal pneumonia. He responds to antibiotics but has rapidly reaccumulating ascites, where large volume paracentesis (of 4‐5 liters of transudative fluid) is required every 3 to 4 days to promote weaning trials. On his 20th hospital day, the patient develops fulminant septic shock, and work‐up reveals free air in the abdomen. A Board‐certified critical care surgeon meets with the family and informs them that he is willing to attempt exploratory laparotomy, but that operative mortality exceeds 95%. However, he was willing to try because the alternative otherwise is >99% mortality. The family asks for surgery, which reveals a small perforation, but the patient dies shortly thereafter.

In both cases, patients were very likely, if not certain, to die without operative procedures. Equally certain, the (critical care) surgeon in the second case might find case 1s surgeons neglectful. And they might consider operating on case 2with >95% preoperative mortalitymalpractice.

WHY IS SURGERY DIFFERENT FROM CPR? BENEFICENCE VERSUS AUTONOMY MODELS

Why can surgeons withhold potentially life‐saving surgery, whereas most US physicianssurgeons or internistsdo not (routinely) unilaterally withhold CPR or mechanical ventilation?3 A variety of possible reasons may underlie this asymmetry. First, to compel a surgeon to cut another human against his/her judgment would violate the surgeon's autonomy. But why is the act of cutting viewed differently from the act of intubating and ventilating, or compressing and shocking? The bodily integrity of the patient is violated in both. Nobody would take issue with a surgeon who assesses a 2% chance of survival and so does not offer surgery. Yet physicians struggle often with patients/surrogates who demand CPR/mechanical ventilation for similar prognoses.4 In the United States, CPR has crept into general acceptance (ie, when the only other option is death) as a system‐wide default. In the case of surgery, the judgment of the physician is accepted both by patients and the medical establishment, whereas for CPRwith hypothetically identical consequencesthe patient must opt out. Neither model is right or wrong; but the focus in the balance of decision‐making (paternalism/beneficence vs autonomy) is different.

Albert Jonsen introduced the rule of rescue which suggests that we have an instinctive response to rescue the doomed.5 Surgeons can make the reasonable argument that, in some cases, surgery is only likely to hasten death, and so beneficence requires that they not provide it. The same argument cannot be made for CPR; we do not provide it until patients have already died. And some (albeit small) fraction of the sickest patients survive. For example, 6.4% of those on 2 or more vasopressors who arrest, survive hospitalization.4 Another distinction between CPR and surgery is that when a physician does not withhold CPR for a patient who he thinks is not likely to benefit, he is ordinarily not the party providing the CPR. Most hospitals have teams of individuals who may or may not know the patient and the precise pathophysiology and ethics of their case. So there is greater physical distance (than with surgery) between making the decision and performing the procedure. Moreover, the process of informed consent is temporally proximate and prior to the need for surgery, whereas informed consent is not uniformly obtained a priori, and never after cardiac arrest in a patient who has not previously opted out.

PROBLEMS INHERENT IN BOTH EXTREMES

Viewed through the prism of ethical principlism,1 the ability to withhold surgery may be viewed as beneficence‐strong/autonomy‐weak (BS/AW) whereas prohibiting physicians from withholding CPR when it is only likely to prolong death is beneficence‐weak/autonomy‐strong (BW/AS). These extremes have definable risks that can be named and minimized.

RECOMMENDATIONS FOR MINIMIZING ETHICAL RISKS

Each day, hospitalists interact with a variety of specialists and sub‐specialists to provide consultative or procedural assistance in care of their patients. Physicians have a duty to practice beneficently and to simultaneously respect patients' autonomy.1 Whether to offer a treatment is a function of many variables, but when benefits approach zero, or when risks substantially outweigh benefits, physicians may justifiably withhold therapies without assent or consent of patients.2 The purpose of this article is to explore why it is accepted practice in the United States to permit unilateral withholding of some potentially life‐prolonging treatments (eg, surgery as the paradigm), while it is not common practice for other critical care procedures (eg, cardiopulmonary resuscitation [CPR]). We offer that these examples demonstrate the tension of 2 pillars of medical ethical conduct, namely beneficence and respect of autonomy.1

Consider 2 real cases that demonstrated a juxtaposition of diametrically opposing views of thoughtful, capable surgeons asked to provide life‐saving surgery to critically ill patients.

CASE 1

A 33‐year‐old man, with a history of obesity, presents with mild epigastric pain and hematemesis of a day's duration. Endoscopic evaluation demonstrates a deep gastric ulcer with visible vessel that is injected with epinephrine. He is transferred to the medical intensive care unit (ICU) for monitoring and has an uneventful first 24 hours. On his second hospital day, he develops severe epigastric pain of sudden onset, accompanied by light‐headedness. He is diaphoretic and dyspneic, sitting bolt upright. His body mass index (BMI) is 40 kg/m², and his vital signs are: 130/min, 140/80 mmHg, 30/min, 99.0F. Examination is normal except for severe upper abdominal tenderness, absent bowel sounds, and voluntary guarding. Abdominal computed tomography demonstrates a perforation, free air, and a loculated collection adjacent to the original ulcer. He is treated with 4 liters of crystalloids, oxygen, and an emergent surgical consultation is performed. The examining surgeon confirms the diagnosis of acute perforation, but asserts that his operative risk of mortality, due to obesity, is excessive. He will never get off the ventilator. He advises watchful waiting. The medical intensivist believes the patient will die without surgery; he asks for a second opinion. A more senior colleague assesses the patient and reiterates the first surgeon's opinion. The intensivist begins preparations to transfer the patient to the nearest tertiary care center for a third opinion, when the surgeons reverse themselves. The patient is taken to surgery where the collection is removed, with partial gastrectomy. He is extubated in the recovery room, spends 12 hours in the ICU, and is transferred to the wards where he undergoes an uneventful recovery.

CASE 2

A 50‐year‐old man, with a history of end‐stage alcoholic cirrhosis, presents to the intensive care unit with respiratory embarrassment associated with tense ascites, complicated by pneumococcal pneumonia. He responds to antibiotics but has rapidly reaccumulating ascites, where large volume paracentesis (of 4‐5 liters of transudative fluid) is required every 3 to 4 days to promote weaning trials. On his 20th hospital day, the patient develops fulminant septic shock, and work‐up reveals free air in the abdomen. A Board‐certified critical care surgeon meets with the family and informs them that he is willing to attempt exploratory laparotomy, but that operative mortality exceeds 95%. However, he was willing to try because the alternative otherwise is >99% mortality. The family asks for surgery, which reveals a small perforation, but the patient dies shortly thereafter.

In both cases, patients were very likely, if not certain, to die without operative procedures. Equally certain, the (critical care) surgeon in the second case might find case 1s surgeons neglectful. And they might consider operating on case 2with >95% preoperative mortalitymalpractice.

WHY IS SURGERY DIFFERENT FROM CPR? BENEFICENCE VERSUS AUTONOMY MODELS

Why can surgeons withhold potentially life‐saving surgery, whereas most US physicianssurgeons or internistsdo not (routinely) unilaterally withhold CPR or mechanical ventilation?3 A variety of possible reasons may underlie this asymmetry. First, to compel a surgeon to cut another human against his/her judgment would violate the surgeon's autonomy. But why is the act of cutting viewed differently from the act of intubating and ventilating, or compressing and shocking? The bodily integrity of the patient is violated in both. Nobody would take issue with a surgeon who assesses a 2% chance of survival and so does not offer surgery. Yet physicians struggle often with patients/surrogates who demand CPR/mechanical ventilation for similar prognoses.4 In the United States, CPR has crept into general acceptance (ie, when the only other option is death) as a system‐wide default. In the case of surgery, the judgment of the physician is accepted both by patients and the medical establishment, whereas for CPRwith hypothetically identical consequencesthe patient must opt out. Neither model is right or wrong; but the focus in the balance of decision‐making (paternalism/beneficence vs autonomy) is different.

Albert Jonsen introduced the rule of rescue which suggests that we have an instinctive response to rescue the doomed.5 Surgeons can make the reasonable argument that, in some cases, surgery is only likely to hasten death, and so beneficence requires that they not provide it. The same argument cannot be made for CPR; we do not provide it until patients have already died. And some (albeit small) fraction of the sickest patients survive. For example, 6.4% of those on 2 or more vasopressors who arrest, survive hospitalization.4 Another distinction between CPR and surgery is that when a physician does not withhold CPR for a patient who he thinks is not likely to benefit, he is ordinarily not the party providing the CPR. Most hospitals have teams of individuals who may or may not know the patient and the precise pathophysiology and ethics of their case. So there is greater physical distance (than with surgery) between making the decision and performing the procedure. Moreover, the process of informed consent is temporally proximate and prior to the need for surgery, whereas informed consent is not uniformly obtained a priori, and never after cardiac arrest in a patient who has not previously opted out.

PROBLEMS INHERENT IN BOTH EXTREMES

Viewed through the prism of ethical principlism,1 the ability to withhold surgery may be viewed as beneficence‐strong/autonomy‐weak (BS/AW) whereas prohibiting physicians from withholding CPR when it is only likely to prolong death is beneficence‐weak/autonomy‐strong (BW/AS). These extremes have definable risks that can be named and minimized.

RECOMMENDATIONS FOR MINIMIZING ETHICAL RISKS

Each day, hospitalists interact with a variety of specialists and sub‐specialists to provide consultative or procedural assistance in care of their patients. Physicians have a duty to practice beneficently and to simultaneously respect patients' autonomy.1 Whether to offer a treatment is a function of many variables, but when benefits approach zero, or when risks substantially outweigh benefits, physicians may justifiably withhold therapies without assent or consent of patients.2 The purpose of this article is to explore why it is accepted practice in the United States to permit unilateral withholding of some potentially life‐prolonging treatments (eg, surgery as the paradigm), while it is not common practice for other critical care procedures (eg, cardiopulmonary resuscitation [CPR]). We offer that these examples demonstrate the tension of 2 pillars of medical ethical conduct, namely beneficence and respect of autonomy.1

Consider 2 real cases that demonstrated a juxtaposition of diametrically opposing views of thoughtful, capable surgeons asked to provide life‐saving surgery to critically ill patients.

CASE 1

A 33‐year‐old man, with a history of obesity, presents with mild epigastric pain and hematemesis of a day's duration. Endoscopic evaluation demonstrates a deep gastric ulcer with visible vessel that is injected with epinephrine. He is transferred to the medical intensive care unit (ICU) for monitoring and has an uneventful first 24 hours. On his second hospital day, he develops severe epigastric pain of sudden onset, accompanied by light‐headedness. He is diaphoretic and dyspneic, sitting bolt upright. His body mass index (BMI) is 40 kg/m², and his vital signs are: 130/min, 140/80 mmHg, 30/min, 99.0F. Examination is normal except for severe upper abdominal tenderness, absent bowel sounds, and voluntary guarding. Abdominal computed tomography demonstrates a perforation, free air, and a loculated collection adjacent to the original ulcer. He is treated with 4 liters of crystalloids, oxygen, and an emergent surgical consultation is performed. The examining surgeon confirms the diagnosis of acute perforation, but asserts that his operative risk of mortality, due to obesity, is excessive. He will never get off the ventilator. He advises watchful waiting. The medical intensivist believes the patient will die without surgery; he asks for a second opinion. A more senior colleague assesses the patient and reiterates the first surgeon's opinion. The intensivist begins preparations to transfer the patient to the nearest tertiary care center for a third opinion, when the surgeons reverse themselves. The patient is taken to surgery where the collection is removed, with partial gastrectomy. He is extubated in the recovery room, spends 12 hours in the ICU, and is transferred to the wards where he undergoes an uneventful recovery.

CASE 2

A 50‐year‐old man, with a history of end‐stage alcoholic cirrhosis, presents to the intensive care unit with respiratory embarrassment associated with tense ascites, complicated by pneumococcal pneumonia. He responds to antibiotics but has rapidly reaccumulating ascites, where large volume paracentesis (of 4‐5 liters of transudative fluid) is required every 3 to 4 days to promote weaning trials. On his 20th hospital day, the patient develops fulminant septic shock, and work‐up reveals free air in the abdomen. A Board‐certified critical care surgeon meets with the family and informs them that he is willing to attempt exploratory laparotomy, but that operative mortality exceeds 95%. However, he was willing to try because the alternative otherwise is >99% mortality. The family asks for surgery, which reveals a small perforation, but the patient dies shortly thereafter.

In both cases, patients were very likely, if not certain, to die without operative procedures. Equally certain, the (critical care) surgeon in the second case might find case 1s surgeons neglectful. And they might consider operating on case 2with >95% preoperative mortalitymalpractice.

WHY IS SURGERY DIFFERENT FROM CPR? BENEFICENCE VERSUS AUTONOMY MODELS

Why can surgeons withhold potentially life‐saving surgery, whereas most US physicianssurgeons or internistsdo not (routinely) unilaterally withhold CPR or mechanical ventilation?3 A variety of possible reasons may underlie this asymmetry. First, to compel a surgeon to cut another human against his/her judgment would violate the surgeon's autonomy. But why is the act of cutting viewed differently from the act of intubating and ventilating, or compressing and shocking? The bodily integrity of the patient is violated in both. Nobody would take issue with a surgeon who assesses a 2% chance of survival and so does not offer surgery. Yet physicians struggle often with patients/surrogates who demand CPR/mechanical ventilation for similar prognoses.4 In the United States, CPR has crept into general acceptance (ie, when the only other option is death) as a system‐wide default. In the case of surgery, the judgment of the physician is accepted both by patients and the medical establishment, whereas for CPRwith hypothetically identical consequencesthe patient must opt out. Neither model is right or wrong; but the focus in the balance of decision‐making (paternalism/beneficence vs autonomy) is different.

Albert Jonsen introduced the rule of rescue which suggests that we have an instinctive response to rescue the doomed.5 Surgeons can make the reasonable argument that, in some cases, surgery is only likely to hasten death, and so beneficence requires that they not provide it. The same argument cannot be made for CPR; we do not provide it until patients have already died. And some (albeit small) fraction of the sickest patients survive. For example, 6.4% of those on 2 or more vasopressors who arrest, survive hospitalization.4 Another distinction between CPR and surgery is that when a physician does not withhold CPR for a patient who he thinks is not likely to benefit, he is ordinarily not the party providing the CPR. Most hospitals have teams of individuals who may or may not know the patient and the precise pathophysiology and ethics of their case. So there is greater physical distance (than with surgery) between making the decision and performing the procedure. Moreover, the process of informed consent is temporally proximate and prior to the need for surgery, whereas informed consent is not uniformly obtained a priori, and never after cardiac arrest in a patient who has not previously opted out.

PROBLEMS INHERENT IN BOTH EXTREMES

Viewed through the prism of ethical principlism,1 the ability to withhold surgery may be viewed as beneficence‐strong/autonomy‐weak (BS/AW) whereas prohibiting physicians from withholding CPR when it is only likely to prolong death is beneficence‐weak/autonomy‐strong (BW/AS). These extremes have definable risks that can be named and minimized.

RECOMMENDATIONS FOR MINIMIZING ETHICAL RISKS

Patients who are hospitalized for acute medical illness are at an increased risk of developing venous thromboembolism (VTE), which comprises deep‐vein thrombosis (DVT) and pulmonary embolism (PE).13 In a recent real‐world study of 158,325 US medical patients by Spyropoulos et al,4 4.0% of patients developed DVT, 1.5% developed PE, and 0.2% developed both DVT and PE. Furthermore, results from a population‐based case‐control study indicate that hospitalization for medical illness accounted for a proportion of VTE events similar to that of hospitalization for surgery (22% and 24%, respectively).5

Thromboprophylaxis reduces VTE incidence in at‐risk medical patients and is recommended according to evidence‐based guidelines from the American College of Chest Physicians (ACCP).1 The ACCP guidelines advocate that acutely ill medical patients admitted to the hospital with congestive heart failure (CHF) or severe lung disease/chronic obstructive pulmonary disease (COPD) or those who are confined to bed and have one or more additional risk factors (including active cancer, previous VTE, sepsis, acute neurologic disease, or inflammatory bowel disease) receive pharmacological prophylaxis with lowmolecular weight heparin (LMWH), low‐dose unfractionated heparin (UFH), or fondaparinux.1 Although guidelines provide recommendations for the duration of prophylaxis after major orthopedic surgery, such recommendations are unavailable for medical patients. In clinical trials of acutely ill medical patients, prophylaxis regimens found to be effective were provided for a duration of hospitalization of 6‐14 days.68 The mean length of hospital stay for medical illnesses is decreasing and is currently shorter than 6‐14 days.9, 10

In clinical practice, the duration of VTE risk during and after hospitalization is not well understood in medical patients, particularly in the context of shortening hospital stays. Such information could, however, provide insight into whether current thromboprophylaxis practices reflect real‐world need. To gain a greater understanding of the period during which patients are at risk of VTE, this retrospective, observational study assessed the incidence and time course of symptomatic VTE events during and after hospitalization in a large population of US medical patients.

METHODS

Analysis

The risk of VTE was estimated across an evaluation period of 180 days by measuring VTE occurrence and person‐time exposure. Inpatient VTE occurrence was defined as any nonprimary diagnosis of DVT and/or PE during the at‐risk hospitalization. VTE after discharge was defined as an ICD‐9‐CM diagnosis code, whether primary or secondary, for DVT or PE in the evaluation period during an emergency room or inpatient admission, or on an outpatient claim with 1 or more of the following confirmatory events: an emergency room or inpatient admission for VTE within 2 days of the outpatient diagnosis; a prescription claim for enoxaparin, fondaparinux, or UFH within 15 days after diagnosis; or a prescription claim for warfarin within 15 days after diagnosis and no evidence of atrial fibrillation or atrial flutter in the 6 months before the outpatient diagnosis for DVT or PE. Person‐time exposure was measured as the number of days from the hospital admission date to the first occurrence of VTE, or censoring at a subsequent at‐risk hospitalization, death, or 180 days after admission.

Cumulative risk of VTE over the 180‐day evaluation period was calculated by the Kaplan‐Meier product limit method of survival analysis and displayed for deciles of cumulative risk at 180 days after the hospital admission date. The risk of VTE at each point of time during the evaluation period (the hazard function) was first calculated on a daily basis and then smoothed via LOESS regression, a locally weighted regression procedure.

RESULTS

Patient Demographics

A total of 11,139 medical patients were included in the analysis (Figure 1), with a mean standard deviation (SD) age of 67.6 13.9 years, and 51.6% were women (Table 1). Of the reasons for admission to the hospital, 51.5% of patients were admitted for severe lung disease/COPD, 20.1% were admitted for cancer, 15.3% were admitted for CHF, and 13.1% were admitted for severe infectious disease. Most patients were treated in an urban hospital (87.5%), in a hospital without teaching status (87.9%), and in the South Census region (74.1%). The majority of patients were treated in medium‐sized to large care facilities. Risk factors for VTE during the baseline period included hospitalization for a medical condition with a high risk for VTE (75.6%), a prior at‐risk hospitalization (18.6%), cancer therapy (10.0% of all medical patients combined and 18.5% of cancer patients), trauma (9.2%), and previous VTE (4.3%).

Figure 1

Table 1.Summary of Patient Demoghraphics and Characteristics

Characteristic	Medical Patients (N = 11,139)
NOTE: All values are presented as no. (%) unless indicated otherwise. 3 Abbreviations: CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; SD, standard deviation. * This study used only the Commercial and Medicare supplemental databases.
Gender
Men	5389 (48.4)
Women	5750 (51.6)
Reason for hospitalization
Cancer	2243 (20.1)
CHF	1705 (15.3)
Severe lung disease/COPD	5736 (51.5)
Severe infectious disease	1455 (13.1)
Age group, years
1834	230 (2.1)
3544	442 (4.0)
4554	1188 (10.7)
5564	2644 (23.7)
6574	2657 (23.9)
7584	2969 (26.7)
85 years	1009 (9.1)
Median age SD, years	67.6 13.9
Primary payer*
Medicare	6819 (61.2)
Commercial	4320 (38.8)
Geographical area
Northeast	122 (1.1)
North Central	2649 (23.8)
South	8258 (74.1)
West	110 (1.0)
Urban location	9743 (87.5)
Teaching hospital	1345 (12.1)
Licensed bed size
1199	1621 (14.6)
200299	2869 (25.8)
300499	4005 (36.0)
500	2644 (23.7)

Time Course of VTE Risk and Hazard Function

Overall, there were 366 symptomatic VTE events, representing a VTE rate of 3.3%. These events comprised 241 DVT‐only events, 98 PE‐only events, and 27 events with evidence of both DVT and PE. In total, 43.4% of events occurred during hospitalization (Figure 2). The VTE rate was 5.7% in patients with cancer (30.5% of events occurring in hospital), 4.3% with infectious disease (61.9% in hospital), 3.1% with CHF (54.7% in hospital), and 2.1% with severe lung disease/COPD (42.6% in hospital). The highest number of VTE events, 97 events (62 DVT only, 26 PE only, and 9 events both DVT and PE), occurred in the first 9 days after the hospital admission date, of which 87.6% were during hospitalization. During days 10‐19, there were 82 VTE events (50 DVT only, 24 PE only, and 8 both DVT and PE), 70.7% of which occurred in the hospital. Over the following 10‐day periods, VTE incidence gradually declined (Figure 2) and fluctuated at a background level of 4‐8 events during each 10‐day interval from 120 to 180 days.

Figure 2

The cumulative probability of VTE among all patients was 0.035 (Figure 3A). Half of the VTE risk had accumulated by day 23, and 75% had accumulated by day 71. By day 30, the proportion of cumulative risk was 52.6% overall, and ranged from 41.9% with cancer to 72.9% with infectious disease (Figure 3).

Figure 3

The VTE hazard peaked at approximately 1.05 VTE events per 1000 person‐days on day 8 after the hospital admission date overall (Figure 4A). The cumulative hazard at the peak day was 18.2% of the total VTE hazard over the 180‐day evaluation period. The hazard peak ranged from day 7 in patients with severe lung disease/COPD to day 12 in patients with infectious disease (Figure 4B). The cumulative hazard at the peak day was 39.7% for patients with infectious disease, 29.2% for patients with CHF, and approximately 19% for cancer or severe lung disease/COPD. After the peak risk day, the VTE hazard function decreased until the curve reached an inflection point, at day 28, when the cumulative risk was 51.8% (Figure 4A). After the inflection point, the VTE hazard increased to 0.3 VTE events per 1000 person‐days at approximately day 40‐45 and then decreased to <0.2 events per 1000 person‐days. The timing of the inflection varied by approximately 1 week across the medical illnesses (ranging from day 25 for severe lung disease/COPD to day 33 for CHF), with the cumulative risk at the inflection point ranging from 41.9% with cancer to 72.9% with infectious disease.

Figure 4

DISCUSSION

The results from this large, real‐world study provide new insights into the duration of risk of symptomatic VTE in medical patients and demonstrate that the number of VTE events was highest during days 0‐19, with the peak of VTE hazard at day 8. Half of the total 180‐day cumulative risk had been incurred by day 23 after hospital admission, and the period of greatest increased risk extended up to at least 30 days. Importantly, more than half of VTE events occurred after discharge (56.6%). A particularly high proportion of VTE events (69.5%) had occurred after discharge in patients with cancer. Although it was assumed that most VTE events that could be reasonably attributed to an at‐risk hospitalization would occur within 90 days as shown previously,4, 11 the 180‐day evaluation period was used to examine whether there was a prolonged period of continually diminished VTE risk from 90 to 180 days. Thus, events occurring within the later portions of the evaluation period may or may not have been attributable to the index hospitalization, potentially reflecting a background rate of VTE as noted above. Although these events are included in our estimate of the 180‐day cumulative risk of VTE, interpretation of the study results excluding such events is possible by examining the cumulative risk that had been incurred at each time point during the evaluation period.

Few other studies have assessed the duration of VTE risk in hospitalized medical patients. In a study by Spyropoulos et al,4 the median time to a DVT and/or PE event was 74 days, ranging from 62 days in severe infectious disease to 126 days in CHF. In another observational study that included patients who had recently been hospitalized but had not undergone surgery, 66.9% of patients who experienced DVT and/or PE events were diagnosed with DVT and/or PE within the first month after hospital discharge; 19.9% between months 1 and 2, and 13.2% between months 2 and 3.12

Fewer than half of the patients in the present study received thromboprophylaxis, which is consistent with other studies demonstrating the low prophylaxis rates in medical inpatients.9, 1315 In a recently published US study of discharge records that included 22,455 medical inpatients, prophylaxis rates were 59.4% in patients with CHF, 52.3% with cancer, 45.8% with severe lung disease/COPD, and 40.4% with infectious disease.14 Fewer than 10% of patients in the present study received prophylaxis after discharge, a result that is consistent with other studies.4, 9

The effect of extended prophylaxis in acutely ill medical patients with the LMWH enoxaparin beyond 6‐14 days has been investigated in the EXCLAIM study.16 This trial included approximately 5800 acutely ill medical patients at significant risk of developing VTE due to a recent reduction in mobility. Patients in the extended prophylaxis group had a lower risk of VTE (2.5% vs 4% for placebo; absolute risk reduction 1.5% [95.8% confidence interval 2.54% to 0.52%]), but had increased major bleeding events (0.8% vs 0.3% for placebo; absolute risk difference favoring placebo, 0.51% [95% confidence interval, 0.12% to 0.89%]). The patient populations with most benefit from an additional 28 days prophylaxis with enoxaparin, in addition to the usual short‐term prophylaxis of 10 days, were patients with restricted mobility (level 1; total bed rest/sedentary), elderly patients (age >75 years), and women. A limitation of the EXCLAIM trial is that estimates of efficacy and safety are difficult to interpret: after an interim analysis of adjudicated efficacy and safety outcomes, amendments were made to the original study protocol by changing eligibility criteria for patients with level 2 immobility (level 1 with bathroom privileges).16

The optimal duration of prophylaxis for medical patients has not been determined; prophylaxis is generally administered to at‐risk medical patients for the duration of hospitalization. In the current study, mean length of stay was 5.3 5.3 days overall. As hospital stays shorten, many medical patients who are prescribed inpatient prophylaxis alone are unlikely to receive the standard 6‐14 days of prophylaxis shown to be effective in clinical trials.68 Furthermore, the extended period of VTE risk in the present study and the finding that 56.6% of events occurred after discharge also suggest that current practices for inpatient prophylaxis alone may need to be evaluated.

This study reports real‐world data from a large, well‐defined population and obtains the incidence of symptomatic VTE events. Even though certain demographic data deviate from the national averagefor example, 74.1% of patients were treated in the South Census region, whereas this region is served by 37.6% of US hospitals17; 87.5% of hospitals had an urban location (compared with 60.1% of US hospitals18), and 85.4% of hospitals had a licensed bed size of at least 200 beds (compared with 28.2% of US hospitals, with the average US hospital having fewer than 100 beds19)these data may be beneficial in guiding policy and health care strategies for gaining understanding of the duration of risk for VTE.

Limitations of the study include characterization of the VTE risk period through examination of the cumulative risk and hazard of VTE across time, as the actual VTE risk period cannot be determined with exact precision. We used ICD‐9‐CM diagnosis coding to identify VTE. Since many cases of PE are asymptomatic and detected at autopsy,20 our approach may have missed such cases, as they would not have been recorded within the database. Furthermore, validation studies suggest that suboptimal specificity exists for ICD‐9‐CM diagnosis codes used to identify VTE.21 In an attempt to improve the specificity of our VTE identification algorithm, we required that post‐discharge VTE was recorded either during an emergency room or subsequent inpatient admission (which would be indicative of acute care for VTE) or on an outpatient claim with subsequent evidence of treatment for VTE. The true sensitivity and specificity of the VTE identification algorithms used for this study remain unknown, however, so the study findings should be interpreted in light of this limitation. The databases used for the analysis may not be representative of the US population as a whole; for example, this study used claims data from commercial and Medicare supplemental databases, which do not include Medicaid patients. Another limitation was that outpatient mechanical prophylaxis, such as graded compression stockings, was not captured due to over‐the‐counter availability. In addition, appropriateness of prophylaxis was not determined in this study, because these data could not be obtained from the claims database used. Further studies are warranted to obtain information on the incidence of VTE after hospitalization for medical illness in patients who received appropriate prophylaxis during hospitalization.

Finally, all dosages of a pharmacological agent were considered prophylactic only if a VTE event did not occur, with the exception of warfarin; any dose of warfarin was considered for prophylaxis, regardless of a VTE diagnosis. Warfarin may be used for purposes other than VTE prophylaxis (eg, prophylaxis for a thromboembolic cerebrovascular accident). The data source does not allow for identifying the exact reason for anticoagulation therapy with warfarin. Nonetheless, warfarin therapy will confer a decreased risk of VTE regardless of its purpose.

Results from this large cohort of medical patients indicate that symptomatic VTE risk is highest within the first 19 days after hospital admission (a period that may encompass both the duration of hospitalization as well as the period after discharge) with a considerable risk of VTE extending into the period after discharge. Receiving appropriate prophylaxis in‐hospital remains of great importance to prevent inpatient and likely post‐discharge VTE in patients with acute medical illness. In addition, given the time course of VTE events, with VTE incidence peaking at 8 days but with increased risk extending to 30 days, and the number of out‐of‐hospital VTE events incurred, the results of this study suggest that future research is warranted to investigate the risks and benefits of improving thromboprophylaxis practices in the period after hospitalization.

Patients who are hospitalized for acute medical illness are at an increased risk of developing venous thromboembolism (VTE), which comprises deep‐vein thrombosis (DVT) and pulmonary embolism (PE).13 In a recent real‐world study of 158,325 US medical patients by Spyropoulos et al,4 4.0% of patients developed DVT, 1.5% developed PE, and 0.2% developed both DVT and PE. Furthermore, results from a population‐based case‐control study indicate that hospitalization for medical illness accounted for a proportion of VTE events similar to that of hospitalization for surgery (22% and 24%, respectively).5

Thromboprophylaxis reduces VTE incidence in at‐risk medical patients and is recommended according to evidence‐based guidelines from the American College of Chest Physicians (ACCP).1 The ACCP guidelines advocate that acutely ill medical patients admitted to the hospital with congestive heart failure (CHF) or severe lung disease/chronic obstructive pulmonary disease (COPD) or those who are confined to bed and have one or more additional risk factors (including active cancer, previous VTE, sepsis, acute neurologic disease, or inflammatory bowel disease) receive pharmacological prophylaxis with lowmolecular weight heparin (LMWH), low‐dose unfractionated heparin (UFH), or fondaparinux.1 Although guidelines provide recommendations for the duration of prophylaxis after major orthopedic surgery, such recommendations are unavailable for medical patients. In clinical trials of acutely ill medical patients, prophylaxis regimens found to be effective were provided for a duration of hospitalization of 6‐14 days.68 The mean length of hospital stay for medical illnesses is decreasing and is currently shorter than 6‐14 days.9, 10

In clinical practice, the duration of VTE risk during and after hospitalization is not well understood in medical patients, particularly in the context of shortening hospital stays. Such information could, however, provide insight into whether current thromboprophylaxis practices reflect real‐world need. To gain a greater understanding of the period during which patients are at risk of VTE, this retrospective, observational study assessed the incidence and time course of symptomatic VTE events during and after hospitalization in a large population of US medical patients.

METHODS

Analysis

The risk of VTE was estimated across an evaluation period of 180 days by measuring VTE occurrence and person‐time exposure. Inpatient VTE occurrence was defined as any nonprimary diagnosis of DVT and/or PE during the at‐risk hospitalization. VTE after discharge was defined as an ICD‐9‐CM diagnosis code, whether primary or secondary, for DVT or PE in the evaluation period during an emergency room or inpatient admission, or on an outpatient claim with 1 or more of the following confirmatory events: an emergency room or inpatient admission for VTE within 2 days of the outpatient diagnosis; a prescription claim for enoxaparin, fondaparinux, or UFH within 15 days after diagnosis; or a prescription claim for warfarin within 15 days after diagnosis and no evidence of atrial fibrillation or atrial flutter in the 6 months before the outpatient diagnosis for DVT or PE. Person‐time exposure was measured as the number of days from the hospital admission date to the first occurrence of VTE, or censoring at a subsequent at‐risk hospitalization, death, or 180 days after admission.

Cumulative risk of VTE over the 180‐day evaluation period was calculated by the Kaplan‐Meier product limit method of survival analysis and displayed for deciles of cumulative risk at 180 days after the hospital admission date. The risk of VTE at each point of time during the evaluation period (the hazard function) was first calculated on a daily basis and then smoothed via LOESS regression, a locally weighted regression procedure.

RESULTS

Patient Demographics

A total of 11,139 medical patients were included in the analysis (Figure 1), with a mean standard deviation (SD) age of 67.6 13.9 years, and 51.6% were women (Table 1). Of the reasons for admission to the hospital, 51.5% of patients were admitted for severe lung disease/COPD, 20.1% were admitted for cancer, 15.3% were admitted for CHF, and 13.1% were admitted for severe infectious disease. Most patients were treated in an urban hospital (87.5%), in a hospital without teaching status (87.9%), and in the South Census region (74.1%). The majority of patients were treated in medium‐sized to large care facilities. Risk factors for VTE during the baseline period included hospitalization for a medical condition with a high risk for VTE (75.6%), a prior at‐risk hospitalization (18.6%), cancer therapy (10.0% of all medical patients combined and 18.5% of cancer patients), trauma (9.2%), and previous VTE (4.3%).

Figure 1

Table 1.Summary of Patient Demoghraphics and Characteristics

Characteristic	Medical Patients (N = 11,139)
NOTE: All values are presented as no. (%) unless indicated otherwise. 3 Abbreviations: CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; SD, standard deviation. * This study used only the Commercial and Medicare supplemental databases.
Gender
Men	5389 (48.4)
Women	5750 (51.6)
Reason for hospitalization
Cancer	2243 (20.1)
CHF	1705 (15.3)
Severe lung disease/COPD	5736 (51.5)
Severe infectious disease	1455 (13.1)
Age group, years
1834	230 (2.1)
3544	442 (4.0)
4554	1188 (10.7)
5564	2644 (23.7)
6574	2657 (23.9)
7584	2969 (26.7)
85 years	1009 (9.1)
Median age SD, years	67.6 13.9
Primary payer*
Medicare	6819 (61.2)
Commercial	4320 (38.8)
Geographical area
Northeast	122 (1.1)
North Central	2649 (23.8)
South	8258 (74.1)
West	110 (1.0)
Urban location	9743 (87.5)
Teaching hospital	1345 (12.1)
Licensed bed size
1199	1621 (14.6)
200299	2869 (25.8)
300499	4005 (36.0)
500	2644 (23.7)

Time Course of VTE Risk and Hazard Function

Overall, there were 366 symptomatic VTE events, representing a VTE rate of 3.3%. These events comprised 241 DVT‐only events, 98 PE‐only events, and 27 events with evidence of both DVT and PE. In total, 43.4% of events occurred during hospitalization (Figure 2). The VTE rate was 5.7% in patients with cancer (30.5% of events occurring in hospital), 4.3% with infectious disease (61.9% in hospital), 3.1% with CHF (54.7% in hospital), and 2.1% with severe lung disease/COPD (42.6% in hospital). The highest number of VTE events, 97 events (62 DVT only, 26 PE only, and 9 events both DVT and PE), occurred in the first 9 days after the hospital admission date, of which 87.6% were during hospitalization. During days 10‐19, there were 82 VTE events (50 DVT only, 24 PE only, and 8 both DVT and PE), 70.7% of which occurred in the hospital. Over the following 10‐day periods, VTE incidence gradually declined (Figure 2) and fluctuated at a background level of 4‐8 events during each 10‐day interval from 120 to 180 days.

Figure 2

The cumulative probability of VTE among all patients was 0.035 (Figure 3A). Half of the VTE risk had accumulated by day 23, and 75% had accumulated by day 71. By day 30, the proportion of cumulative risk was 52.6% overall, and ranged from 41.9% with cancer to 72.9% with infectious disease (Figure 3).

Figure 3

The VTE hazard peaked at approximately 1.05 VTE events per 1000 person‐days on day 8 after the hospital admission date overall (Figure 4A). The cumulative hazard at the peak day was 18.2% of the total VTE hazard over the 180‐day evaluation period. The hazard peak ranged from day 7 in patients with severe lung disease/COPD to day 12 in patients with infectious disease (Figure 4B). The cumulative hazard at the peak day was 39.7% for patients with infectious disease, 29.2% for patients with CHF, and approximately 19% for cancer or severe lung disease/COPD. After the peak risk day, the VTE hazard function decreased until the curve reached an inflection point, at day 28, when the cumulative risk was 51.8% (Figure 4A). After the inflection point, the VTE hazard increased to 0.3 VTE events per 1000 person‐days at approximately day 40‐45 and then decreased to <0.2 events per 1000 person‐days. The timing of the inflection varied by approximately 1 week across the medical illnesses (ranging from day 25 for severe lung disease/COPD to day 33 for CHF), with the cumulative risk at the inflection point ranging from 41.9% with cancer to 72.9% with infectious disease.

Figure 4

DISCUSSION

The results from this large, real‐world study provide new insights into the duration of risk of symptomatic VTE in medical patients and demonstrate that the number of VTE events was highest during days 0‐19, with the peak of VTE hazard at day 8. Half of the total 180‐day cumulative risk had been incurred by day 23 after hospital admission, and the period of greatest increased risk extended up to at least 30 days. Importantly, more than half of VTE events occurred after discharge (56.6%). A particularly high proportion of VTE events (69.5%) had occurred after discharge in patients with cancer. Although it was assumed that most VTE events that could be reasonably attributed to an at‐risk hospitalization would occur within 90 days as shown previously,4, 11 the 180‐day evaluation period was used to examine whether there was a prolonged period of continually diminished VTE risk from 90 to 180 days. Thus, events occurring within the later portions of the evaluation period may or may not have been attributable to the index hospitalization, potentially reflecting a background rate of VTE as noted above. Although these events are included in our estimate of the 180‐day cumulative risk of VTE, interpretation of the study results excluding such events is possible by examining the cumulative risk that had been incurred at each time point during the evaluation period.

Few other studies have assessed the duration of VTE risk in hospitalized medical patients. In a study by Spyropoulos et al,4 the median time to a DVT and/or PE event was 74 days, ranging from 62 days in severe infectious disease to 126 days in CHF. In another observational study that included patients who had recently been hospitalized but had not undergone surgery, 66.9% of patients who experienced DVT and/or PE events were diagnosed with DVT and/or PE within the first month after hospital discharge; 19.9% between months 1 and 2, and 13.2% between months 2 and 3.12

Fewer than half of the patients in the present study received thromboprophylaxis, which is consistent with other studies demonstrating the low prophylaxis rates in medical inpatients.9, 1315 In a recently published US study of discharge records that included 22,455 medical inpatients, prophylaxis rates were 59.4% in patients with CHF, 52.3% with cancer, 45.8% with severe lung disease/COPD, and 40.4% with infectious disease.14 Fewer than 10% of patients in the present study received prophylaxis after discharge, a result that is consistent with other studies.4, 9

The effect of extended prophylaxis in acutely ill medical patients with the LMWH enoxaparin beyond 6‐14 days has been investigated in the EXCLAIM study.16 This trial included approximately 5800 acutely ill medical patients at significant risk of developing VTE due to a recent reduction in mobility. Patients in the extended prophylaxis group had a lower risk of VTE (2.5% vs 4% for placebo; absolute risk reduction 1.5% [95.8% confidence interval 2.54% to 0.52%]), but had increased major bleeding events (0.8% vs 0.3% for placebo; absolute risk difference favoring placebo, 0.51% [95% confidence interval, 0.12% to 0.89%]). The patient populations with most benefit from an additional 28 days prophylaxis with enoxaparin, in addition to the usual short‐term prophylaxis of 10 days, were patients with restricted mobility (level 1; total bed rest/sedentary), elderly patients (age >75 years), and women. A limitation of the EXCLAIM trial is that estimates of efficacy and safety are difficult to interpret: after an interim analysis of adjudicated efficacy and safety outcomes, amendments were made to the original study protocol by changing eligibility criteria for patients with level 2 immobility (level 1 with bathroom privileges).16

The optimal duration of prophylaxis for medical patients has not been determined; prophylaxis is generally administered to at‐risk medical patients for the duration of hospitalization. In the current study, mean length of stay was 5.3 5.3 days overall. As hospital stays shorten, many medical patients who are prescribed inpatient prophylaxis alone are unlikely to receive the standard 6‐14 days of prophylaxis shown to be effective in clinical trials.68 Furthermore, the extended period of VTE risk in the present study and the finding that 56.6% of events occurred after discharge also suggest that current practices for inpatient prophylaxis alone may need to be evaluated.

This study reports real‐world data from a large, well‐defined population and obtains the incidence of symptomatic VTE events. Even though certain demographic data deviate from the national averagefor example, 74.1% of patients were treated in the South Census region, whereas this region is served by 37.6% of US hospitals17; 87.5% of hospitals had an urban location (compared with 60.1% of US hospitals18), and 85.4% of hospitals had a licensed bed size of at least 200 beds (compared with 28.2% of US hospitals, with the average US hospital having fewer than 100 beds19)these data may be beneficial in guiding policy and health care strategies for gaining understanding of the duration of risk for VTE.

Limitations of the study include characterization of the VTE risk period through examination of the cumulative risk and hazard of VTE across time, as the actual VTE risk period cannot be determined with exact precision. We used ICD‐9‐CM diagnosis coding to identify VTE. Since many cases of PE are asymptomatic and detected at autopsy,20 our approach may have missed such cases, as they would not have been recorded within the database. Furthermore, validation studies suggest that suboptimal specificity exists for ICD‐9‐CM diagnosis codes used to identify VTE.21 In an attempt to improve the specificity of our VTE identification algorithm, we required that post‐discharge VTE was recorded either during an emergency room or subsequent inpatient admission (which would be indicative of acute care for VTE) or on an outpatient claim with subsequent evidence of treatment for VTE. The true sensitivity and specificity of the VTE identification algorithms used for this study remain unknown, however, so the study findings should be interpreted in light of this limitation. The databases used for the analysis may not be representative of the US population as a whole; for example, this study used claims data from commercial and Medicare supplemental databases, which do not include Medicaid patients. Another limitation was that outpatient mechanical prophylaxis, such as graded compression stockings, was not captured due to over‐the‐counter availability. In addition, appropriateness of prophylaxis was not determined in this study, because these data could not be obtained from the claims database used. Further studies are warranted to obtain information on the incidence of VTE after hospitalization for medical illness in patients who received appropriate prophylaxis during hospitalization.

Finally, all dosages of a pharmacological agent were considered prophylactic only if a VTE event did not occur, with the exception of warfarin; any dose of warfarin was considered for prophylaxis, regardless of a VTE diagnosis. Warfarin may be used for purposes other than VTE prophylaxis (eg, prophylaxis for a thromboembolic cerebrovascular accident). The data source does not allow for identifying the exact reason for anticoagulation therapy with warfarin. Nonetheless, warfarin therapy will confer a decreased risk of VTE regardless of its purpose.

Results from this large cohort of medical patients indicate that symptomatic VTE risk is highest within the first 19 days after hospital admission (a period that may encompass both the duration of hospitalization as well as the period after discharge) with a considerable risk of VTE extending into the period after discharge. Receiving appropriate prophylaxis in‐hospital remains of great importance to prevent inpatient and likely post‐discharge VTE in patients with acute medical illness. In addition, given the time course of VTE events, with VTE incidence peaking at 8 days but with increased risk extending to 30 days, and the number of out‐of‐hospital VTE events incurred, the results of this study suggest that future research is warranted to investigate the risks and benefits of improving thromboprophylaxis practices in the period after hospitalization.

Patients who are hospitalized for acute medical illness are at an increased risk of developing venous thromboembolism (VTE), which comprises deep‐vein thrombosis (DVT) and pulmonary embolism (PE).13 In a recent real‐world study of 158,325 US medical patients by Spyropoulos et al,4 4.0% of patients developed DVT, 1.5% developed PE, and 0.2% developed both DVT and PE. Furthermore, results from a population‐based case‐control study indicate that hospitalization for medical illness accounted for a proportion of VTE events similar to that of hospitalization for surgery (22% and 24%, respectively).5

Thromboprophylaxis reduces VTE incidence in at‐risk medical patients and is recommended according to evidence‐based guidelines from the American College of Chest Physicians (ACCP).1 The ACCP guidelines advocate that acutely ill medical patients admitted to the hospital with congestive heart failure (CHF) or severe lung disease/chronic obstructive pulmonary disease (COPD) or those who are confined to bed and have one or more additional risk factors (including active cancer, previous VTE, sepsis, acute neurologic disease, or inflammatory bowel disease) receive pharmacological prophylaxis with lowmolecular weight heparin (LMWH), low‐dose unfractionated heparin (UFH), or fondaparinux.1 Although guidelines provide recommendations for the duration of prophylaxis after major orthopedic surgery, such recommendations are unavailable for medical patients. In clinical trials of acutely ill medical patients, prophylaxis regimens found to be effective were provided for a duration of hospitalization of 6‐14 days.68 The mean length of hospital stay for medical illnesses is decreasing and is currently shorter than 6‐14 days.9, 10

In clinical practice, the duration of VTE risk during and after hospitalization is not well understood in medical patients, particularly in the context of shortening hospital stays. Such information could, however, provide insight into whether current thromboprophylaxis practices reflect real‐world need. To gain a greater understanding of the period during which patients are at risk of VTE, this retrospective, observational study assessed the incidence and time course of symptomatic VTE events during and after hospitalization in a large population of US medical patients.

METHODS

Analysis

The risk of VTE was estimated across an evaluation period of 180 days by measuring VTE occurrence and person‐time exposure. Inpatient VTE occurrence was defined as any nonprimary diagnosis of DVT and/or PE during the at‐risk hospitalization. VTE after discharge was defined as an ICD‐9‐CM diagnosis code, whether primary or secondary, for DVT or PE in the evaluation period during an emergency room or inpatient admission, or on an outpatient claim with 1 or more of the following confirmatory events: an emergency room or inpatient admission for VTE within 2 days of the outpatient diagnosis; a prescription claim for enoxaparin, fondaparinux, or UFH within 15 days after diagnosis; or a prescription claim for warfarin within 15 days after diagnosis and no evidence of atrial fibrillation or atrial flutter in the 6 months before the outpatient diagnosis for DVT or PE. Person‐time exposure was measured as the number of days from the hospital admission date to the first occurrence of VTE, or censoring at a subsequent at‐risk hospitalization, death, or 180 days after admission.

Cumulative risk of VTE over the 180‐day evaluation period was calculated by the Kaplan‐Meier product limit method of survival analysis and displayed for deciles of cumulative risk at 180 days after the hospital admission date. The risk of VTE at each point of time during the evaluation period (the hazard function) was first calculated on a daily basis and then smoothed via LOESS regression, a locally weighted regression procedure.

RESULTS

Patient Demographics

A total of 11,139 medical patients were included in the analysis (Figure 1), with a mean standard deviation (SD) age of 67.6 13.9 years, and 51.6% were women (Table 1). Of the reasons for admission to the hospital, 51.5% of patients were admitted for severe lung disease/COPD, 20.1% were admitted for cancer, 15.3% were admitted for CHF, and 13.1% were admitted for severe infectious disease. Most patients were treated in an urban hospital (87.5%), in a hospital without teaching status (87.9%), and in the South Census region (74.1%). The majority of patients were treated in medium‐sized to large care facilities. Risk factors for VTE during the baseline period included hospitalization for a medical condition with a high risk for VTE (75.6%), a prior at‐risk hospitalization (18.6%), cancer therapy (10.0% of all medical patients combined and 18.5% of cancer patients), trauma (9.2%), and previous VTE (4.3%).

Figure 1

Table 1.Summary of Patient Demoghraphics and Characteristics

Characteristic	Medical Patients (N = 11,139)
NOTE: All values are presented as no. (%) unless indicated otherwise. 3 Abbreviations: CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease; SD, standard deviation. * This study used only the Commercial and Medicare supplemental databases.
Gender
Men	5389 (48.4)
Women	5750 (51.6)
Reason for hospitalization
Cancer	2243 (20.1)
CHF	1705 (15.3)
Severe lung disease/COPD	5736 (51.5)
Severe infectious disease	1455 (13.1)
Age group, years
1834	230 (2.1)
3544	442 (4.0)
4554	1188 (10.7)
5564	2644 (23.7)
6574	2657 (23.9)
7584	2969 (26.7)
85 years	1009 (9.1)
Median age SD, years	67.6 13.9
Primary payer*
Medicare	6819 (61.2)
Commercial	4320 (38.8)
Geographical area
Northeast	122 (1.1)
North Central	2649 (23.8)
South	8258 (74.1)
West	110 (1.0)
Urban location	9743 (87.5)
Teaching hospital	1345 (12.1)
Licensed bed size
1199	1621 (14.6)
200299	2869 (25.8)
300499	4005 (36.0)
500	2644 (23.7)

Time Course of VTE Risk and Hazard Function

Overall, there were 366 symptomatic VTE events, representing a VTE rate of 3.3%. These events comprised 241 DVT‐only events, 98 PE‐only events, and 27 events with evidence of both DVT and PE. In total, 43.4% of events occurred during hospitalization (Figure 2). The VTE rate was 5.7% in patients with cancer (30.5% of events occurring in hospital), 4.3% with infectious disease (61.9% in hospital), 3.1% with CHF (54.7% in hospital), and 2.1% with severe lung disease/COPD (42.6% in hospital). The highest number of VTE events, 97 events (62 DVT only, 26 PE only, and 9 events both DVT and PE), occurred in the first 9 days after the hospital admission date, of which 87.6% were during hospitalization. During days 10‐19, there were 82 VTE events (50 DVT only, 24 PE only, and 8 both DVT and PE), 70.7% of which occurred in the hospital. Over the following 10‐day periods, VTE incidence gradually declined (Figure 2) and fluctuated at a background level of 4‐8 events during each 10‐day interval from 120 to 180 days.

Figure 2

The cumulative probability of VTE among all patients was 0.035 (Figure 3A). Half of the VTE risk had accumulated by day 23, and 75% had accumulated by day 71. By day 30, the proportion of cumulative risk was 52.6% overall, and ranged from 41.9% with cancer to 72.9% with infectious disease (Figure 3).

Figure 3

The VTE hazard peaked at approximately 1.05 VTE events per 1000 person‐days on day 8 after the hospital admission date overall (Figure 4A). The cumulative hazard at the peak day was 18.2% of the total VTE hazard over the 180‐day evaluation period. The hazard peak ranged from day 7 in patients with severe lung disease/COPD to day 12 in patients with infectious disease (Figure 4B). The cumulative hazard at the peak day was 39.7% for patients with infectious disease, 29.2% for patients with CHF, and approximately 19% for cancer or severe lung disease/COPD. After the peak risk day, the VTE hazard function decreased until the curve reached an inflection point, at day 28, when the cumulative risk was 51.8% (Figure 4A). After the inflection point, the VTE hazard increased to 0.3 VTE events per 1000 person‐days at approximately day 40‐45 and then decreased to <0.2 events per 1000 person‐days. The timing of the inflection varied by approximately 1 week across the medical illnesses (ranging from day 25 for severe lung disease/COPD to day 33 for CHF), with the cumulative risk at the inflection point ranging from 41.9% with cancer to 72.9% with infectious disease.

Figure 4

DISCUSSION

The results from this large, real‐world study provide new insights into the duration of risk of symptomatic VTE in medical patients and demonstrate that the number of VTE events was highest during days 0‐19, with the peak of VTE hazard at day 8. Half of the total 180‐day cumulative risk had been incurred by day 23 after hospital admission, and the period of greatest increased risk extended up to at least 30 days. Importantly, more than half of VTE events occurred after discharge (56.6%). A particularly high proportion of VTE events (69.5%) had occurred after discharge in patients with cancer. Although it was assumed that most VTE events that could be reasonably attributed to an at‐risk hospitalization would occur within 90 days as shown previously,4, 11 the 180‐day evaluation period was used to examine whether there was a prolonged period of continually diminished VTE risk from 90 to 180 days. Thus, events occurring within the later portions of the evaluation period may or may not have been attributable to the index hospitalization, potentially reflecting a background rate of VTE as noted above. Although these events are included in our estimate of the 180‐day cumulative risk of VTE, interpretation of the study results excluding such events is possible by examining the cumulative risk that had been incurred at each time point during the evaluation period.

Few other studies have assessed the duration of VTE risk in hospitalized medical patients. In a study by Spyropoulos et al,4 the median time to a DVT and/or PE event was 74 days, ranging from 62 days in severe infectious disease to 126 days in CHF. In another observational study that included patients who had recently been hospitalized but had not undergone surgery, 66.9% of patients who experienced DVT and/or PE events were diagnosed with DVT and/or PE within the first month after hospital discharge; 19.9% between months 1 and 2, and 13.2% between months 2 and 3.12

Fewer than half of the patients in the present study received thromboprophylaxis, which is consistent with other studies demonstrating the low prophylaxis rates in medical inpatients.9, 1315 In a recently published US study of discharge records that included 22,455 medical inpatients, prophylaxis rates were 59.4% in patients with CHF, 52.3% with cancer, 45.8% with severe lung disease/COPD, and 40.4% with infectious disease.14 Fewer than 10% of patients in the present study received prophylaxis after discharge, a result that is consistent with other studies.4, 9

The effect of extended prophylaxis in acutely ill medical patients with the LMWH enoxaparin beyond 6‐14 days has been investigated in the EXCLAIM study.16 This trial included approximately 5800 acutely ill medical patients at significant risk of developing VTE due to a recent reduction in mobility. Patients in the extended prophylaxis group had a lower risk of VTE (2.5% vs 4% for placebo; absolute risk reduction 1.5% [95.8% confidence interval 2.54% to 0.52%]), but had increased major bleeding events (0.8% vs 0.3% for placebo; absolute risk difference favoring placebo, 0.51% [95% confidence interval, 0.12% to 0.89%]). The patient populations with most benefit from an additional 28 days prophylaxis with enoxaparin, in addition to the usual short‐term prophylaxis of 10 days, were patients with restricted mobility (level 1; total bed rest/sedentary), elderly patients (age >75 years), and women. A limitation of the EXCLAIM trial is that estimates of efficacy and safety are difficult to interpret: after an interim analysis of adjudicated efficacy and safety outcomes, amendments were made to the original study protocol by changing eligibility criteria for patients with level 2 immobility (level 1 with bathroom privileges).16

The optimal duration of prophylaxis for medical patients has not been determined; prophylaxis is generally administered to at‐risk medical patients for the duration of hospitalization. In the current study, mean length of stay was 5.3 5.3 days overall. As hospital stays shorten, many medical patients who are prescribed inpatient prophylaxis alone are unlikely to receive the standard 6‐14 days of prophylaxis shown to be effective in clinical trials.68 Furthermore, the extended period of VTE risk in the present study and the finding that 56.6% of events occurred after discharge also suggest that current practices for inpatient prophylaxis alone may need to be evaluated.

This study reports real‐world data from a large, well‐defined population and obtains the incidence of symptomatic VTE events. Even though certain demographic data deviate from the national averagefor example, 74.1% of patients were treated in the South Census region, whereas this region is served by 37.6% of US hospitals17; 87.5% of hospitals had an urban location (compared with 60.1% of US hospitals18), and 85.4% of hospitals had a licensed bed size of at least 200 beds (compared with 28.2% of US hospitals, with the average US hospital having fewer than 100 beds19)these data may be beneficial in guiding policy and health care strategies for gaining understanding of the duration of risk for VTE.

Limitations of the study include characterization of the VTE risk period through examination of the cumulative risk and hazard of VTE across time, as the actual VTE risk period cannot be determined with exact precision. We used ICD‐9‐CM diagnosis coding to identify VTE. Since many cases of PE are asymptomatic and detected at autopsy,20 our approach may have missed such cases, as they would not have been recorded within the database. Furthermore, validation studies suggest that suboptimal specificity exists for ICD‐9‐CM diagnosis codes used to identify VTE.21 In an attempt to improve the specificity of our VTE identification algorithm, we required that post‐discharge VTE was recorded either during an emergency room or subsequent inpatient admission (which would be indicative of acute care for VTE) or on an outpatient claim with subsequent evidence of treatment for VTE. The true sensitivity and specificity of the VTE identification algorithms used for this study remain unknown, however, so the study findings should be interpreted in light of this limitation. The databases used for the analysis may not be representative of the US population as a whole; for example, this study used claims data from commercial and Medicare supplemental databases, which do not include Medicaid patients. Another limitation was that outpatient mechanical prophylaxis, such as graded compression stockings, was not captured due to over‐the‐counter availability. In addition, appropriateness of prophylaxis was not determined in this study, because these data could not be obtained from the claims database used. Further studies are warranted to obtain information on the incidence of VTE after hospitalization for medical illness in patients who received appropriate prophylaxis during hospitalization.

Finally, all dosages of a pharmacological agent were considered prophylactic only if a VTE event did not occur, with the exception of warfarin; any dose of warfarin was considered for prophylaxis, regardless of a VTE diagnosis. Warfarin may be used for purposes other than VTE prophylaxis (eg, prophylaxis for a thromboembolic cerebrovascular accident). The data source does not allow for identifying the exact reason for anticoagulation therapy with warfarin. Nonetheless, warfarin therapy will confer a decreased risk of VTE regardless of its purpose.

Results from this large cohort of medical patients indicate that symptomatic VTE risk is highest within the first 19 days after hospital admission (a period that may encompass both the duration of hospitalization as well as the period after discharge) with a considerable risk of VTE extending into the period after discharge. Receiving appropriate prophylaxis in‐hospital remains of great importance to prevent inpatient and likely post‐discharge VTE in patients with acute medical illness. In addition, given the time course of VTE events, with VTE incidence peaking at 8 days but with increased risk extending to 30 days, and the number of out‐of‐hospital VTE events incurred, the results of this study suggest that future research is warranted to investigate the risks and benefits of improving thromboprophylaxis practices in the period after hospitalization.

Pain is considered the fifth vital sign, and the appropriate management of pain is fundamental for patient care. Uncontrolled pain has adverse negative physiological consequences,14 and better pain control in hospitalized patients has been associated with decreased length of stay and improved recovery and physical comfort.36 However, many patients fail to receive state‐of‐the‐art pain relief7; for example, in a study of 176 hospitalized patients with cancer, 46% reported severe pain at the time of the interview.8 The prevalence of pain in hospitalized patients with other diagnoses besides cancer likewise remains high.9

Over the last 2 decades, the quality of care in pain management has gained increasing attention. In 2000, the Joint Commission unveiled an official statement for the purpose of improving the quality of pain management.10 The Joint Commission's 6 core principles include the right of pain assessment and treatment, institution of organizational procedures to assess pain, provision of care of persons with pain, general education, continuity of pain management after hospital discharge, and inclusion of pain management as a performance measure. Pain management is now a hospital accountability indicator. Although multiple initiatives have been undertaken to improve pain control, however, challenges still remain. Effective strategies for consensus development are still needed to prioritize interventions.

The nominal group technique (NGT) is a brainstorming tool for quality improvement; NGT is a highly structured small group discussion used to elicit and prioritize a list of answers to a specific question.1115 We conducted an NGT session to identify the multiple challenges, barriers, and perspectives of healthcare providers in managing pain among hospitalized patients. The ultimate goals were to identify potential areas for pain management improvement, build consensus among caregivers, and introduce the NGT as a tool to elicit caregivers' ideas for quality improvement.

METHODS

In a multistep process, we first identified areas for quality improvement interventions in pain management by using the NGT in hospitalized patients, then organized the information by using an iterative consensus development process, and finally displayed the main findings using a Fishbone diagram.

RESULTS

The 27 health workers representing the 3 units completing the nominal group sessions generated a total of 94 ideas. The Fishbone diagram shown in Figure 1 shows each service's top 5 rankings of the elements perceived contributing most to uncontrolled pain; the elements were organized into 3 main factors and 7 priority themes identified during the iterative process.

Figure 1

The main categories illustrate system factors (timeliness, communication, pain assessment; Figure 1, top portion), human factors (knowledge and experience, provider bias, patient factors; Figure 1, bottom portion), and an interface between system and human factors (standardization; Figure 1, center left). The remaining 79 ideas, non‐top 5 for each unit, fell into the following priority themes: provider bias (n = 6), knowledge and experience (n = 12), pain assessment (n = 7), communication (n = 10), timeliness (n = 14), standardization (or policies and practice variation) (n = 13), and patient factors (n = 17); the ideas and representative examples are shown in Figure 2.

Figure 2

DISCUSSION

Using a brainstorming tool for idea generation (NGT) in quality improvement, we identified almost 100 causes of uncontrolled pain management in hospital units. We identified 7 priority themes along 3 main factors: system factors, human factors, and an interface of system and human factors. Timeliness and education emerged from 2 of the 3 services as top priorities, though they were unique and specific to the providers and patient populations within each service. The third service yielded surprisingly different priorities, with patient issues and provider bias foremost in the minds of the staff caring for these patients.

In the decade since To Err Is Human was released by the Institute of Medicine,18 healthcare improvement work has become commonplace and routine. Projects start by soliciting staff views about possible areas for improvement, usually during group meetings; however, this process is informal and not systematic. Unstructured group meetings and brainstorming have some limitations when used to uncover creative ideas for healthcare improvement.19 The literature has consistently reported that groups produce fewer ideas than an equivalent number of individuals working alone.20 In a meta‐analysis, Mullen et al21 found that interacting groups usually produced ideas of poorer quality than did nominal groups. Interpersonal interactions in a multidisciplinary team may be influenced by perceived roles and dominant personalities, and can impede a collaborative, critical discourse.22

In contrast, in NGT sessions, the weight of each member's opinion is the same, and it appears that process loss is less likely to occur.17 Moreover, the highly structured format of NGT provides an opportunity for group members to achieve a substantial amount of work in a relatively short time. Another advantage of NGT is the deliberate avoidance of interference or interpretation from a moderator or facilitator who, in the case of NGT, has the responsibility to explore but not interfere with or influence the members of the group.13 However, the NGT has some limitations. The composition and representativeness of participants may limit the generalizability of the findings. Also, it requires training and preparation, restricts the discussion to a single topic, and may not allow further elaboration of other ideas.13 Despite its potential benefits, NGT is relatively underutilized in quality improvement initiatives. To the best of our knowledge, this is the first study that has utilized NGT to elicit ideas about potential areas for pain control improvement. NGT is a good method for achieving local solutions to local problems; teams of healthcare providers should consider using NGT to address their own challenges and barriers.

In our study, timeliness, knowledge, and experience were considered the top priorities to improve pain management. Our findings are similar to a Canadian study, where delay of more than an hour to administer analgesia was considered one of the most important factors to provide good pain control management.23 Also our findings are coherent with a recent survey of 225 hospitals in the United Kingdom, where perceived lack of training was a highly ranked contributing factor for suboptimal postoperative pain management.24

Our study also identified other interesting observations that deserve comment. Examining the top priorities among the 3 services, there is some dissonance of reasoning underlying the inadequate pain control. We can speculate several reasons. First, pain control management, like any other condition, happens within a specific context with unique problems or barriers that prevent the delivery of the best care for each service. The other possible explanation is a silo effect. It is possible that workers of the same service represent a relatively homogenous social group despite differing training and backgrounds; they live the same experiences and face similar problems. Workers of one silo may work in parallel but do not interact with members of another silo, so they do not have opportunities to share their experiences or compare their beliefs. Finally, it is possible that the greater number of groups used in this study resulted in a broader array of issues. Studies have confirmed that the presence of several groups using NGT can produce a larger pool of issues, with more variation.16 Thus, the ideas generated in our NGT can be easily grouped into 2 broader categories: human factors (knowledge, experience, provider bias, and patient factors) and system factors (timeliness, communication, and pain assessment).

Our study has some limitations. The study was conducted in a single institution and in a limited number of inpatient units. The list of potential areas for improvement generated in this study may require further confirmation and validation at other institutions and with other sources of information (actual pain assessments, timeliness, knowledge, patient satisfaction). Importantly, the study design only involved healthcare providers and did not involve other stakeholders such as the patients or their families.

Despite these limitations, we propose that the NGT is a good alternative to unstructured brainstorming to systematically identify, characterize, categorize, and prioritize ideas behind inadequate pain management. The NGT is a valuable tool in conducting a robust formative assessment to better understand the multiple challenges, barriers, and perspectives of healthcare providers in guiding quality improvement interventions in a systematic and less biased manner.

CONCLUSIONS

In conclusion, healthcare workers have clear ideas about potential areas for improvement in pain management. Using the NGT, we identified 7 potential areas for improvement encompassed within human and system factors. Knowledge and timeliness emerged from 2 of the 3 clinical services as top priorities, whereas the third group identified disparate concerns suggesting provider bias and patient issues. We believe the nominal group technique is an efficient tool to uncover general and context‐specific priorities and to guide quality improvement work.

Pain is considered the fifth vital sign, and the appropriate management of pain is fundamental for patient care. Uncontrolled pain has adverse negative physiological consequences,14 and better pain control in hospitalized patients has been associated with decreased length of stay and improved recovery and physical comfort.36 However, many patients fail to receive state‐of‐the‐art pain relief7; for example, in a study of 176 hospitalized patients with cancer, 46% reported severe pain at the time of the interview.8 The prevalence of pain in hospitalized patients with other diagnoses besides cancer likewise remains high.9

Over the last 2 decades, the quality of care in pain management has gained increasing attention. In 2000, the Joint Commission unveiled an official statement for the purpose of improving the quality of pain management.10 The Joint Commission's 6 core principles include the right of pain assessment and treatment, institution of organizational procedures to assess pain, provision of care of persons with pain, general education, continuity of pain management after hospital discharge, and inclusion of pain management as a performance measure. Pain management is now a hospital accountability indicator. Although multiple initiatives have been undertaken to improve pain control, however, challenges still remain. Effective strategies for consensus development are still needed to prioritize interventions.

The nominal group technique (NGT) is a brainstorming tool for quality improvement; NGT is a highly structured small group discussion used to elicit and prioritize a list of answers to a specific question.1115 We conducted an NGT session to identify the multiple challenges, barriers, and perspectives of healthcare providers in managing pain among hospitalized patients. The ultimate goals were to identify potential areas for pain management improvement, build consensus among caregivers, and introduce the NGT as a tool to elicit caregivers' ideas for quality improvement.

METHODS

In a multistep process, we first identified areas for quality improvement interventions in pain management by using the NGT in hospitalized patients, then organized the information by using an iterative consensus development process, and finally displayed the main findings using a Fishbone diagram.

RESULTS

The 27 health workers representing the 3 units completing the nominal group sessions generated a total of 94 ideas. The Fishbone diagram shown in Figure 1 shows each service's top 5 rankings of the elements perceived contributing most to uncontrolled pain; the elements were organized into 3 main factors and 7 priority themes identified during the iterative process.

Figure 1

The main categories illustrate system factors (timeliness, communication, pain assessment; Figure 1, top portion), human factors (knowledge and experience, provider bias, patient factors; Figure 1, bottom portion), and an interface between system and human factors (standardization; Figure 1, center left). The remaining 79 ideas, non‐top 5 for each unit, fell into the following priority themes: provider bias (n = 6), knowledge and experience (n = 12), pain assessment (n = 7), communication (n = 10), timeliness (n = 14), standardization (or policies and practice variation) (n = 13), and patient factors (n = 17); the ideas and representative examples are shown in Figure 2.

Figure 2

DISCUSSION

Using a brainstorming tool for idea generation (NGT) in quality improvement, we identified almost 100 causes of uncontrolled pain management in hospital units. We identified 7 priority themes along 3 main factors: system factors, human factors, and an interface of system and human factors. Timeliness and education emerged from 2 of the 3 services as top priorities, though they were unique and specific to the providers and patient populations within each service. The third service yielded surprisingly different priorities, with patient issues and provider bias foremost in the minds of the staff caring for these patients.

In the decade since To Err Is Human was released by the Institute of Medicine,18 healthcare improvement work has become commonplace and routine. Projects start by soliciting staff views about possible areas for improvement, usually during group meetings; however, this process is informal and not systematic. Unstructured group meetings and brainstorming have some limitations when used to uncover creative ideas for healthcare improvement.19 The literature has consistently reported that groups produce fewer ideas than an equivalent number of individuals working alone.20 In a meta‐analysis, Mullen et al21 found that interacting groups usually produced ideas of poorer quality than did nominal groups. Interpersonal interactions in a multidisciplinary team may be influenced by perceived roles and dominant personalities, and can impede a collaborative, critical discourse.22

In contrast, in NGT sessions, the weight of each member's opinion is the same, and it appears that process loss is less likely to occur.17 Moreover, the highly structured format of NGT provides an opportunity for group members to achieve a substantial amount of work in a relatively short time. Another advantage of NGT is the deliberate avoidance of interference or interpretation from a moderator or facilitator who, in the case of NGT, has the responsibility to explore but not interfere with or influence the members of the group.13 However, the NGT has some limitations. The composition and representativeness of participants may limit the generalizability of the findings. Also, it requires training and preparation, restricts the discussion to a single topic, and may not allow further elaboration of other ideas.13 Despite its potential benefits, NGT is relatively underutilized in quality improvement initiatives. To the best of our knowledge, this is the first study that has utilized NGT to elicit ideas about potential areas for pain control improvement. NGT is a good method for achieving local solutions to local problems; teams of healthcare providers should consider using NGT to address their own challenges and barriers.

In our study, timeliness, knowledge, and experience were considered the top priorities to improve pain management. Our findings are similar to a Canadian study, where delay of more than an hour to administer analgesia was considered one of the most important factors to provide good pain control management.23 Also our findings are coherent with a recent survey of 225 hospitals in the United Kingdom, where perceived lack of training was a highly ranked contributing factor for suboptimal postoperative pain management.24

Our study also identified other interesting observations that deserve comment. Examining the top priorities among the 3 services, there is some dissonance of reasoning underlying the inadequate pain control. We can speculate several reasons. First, pain control management, like any other condition, happens within a specific context with unique problems or barriers that prevent the delivery of the best care for each service. The other possible explanation is a silo effect. It is possible that workers of the same service represent a relatively homogenous social group despite differing training and backgrounds; they live the same experiences and face similar problems. Workers of one silo may work in parallel but do not interact with members of another silo, so they do not have opportunities to share their experiences or compare their beliefs. Finally, it is possible that the greater number of groups used in this study resulted in a broader array of issues. Studies have confirmed that the presence of several groups using NGT can produce a larger pool of issues, with more variation.16 Thus, the ideas generated in our NGT can be easily grouped into 2 broader categories: human factors (knowledge, experience, provider bias, and patient factors) and system factors (timeliness, communication, and pain assessment).

Our study has some limitations. The study was conducted in a single institution and in a limited number of inpatient units. The list of potential areas for improvement generated in this study may require further confirmation and validation at other institutions and with other sources of information (actual pain assessments, timeliness, knowledge, patient satisfaction). Importantly, the study design only involved healthcare providers and did not involve other stakeholders such as the patients or their families.

Despite these limitations, we propose that the NGT is a good alternative to unstructured brainstorming to systematically identify, characterize, categorize, and prioritize ideas behind inadequate pain management. The NGT is a valuable tool in conducting a robust formative assessment to better understand the multiple challenges, barriers, and perspectives of healthcare providers in guiding quality improvement interventions in a systematic and less biased manner.

CONCLUSIONS

In conclusion, healthcare workers have clear ideas about potential areas for improvement in pain management. Using the NGT, we identified 7 potential areas for improvement encompassed within human and system factors. Knowledge and timeliness emerged from 2 of the 3 clinical services as top priorities, whereas the third group identified disparate concerns suggesting provider bias and patient issues. We believe the nominal group technique is an efficient tool to uncover general and context‐specific priorities and to guide quality improvement work.

Pain is considered the fifth vital sign, and the appropriate management of pain is fundamental for patient care. Uncontrolled pain has adverse negative physiological consequences,14 and better pain control in hospitalized patients has been associated with decreased length of stay and improved recovery and physical comfort.36 However, many patients fail to receive state‐of‐the‐art pain relief7; for example, in a study of 176 hospitalized patients with cancer, 46% reported severe pain at the time of the interview.8 The prevalence of pain in hospitalized patients with other diagnoses besides cancer likewise remains high.9

Over the last 2 decades, the quality of care in pain management has gained increasing attention. In 2000, the Joint Commission unveiled an official statement for the purpose of improving the quality of pain management.10 The Joint Commission's 6 core principles include the right of pain assessment and treatment, institution of organizational procedures to assess pain, provision of care of persons with pain, general education, continuity of pain management after hospital discharge, and inclusion of pain management as a performance measure. Pain management is now a hospital accountability indicator. Although multiple initiatives have been undertaken to improve pain control, however, challenges still remain. Effective strategies for consensus development are still needed to prioritize interventions.

The nominal group technique (NGT) is a brainstorming tool for quality improvement; NGT is a highly structured small group discussion used to elicit and prioritize a list of answers to a specific question.1115 We conducted an NGT session to identify the multiple challenges, barriers, and perspectives of healthcare providers in managing pain among hospitalized patients. The ultimate goals were to identify potential areas for pain management improvement, build consensus among caregivers, and introduce the NGT as a tool to elicit caregivers' ideas for quality improvement.

METHODS

In a multistep process, we first identified areas for quality improvement interventions in pain management by using the NGT in hospitalized patients, then organized the information by using an iterative consensus development process, and finally displayed the main findings using a Fishbone diagram.

RESULTS

The 27 health workers representing the 3 units completing the nominal group sessions generated a total of 94 ideas. The Fishbone diagram shown in Figure 1 shows each service's top 5 rankings of the elements perceived contributing most to uncontrolled pain; the elements were organized into 3 main factors and 7 priority themes identified during the iterative process.

Figure 1

The main categories illustrate system factors (timeliness, communication, pain assessment; Figure 1, top portion), human factors (knowledge and experience, provider bias, patient factors; Figure 1, bottom portion), and an interface between system and human factors (standardization; Figure 1, center left). The remaining 79 ideas, non‐top 5 for each unit, fell into the following priority themes: provider bias (n = 6), knowledge and experience (n = 12), pain assessment (n = 7), communication (n = 10), timeliness (n = 14), standardization (or policies and practice variation) (n = 13), and patient factors (n = 17); the ideas and representative examples are shown in Figure 2.

Figure 2

DISCUSSION

Using a brainstorming tool for idea generation (NGT) in quality improvement, we identified almost 100 causes of uncontrolled pain management in hospital units. We identified 7 priority themes along 3 main factors: system factors, human factors, and an interface of system and human factors. Timeliness and education emerged from 2 of the 3 services as top priorities, though they were unique and specific to the providers and patient populations within each service. The third service yielded surprisingly different priorities, with patient issues and provider bias foremost in the minds of the staff caring for these patients.

In the decade since To Err Is Human was released by the Institute of Medicine,18 healthcare improvement work has become commonplace and routine. Projects start by soliciting staff views about possible areas for improvement, usually during group meetings; however, this process is informal and not systematic. Unstructured group meetings and brainstorming have some limitations when used to uncover creative ideas for healthcare improvement.19 The literature has consistently reported that groups produce fewer ideas than an equivalent number of individuals working alone.20 In a meta‐analysis, Mullen et al21 found that interacting groups usually produced ideas of poorer quality than did nominal groups. Interpersonal interactions in a multidisciplinary team may be influenced by perceived roles and dominant personalities, and can impede a collaborative, critical discourse.22

In contrast, in NGT sessions, the weight of each member's opinion is the same, and it appears that process loss is less likely to occur.17 Moreover, the highly structured format of NGT provides an opportunity for group members to achieve a substantial amount of work in a relatively short time. Another advantage of NGT is the deliberate avoidance of interference or interpretation from a moderator or facilitator who, in the case of NGT, has the responsibility to explore but not interfere with or influence the members of the group.13 However, the NGT has some limitations. The composition and representativeness of participants may limit the generalizability of the findings. Also, it requires training and preparation, restricts the discussion to a single topic, and may not allow further elaboration of other ideas.13 Despite its potential benefits, NGT is relatively underutilized in quality improvement initiatives. To the best of our knowledge, this is the first study that has utilized NGT to elicit ideas about potential areas for pain control improvement. NGT is a good method for achieving local solutions to local problems; teams of healthcare providers should consider using NGT to address their own challenges and barriers.

In our study, timeliness, knowledge, and experience were considered the top priorities to improve pain management. Our findings are similar to a Canadian study, where delay of more than an hour to administer analgesia was considered one of the most important factors to provide good pain control management.23 Also our findings are coherent with a recent survey of 225 hospitals in the United Kingdom, where perceived lack of training was a highly ranked contributing factor for suboptimal postoperative pain management.24

Our study also identified other interesting observations that deserve comment. Examining the top priorities among the 3 services, there is some dissonance of reasoning underlying the inadequate pain control. We can speculate several reasons. First, pain control management, like any other condition, happens within a specific context with unique problems or barriers that prevent the delivery of the best care for each service. The other possible explanation is a silo effect. It is possible that workers of the same service represent a relatively homogenous social group despite differing training and backgrounds; they live the same experiences and face similar problems. Workers of one silo may work in parallel but do not interact with members of another silo, so they do not have opportunities to share their experiences or compare their beliefs. Finally, it is possible that the greater number of groups used in this study resulted in a broader array of issues. Studies have confirmed that the presence of several groups using NGT can produce a larger pool of issues, with more variation.16 Thus, the ideas generated in our NGT can be easily grouped into 2 broader categories: human factors (knowledge, experience, provider bias, and patient factors) and system factors (timeliness, communication, and pain assessment).

Our study has some limitations. The study was conducted in a single institution and in a limited number of inpatient units. The list of potential areas for improvement generated in this study may require further confirmation and validation at other institutions and with other sources of information (actual pain assessments, timeliness, knowledge, patient satisfaction). Importantly, the study design only involved healthcare providers and did not involve other stakeholders such as the patients or their families.

Despite these limitations, we propose that the NGT is a good alternative to unstructured brainstorming to systematically identify, characterize, categorize, and prioritize ideas behind inadequate pain management. The NGT is a valuable tool in conducting a robust formative assessment to better understand the multiple challenges, barriers, and perspectives of healthcare providers in guiding quality improvement interventions in a systematic and less biased manner.

CONCLUSIONS

In conclusion, healthcare workers have clear ideas about potential areas for improvement in pain management. Using the NGT, we identified 7 potential areas for improvement encompassed within human and system factors. Knowledge and timeliness emerged from 2 of the 3 clinical services as top priorities, whereas the third group identified disparate concerns suggesting provider bias and patient issues. We believe the nominal group technique is an efficient tool to uncover general and context‐specific priorities and to guide quality improvement work.

Proton pump inhibitors (PPIs) are the third most commonly prescribed class of medication in the United States, with $13.6 billion in yearly sales.1 Despite their effectiveness in treating acid reflux2 and their mortality benefit in the treatment of patients with gastrointestinal bleeding,3 recent literature has identified a number of risks associated with PPIs, including an increased incidence of Clostridium difficile infection,4 decreased effectiveness of clopidogrel in patients with acute coronary syndrome,5 increased risk of community‐ and hospital‐acquired pneumonia, and an increased risk of hip fracture.69 Additionally, in March of 2011, the US Food and Drug Administration (FDA) issued a warning regarding the potential for PPIs to cause low magnesium levels which can, in turn, cause muscle spasms, an irregular heartbeat, and convulsions.10

Inappropriate PPI prescription practice has been demonstrated in the primary care setting,11 as well as in small studies conducted in the hospital setting.1216 We hypothesized that many hospitalized patients receive these medications without having an accepted indication, and examined 2 populations of hospitalized patients, including administrative data from 6.5 million discharges from US university hospitals, to look for appropriate diagnoses justifying their use.

METHODS

We performed a retrospective review of administrative data collected between January 1, 2008 and December 31, 2009 from 2 patient populations: (a) those discharged from Denver Health (DH), a university‐affiliated public safety net hospital in Denver, CO; and (b) patients discharged from 112 academic health centers and 256 of their affiliated hospitals that participate in the University HealthSystem Consortium (UHC). The Colorado Multiple Institution Review Board reviewed and approved the conduct of this study.

Inclusion criteria for both populations were age >18 or <90 years, and hospitalization on a Medicine service. Prisoners and women known to be pregnant were excluded. In both cohorts, if patients had more than 1 admission during the 2‐year study period, only data from the first admission were used.

We recorded demographics, admitting diagnosis, and discharge diagnoses together with information pertaining to the name, route, and duration of administration of all PPIs (ie, omeprazole, lansoprazole, esomeprazole, pantoprazole, rabeprazole). We created a broadly inclusive set of valid indications for PPIs by incorporating diagnoses that could be identified by International Classification of Diseases, Ninth Revision.

(ICD‐9) codes from a number of previously published sources including the National Institute of Clinical Excellence (NICE) guidelines issued by the National Health Service (NHS) of the United Kingdom in 200012, 1721 (Table 1).

To assess the accuracy of the administrative data from DH, we also reviewed the Emergency Department histories, admission histories, progress notes, electronic pharmacy records, endoscopy reports, and discharge summaries of 123 patients randomly selected (ie, a 5% sample) from the group of patients identified by administrative data to have received a PPI without a valid indication, looking for any accepted indication that might have been missed in the administrative data.

All analyses were performed using SAS Enterprise Guide 4.1 (SAS Institute, Cary, NC). A Student t test was used to compare continuous variables and a chi‐square test was used to compare categorical variables. Bonferroni corrections were used for multiple comparisons, such that P values less than 0.01 were considered to be significant for categorical variables.

RESULTS

Inclusion criteria were met by 9875 patients in the Denver Health database and 6,592,100 patients in the UHC database. The demographics and primary discharge diagnoses for these patients are summarized in Table 2.

Only 39% and 27% of the patients in the DH and UHC databases, respectively, had a valid indication for PPIs on the basis of discharge diagnoses (Table 3). In the DH data, if admission ICD‐9 codes were also inspected for valid PPI indications, 1579 (40%) of patients receiving PPIs had a valid indication (admission ICD‐9 codes were not available for patients in the UHC database). Thirty‐one percent of Denver Health patients spent time in the intensive care unit (ICU) during their hospital stay and 65% of those patients received a PPI without a valid indication, as compared to 59% of patients who remained on the General Medicine ward (Table 3).

Higher rates of concurrent C. difficile infections were observed in patients receiving PPIs in both databases; a higher rate of concurrent diagnosis of pneumonia was seen in patients receiving PPIs in the UHC population, with a nonsignificant trend towards the same finding in DH patients (Table 4).

Chart review in the DH population found valid indications for PPIs in 19% of patients who were thought not have a valid indication on the basis of the administrative data (Table 5). For 56% of those in whom no valid indication was confirmed, physicians identified prophylaxis as the justification.

DISCUSSION

The important finding of this study was that the majority of patients in 2 large groups of Medicine patients hospitalized in university‐affiliated hospitals received PPIs without having a valid indication. To our knowledge, the more than 900,000 UHC patients who received a PPI during their hospitalization represent the largest inpatient population evaluated for appropriateness of PPI prescriptions.

Our finding that 41% of the patients admitted to the DH Medicine service received a PPI during their hospital stay is similar to what has been observed by others.9, 14, 22 The rate of PPI prescription was lower in the UHC population (14%) for unclear reasons. By our definition, 61% lacked an adequate diagnosis to justify the prescription of the PPI. After performing a chart review on a randomly selected 5% of these records, we found that the DH administrative database had failed to identify 19% of patients who had a valid indication for receiving a PPI. Adjusting the administrative data accordingly still resulted in 50% of DH patients not having a valid indication for receiving a PPI. This is consistent with the 54% recorded by Batuwitage and colleagues11 in the outpatient setting by direct chart review, as well as a range of 60%‐75% for hospitalized patients in other studies.12, 13, 15, 23, 24

Stomach acidity is believed to provide an important host defense against lower gastrointestinal tract infections including Salmonella, Campylobacter, and Clostridium difficile.25 A recent study by Howell et al26 showed a doseresponse effect between PPI use and C. difficile infection, supporting a causal connection between loss of stomach acidity and development of Clostridium difficile‐associated diarrhea (CDAD). We found that C. difficile infection was more common in both populations of patients receiving PPIs (although the relative risk was much higher in the UHC database) (Table 5). The rate of CDAD in DH patients who received PPIs was 2.6 times higher than in patients who did not receive these acid suppressive agents.

The role of acid suppression in increasing risk for community‐acquired pneumonia is not entirely clear. Theories regarding the loss of an important host defense and bacterial proliferation head the list.6, 8, 27 Gastric and duodenal bacterial overgrowth is significantly more common in patients receiving PPIs than in patients receiving histamine type‐2 (H2) blockers.28 Previous studies have identified an increased rate of hospital‐acquired pneumonia and recurrent community‐acquired pneumonia27 in patients receiving any form of acid suppression therapy, but the risk appears to be greater in patients receiving PPIs than in those receiving H2 receptor antagonists (H2RAs).9 Significantly more patients in the UHC population who were taking PPIs had a concurrent diagnosis of pneumonia, consistent with previous studies alerting to this association6, 8, 9, 27 and consistent with the nonsignificant trend observed in the DH population.

Our study has a number of limitations. Our database comes from a single university‐affiliated public hospital with residents and hospitalists writing orders for all medications. The hospitals in the UHC are also teaching hospitals. Accordingly, our results might not generalize to other settings or reflect prescribing patterns in private, nonteaching hospital environments. Because our study was retrospective, we could not confirm the decision‐making process supporting the prescription of PPIs. Similarly, we could not temporarily relate the existence of the indication with the time the PPI was prescribed. Our list of appropriate indications for prescribing PPIs was developed by reviewing a number of references, and other studies have used slightly different lists (albeit the more commonly recognized indications are the same), but it may be argued that the list either includes or misses diagnoses in error.

While there is considerable debate about the use of PPIs for stress ulcer prophylaxis,29 we specifically chose not to include this as one of our valid indications for PPIs for 4 reasons. First, the American Society of Health‐System Pharmacists (ASHP) Report does not recommend prophylaxis for non‐ICU patients, and only recommends prophylaxis for those ICU patients with a coagulopathy, those requiring mechanical ventilation for more than 48 hours, those with a history of gastrointestinal ulceration or bleeding in the year prior to admission, and those with 2 or more of the following indications: sepsis, ICU stay >1 week, occult bleeding lasting 6 or more days, receiving high‐dose corticosteroids, and selected surgical situations.30 At the time the guideline was written, the authors note that there was insufficient data on PPIs to make any recommendations on their use, but no subsequent guidelines have been issued.30 Second, a review by Mohebbi and Hesch published in 2009, and a meta‐analysis by Lin and colleagues published in 2010, summarize subsequent randomized trials that suggest that PPIs and H2 blockers are, at best, similarly effective at preventing upper gastrointestinal (GI) bleeding among critically ill patients.31, 32 Third, the NICE guidelines do not include stress ulcer prophylaxis as an appropriate indication for PPIs except in the prevention and treatment of NSAID [non‐steroidal anti‐inflammatory drug]‐associated ulcers.19 Finally, H2RAs are currently the only medications with an FDA‐approved indication for stress ulcer prophylaxis. We acknowledge that PPIs may be a reasonable and acceptable choice for stress ulcer prophylaxis in patients who meet indications, but we were unable to identify such patients in either of our administrative databases.

In our Denver Health population, only 31% of our patients spent any time in the intensive care unit, and only a fraction of these would have both an accepted indication for stress ulcer prophylaxis by the ASHP guidelines and an intolerance or contraindication to an H2RA or sulcralfate. While our administrative database lacked the detail necessary to identify this small group of patients, the number of patients who might have been misclassified as not having a valid PPI indication was likely very small. Similar to the findings of previous studies,15, 18, 23, 29 prophylaxis against gastrointestinal bleeding was the stated justification for prescribing the PPI in 56% of the DH patient charts reviewed. It is impossible for us to estimate the number of patients in our administrative database for whom stress ulcer prophylaxis was justified by existing guidelines, as it would be necessary to gather a number of specific clinical details for each patient including: 1) ICU stay; 2) presence of coagulopathy; 3) duration of mechanical ventilation; 4) presence of sepsis; 5) duration of ICU stay; 6) presence of occult bleeding for >6 days; and 7) use of high‐dose corticosteroids. This level of clinical detail would likely only be available through a prospective study design, as has been suggested by other authors.33 Further research into the use, safety, and effectiveness of PPIs specifically for stress ulcer prophylaxis is warranted.

In conclusion, we found that 73% of nearly 1 million Medicine patients discharged from academic medical centers received a PPI without a valid indication during their hospitalization. The implications of our findings are broad. PPIs are more expensive31 than H2RAs and there is increasing evidence that they have significant side effects. In both databases we examined, the rate of C. difficile infection was higher in patients receiving PPIs than others. The prescribing habits of physicians in these university hospital settings appear to be far out of line with published guidelines and evidence‐based practice. Reducing inappropriate prescribing of PPIs would be an important educational and quality assurance project in most institutions.

Proton pump inhibitors (PPIs) are the third most commonly prescribed class of medication in the United States, with $13.6 billion in yearly sales.1 Despite their effectiveness in treating acid reflux2 and their mortality benefit in the treatment of patients with gastrointestinal bleeding,3 recent literature has identified a number of risks associated with PPIs, including an increased incidence of Clostridium difficile infection,4 decreased effectiveness of clopidogrel in patients with acute coronary syndrome,5 increased risk of community‐ and hospital‐acquired pneumonia, and an increased risk of hip fracture.69 Additionally, in March of 2011, the US Food and Drug Administration (FDA) issued a warning regarding the potential for PPIs to cause low magnesium levels which can, in turn, cause muscle spasms, an irregular heartbeat, and convulsions.10

Inappropriate PPI prescription practice has been demonstrated in the primary care setting,11 as well as in small studies conducted in the hospital setting.1216 We hypothesized that many hospitalized patients receive these medications without having an accepted indication, and examined 2 populations of hospitalized patients, including administrative data from 6.5 million discharges from US university hospitals, to look for appropriate diagnoses justifying their use.

METHODS

We performed a retrospective review of administrative data collected between January 1, 2008 and December 31, 2009 from 2 patient populations: (a) those discharged from Denver Health (DH), a university‐affiliated public safety net hospital in Denver, CO; and (b) patients discharged from 112 academic health centers and 256 of their affiliated hospitals that participate in the University HealthSystem Consortium (UHC). The Colorado Multiple Institution Review Board reviewed and approved the conduct of this study.

Inclusion criteria for both populations were age >18 or <90 years, and hospitalization on a Medicine service. Prisoners and women known to be pregnant were excluded. In both cohorts, if patients had more than 1 admission during the 2‐year study period, only data from the first admission were used.

We recorded demographics, admitting diagnosis, and discharge diagnoses together with information pertaining to the name, route, and duration of administration of all PPIs (ie, omeprazole, lansoprazole, esomeprazole, pantoprazole, rabeprazole). We created a broadly inclusive set of valid indications for PPIs by incorporating diagnoses that could be identified by International Classification of Diseases, Ninth Revision.

(ICD‐9) codes from a number of previously published sources including the National Institute of Clinical Excellence (NICE) guidelines issued by the National Health Service (NHS) of the United Kingdom in 200012, 1721 (Table 1).

To assess the accuracy of the administrative data from DH, we also reviewed the Emergency Department histories, admission histories, progress notes, electronic pharmacy records, endoscopy reports, and discharge summaries of 123 patients randomly selected (ie, a 5% sample) from the group of patients identified by administrative data to have received a PPI without a valid indication, looking for any accepted indication that might have been missed in the administrative data.

All analyses were performed using SAS Enterprise Guide 4.1 (SAS Institute, Cary, NC). A Student t test was used to compare continuous variables and a chi‐square test was used to compare categorical variables. Bonferroni corrections were used for multiple comparisons, such that P values less than 0.01 were considered to be significant for categorical variables.

RESULTS

Inclusion criteria were met by 9875 patients in the Denver Health database and 6,592,100 patients in the UHC database. The demographics and primary discharge diagnoses for these patients are summarized in Table 2.

Only 39% and 27% of the patients in the DH and UHC databases, respectively, had a valid indication for PPIs on the basis of discharge diagnoses (Table 3). In the DH data, if admission ICD‐9 codes were also inspected for valid PPI indications, 1579 (40%) of patients receiving PPIs had a valid indication (admission ICD‐9 codes were not available for patients in the UHC database). Thirty‐one percent of Denver Health patients spent time in the intensive care unit (ICU) during their hospital stay and 65% of those patients received a PPI without a valid indication, as compared to 59% of patients who remained on the General Medicine ward (Table 3).

Higher rates of concurrent C. difficile infections were observed in patients receiving PPIs in both databases; a higher rate of concurrent diagnosis of pneumonia was seen in patients receiving PPIs in the UHC population, with a nonsignificant trend towards the same finding in DH patients (Table 4).

Chart review in the DH population found valid indications for PPIs in 19% of patients who were thought not have a valid indication on the basis of the administrative data (Table 5). For 56% of those in whom no valid indication was confirmed, physicians identified prophylaxis as the justification.

DISCUSSION

The important finding of this study was that the majority of patients in 2 large groups of Medicine patients hospitalized in university‐affiliated hospitals received PPIs without having a valid indication. To our knowledge, the more than 900,000 UHC patients who received a PPI during their hospitalization represent the largest inpatient population evaluated for appropriateness of PPI prescriptions.

Our finding that 41% of the patients admitted to the DH Medicine service received a PPI during their hospital stay is similar to what has been observed by others.9, 14, 22 The rate of PPI prescription was lower in the UHC population (14%) for unclear reasons. By our definition, 61% lacked an adequate diagnosis to justify the prescription of the PPI. After performing a chart review on a randomly selected 5% of these records, we found that the DH administrative database had failed to identify 19% of patients who had a valid indication for receiving a PPI. Adjusting the administrative data accordingly still resulted in 50% of DH patients not having a valid indication for receiving a PPI. This is consistent with the 54% recorded by Batuwitage and colleagues11 in the outpatient setting by direct chart review, as well as a range of 60%‐75% for hospitalized patients in other studies.12, 13, 15, 23, 24

Stomach acidity is believed to provide an important host defense against lower gastrointestinal tract infections including Salmonella, Campylobacter, and Clostridium difficile.25 A recent study by Howell et al26 showed a doseresponse effect between PPI use and C. difficile infection, supporting a causal connection between loss of stomach acidity and development of Clostridium difficile‐associated diarrhea (CDAD). We found that C. difficile infection was more common in both populations of patients receiving PPIs (although the relative risk was much higher in the UHC database) (Table 5). The rate of CDAD in DH patients who received PPIs was 2.6 times higher than in patients who did not receive these acid suppressive agents.

The role of acid suppression in increasing risk for community‐acquired pneumonia is not entirely clear. Theories regarding the loss of an important host defense and bacterial proliferation head the list.6, 8, 27 Gastric and duodenal bacterial overgrowth is significantly more common in patients receiving PPIs than in patients receiving histamine type‐2 (H2) blockers.28 Previous studies have identified an increased rate of hospital‐acquired pneumonia and recurrent community‐acquired pneumonia27 in patients receiving any form of acid suppression therapy, but the risk appears to be greater in patients receiving PPIs than in those receiving H2 receptor antagonists (H2RAs).9 Significantly more patients in the UHC population who were taking PPIs had a concurrent diagnosis of pneumonia, consistent with previous studies alerting to this association6, 8, 9, 27 and consistent with the nonsignificant trend observed in the DH population.

Our study has a number of limitations. Our database comes from a single university‐affiliated public hospital with residents and hospitalists writing orders for all medications. The hospitals in the UHC are also teaching hospitals. Accordingly, our results might not generalize to other settings or reflect prescribing patterns in private, nonteaching hospital environments. Because our study was retrospective, we could not confirm the decision‐making process supporting the prescription of PPIs. Similarly, we could not temporarily relate the existence of the indication with the time the PPI was prescribed. Our list of appropriate indications for prescribing PPIs was developed by reviewing a number of references, and other studies have used slightly different lists (albeit the more commonly recognized indications are the same), but it may be argued that the list either includes or misses diagnoses in error.

While there is considerable debate about the use of PPIs for stress ulcer prophylaxis,29 we specifically chose not to include this as one of our valid indications for PPIs for 4 reasons. First, the American Society of Health‐System Pharmacists (ASHP) Report does not recommend prophylaxis for non‐ICU patients, and only recommends prophylaxis for those ICU patients with a coagulopathy, those requiring mechanical ventilation for more than 48 hours, those with a history of gastrointestinal ulceration or bleeding in the year prior to admission, and those with 2 or more of the following indications: sepsis, ICU stay >1 week, occult bleeding lasting 6 or more days, receiving high‐dose corticosteroids, and selected surgical situations.30 At the time the guideline was written, the authors note that there was insufficient data on PPIs to make any recommendations on their use, but no subsequent guidelines have been issued.30 Second, a review by Mohebbi and Hesch published in 2009, and a meta‐analysis by Lin and colleagues published in 2010, summarize subsequent randomized trials that suggest that PPIs and H2 blockers are, at best, similarly effective at preventing upper gastrointestinal (GI) bleeding among critically ill patients.31, 32 Third, the NICE guidelines do not include stress ulcer prophylaxis as an appropriate indication for PPIs except in the prevention and treatment of NSAID [non‐steroidal anti‐inflammatory drug]‐associated ulcers.19 Finally, H2RAs are currently the only medications with an FDA‐approved indication for stress ulcer prophylaxis. We acknowledge that PPIs may be a reasonable and acceptable choice for stress ulcer prophylaxis in patients who meet indications, but we were unable to identify such patients in either of our administrative databases.

In our Denver Health population, only 31% of our patients spent any time in the intensive care unit, and only a fraction of these would have both an accepted indication for stress ulcer prophylaxis by the ASHP guidelines and an intolerance or contraindication to an H2RA or sulcralfate. While our administrative database lacked the detail necessary to identify this small group of patients, the number of patients who might have been misclassified as not having a valid PPI indication was likely very small. Similar to the findings of previous studies,15, 18, 23, 29 prophylaxis against gastrointestinal bleeding was the stated justification for prescribing the PPI in 56% of the DH patient charts reviewed. It is impossible for us to estimate the number of patients in our administrative database for whom stress ulcer prophylaxis was justified by existing guidelines, as it would be necessary to gather a number of specific clinical details for each patient including: 1) ICU stay; 2) presence of coagulopathy; 3) duration of mechanical ventilation; 4) presence of sepsis; 5) duration of ICU stay; 6) presence of occult bleeding for >6 days; and 7) use of high‐dose corticosteroids. This level of clinical detail would likely only be available through a prospective study design, as has been suggested by other authors.33 Further research into the use, safety, and effectiveness of PPIs specifically for stress ulcer prophylaxis is warranted.

In conclusion, we found that 73% of nearly 1 million Medicine patients discharged from academic medical centers received a PPI without a valid indication during their hospitalization. The implications of our findings are broad. PPIs are more expensive31 than H2RAs and there is increasing evidence that they have significant side effects. In both databases we examined, the rate of C. difficile infection was higher in patients receiving PPIs than others. The prescribing habits of physicians in these university hospital settings appear to be far out of line with published guidelines and evidence‐based practice. Reducing inappropriate prescribing of PPIs would be an important educational and quality assurance project in most institutions.

Proton pump inhibitors (PPIs) are the third most commonly prescribed class of medication in the United States, with $13.6 billion in yearly sales.1 Despite their effectiveness in treating acid reflux2 and their mortality benefit in the treatment of patients with gastrointestinal bleeding,3 recent literature has identified a number of risks associated with PPIs, including an increased incidence of Clostridium difficile infection,4 decreased effectiveness of clopidogrel in patients with acute coronary syndrome,5 increased risk of community‐ and hospital‐acquired pneumonia, and an increased risk of hip fracture.69 Additionally, in March of 2011, the US Food and Drug Administration (FDA) issued a warning regarding the potential for PPIs to cause low magnesium levels which can, in turn, cause muscle spasms, an irregular heartbeat, and convulsions.10

Inappropriate PPI prescription practice has been demonstrated in the primary care setting,11 as well as in small studies conducted in the hospital setting.1216 We hypothesized that many hospitalized patients receive these medications without having an accepted indication, and examined 2 populations of hospitalized patients, including administrative data from 6.5 million discharges from US university hospitals, to look for appropriate diagnoses justifying their use.

METHODS

We performed a retrospective review of administrative data collected between January 1, 2008 and December 31, 2009 from 2 patient populations: (a) those discharged from Denver Health (DH), a university‐affiliated public safety net hospital in Denver, CO; and (b) patients discharged from 112 academic health centers and 256 of their affiliated hospitals that participate in the University HealthSystem Consortium (UHC). The Colorado Multiple Institution Review Board reviewed and approved the conduct of this study.

Inclusion criteria for both populations were age >18 or <90 years, and hospitalization on a Medicine service. Prisoners and women known to be pregnant were excluded. In both cohorts, if patients had more than 1 admission during the 2‐year study period, only data from the first admission were used.

We recorded demographics, admitting diagnosis, and discharge diagnoses together with information pertaining to the name, route, and duration of administration of all PPIs (ie, omeprazole, lansoprazole, esomeprazole, pantoprazole, rabeprazole). We created a broadly inclusive set of valid indications for PPIs by incorporating diagnoses that could be identified by International Classification of Diseases, Ninth Revision.

(ICD‐9) codes from a number of previously published sources including the National Institute of Clinical Excellence (NICE) guidelines issued by the National Health Service (NHS) of the United Kingdom in 200012, 1721 (Table 1).

To assess the accuracy of the administrative data from DH, we also reviewed the Emergency Department histories, admission histories, progress notes, electronic pharmacy records, endoscopy reports, and discharge summaries of 123 patients randomly selected (ie, a 5% sample) from the group of patients identified by administrative data to have received a PPI without a valid indication, looking for any accepted indication that might have been missed in the administrative data.

All analyses were performed using SAS Enterprise Guide 4.1 (SAS Institute, Cary, NC). A Student t test was used to compare continuous variables and a chi‐square test was used to compare categorical variables. Bonferroni corrections were used for multiple comparisons, such that P values less than 0.01 were considered to be significant for categorical variables.

RESULTS

Inclusion criteria were met by 9875 patients in the Denver Health database and 6,592,100 patients in the UHC database. The demographics and primary discharge diagnoses for these patients are summarized in Table 2.

Only 39% and 27% of the patients in the DH and UHC databases, respectively, had a valid indication for PPIs on the basis of discharge diagnoses (Table 3). In the DH data, if admission ICD‐9 codes were also inspected for valid PPI indications, 1579 (40%) of patients receiving PPIs had a valid indication (admission ICD‐9 codes were not available for patients in the UHC database). Thirty‐one percent of Denver Health patients spent time in the intensive care unit (ICU) during their hospital stay and 65% of those patients received a PPI without a valid indication, as compared to 59% of patients who remained on the General Medicine ward (Table 3).

Higher rates of concurrent C. difficile infections were observed in patients receiving PPIs in both databases; a higher rate of concurrent diagnosis of pneumonia was seen in patients receiving PPIs in the UHC population, with a nonsignificant trend towards the same finding in DH patients (Table 4).

Chart review in the DH population found valid indications for PPIs in 19% of patients who were thought not have a valid indication on the basis of the administrative data (Table 5). For 56% of those in whom no valid indication was confirmed, physicians identified prophylaxis as the justification.

DISCUSSION

The important finding of this study was that the majority of patients in 2 large groups of Medicine patients hospitalized in university‐affiliated hospitals received PPIs without having a valid indication. To our knowledge, the more than 900,000 UHC patients who received a PPI during their hospitalization represent the largest inpatient population evaluated for appropriateness of PPI prescriptions.

Our finding that 41% of the patients admitted to the DH Medicine service received a PPI during their hospital stay is similar to what has been observed by others.9, 14, 22 The rate of PPI prescription was lower in the UHC population (14%) for unclear reasons. By our definition, 61% lacked an adequate diagnosis to justify the prescription of the PPI. After performing a chart review on a randomly selected 5% of these records, we found that the DH administrative database had failed to identify 19% of patients who had a valid indication for receiving a PPI. Adjusting the administrative data accordingly still resulted in 50% of DH patients not having a valid indication for receiving a PPI. This is consistent with the 54% recorded by Batuwitage and colleagues11 in the outpatient setting by direct chart review, as well as a range of 60%‐75% for hospitalized patients in other studies.12, 13, 15, 23, 24

Stomach acidity is believed to provide an important host defense against lower gastrointestinal tract infections including Salmonella, Campylobacter, and Clostridium difficile.25 A recent study by Howell et al26 showed a doseresponse effect between PPI use and C. difficile infection, supporting a causal connection between loss of stomach acidity and development of Clostridium difficile‐associated diarrhea (CDAD). We found that C. difficile infection was more common in both populations of patients receiving PPIs (although the relative risk was much higher in the UHC database) (Table 5). The rate of CDAD in DH patients who received PPIs was 2.6 times higher than in patients who did not receive these acid suppressive agents.

The role of acid suppression in increasing risk for community‐acquired pneumonia is not entirely clear. Theories regarding the loss of an important host defense and bacterial proliferation head the list.6, 8, 27 Gastric and duodenal bacterial overgrowth is significantly more common in patients receiving PPIs than in patients receiving histamine type‐2 (H2) blockers.28 Previous studies have identified an increased rate of hospital‐acquired pneumonia and recurrent community‐acquired pneumonia27 in patients receiving any form of acid suppression therapy, but the risk appears to be greater in patients receiving PPIs than in those receiving H2 receptor antagonists (H2RAs).9 Significantly more patients in the UHC population who were taking PPIs had a concurrent diagnosis of pneumonia, consistent with previous studies alerting to this association6, 8, 9, 27 and consistent with the nonsignificant trend observed in the DH population.

Our study has a number of limitations. Our database comes from a single university‐affiliated public hospital with residents and hospitalists writing orders for all medications. The hospitals in the UHC are also teaching hospitals. Accordingly, our results might not generalize to other settings or reflect prescribing patterns in private, nonteaching hospital environments. Because our study was retrospective, we could not confirm the decision‐making process supporting the prescription of PPIs. Similarly, we could not temporarily relate the existence of the indication with the time the PPI was prescribed. Our list of appropriate indications for prescribing PPIs was developed by reviewing a number of references, and other studies have used slightly different lists (albeit the more commonly recognized indications are the same), but it may be argued that the list either includes or misses diagnoses in error.

While there is considerable debate about the use of PPIs for stress ulcer prophylaxis,29 we specifically chose not to include this as one of our valid indications for PPIs for 4 reasons. First, the American Society of Health‐System Pharmacists (ASHP) Report does not recommend prophylaxis for non‐ICU patients, and only recommends prophylaxis for those ICU patients with a coagulopathy, those requiring mechanical ventilation for more than 48 hours, those with a history of gastrointestinal ulceration or bleeding in the year prior to admission, and those with 2 or more of the following indications: sepsis, ICU stay >1 week, occult bleeding lasting 6 or more days, receiving high‐dose corticosteroids, and selected surgical situations.30 At the time the guideline was written, the authors note that there was insufficient data on PPIs to make any recommendations on their use, but no subsequent guidelines have been issued.30 Second, a review by Mohebbi and Hesch published in 2009, and a meta‐analysis by Lin and colleagues published in 2010, summarize subsequent randomized trials that suggest that PPIs and H2 blockers are, at best, similarly effective at preventing upper gastrointestinal (GI) bleeding among critically ill patients.31, 32 Third, the NICE guidelines do not include stress ulcer prophylaxis as an appropriate indication for PPIs except in the prevention and treatment of NSAID [non‐steroidal anti‐inflammatory drug]‐associated ulcers.19 Finally, H2RAs are currently the only medications with an FDA‐approved indication for stress ulcer prophylaxis. We acknowledge that PPIs may be a reasonable and acceptable choice for stress ulcer prophylaxis in patients who meet indications, but we were unable to identify such patients in either of our administrative databases.

In our Denver Health population, only 31% of our patients spent any time in the intensive care unit, and only a fraction of these would have both an accepted indication for stress ulcer prophylaxis by the ASHP guidelines and an intolerance or contraindication to an H2RA or sulcralfate. While our administrative database lacked the detail necessary to identify this small group of patients, the number of patients who might have been misclassified as not having a valid PPI indication was likely very small. Similar to the findings of previous studies,15, 18, 23, 29 prophylaxis against gastrointestinal bleeding was the stated justification for prescribing the PPI in 56% of the DH patient charts reviewed. It is impossible for us to estimate the number of patients in our administrative database for whom stress ulcer prophylaxis was justified by existing guidelines, as it would be necessary to gather a number of specific clinical details for each patient including: 1) ICU stay; 2) presence of coagulopathy; 3) duration of mechanical ventilation; 4) presence of sepsis; 5) duration of ICU stay; 6) presence of occult bleeding for >6 days; and 7) use of high‐dose corticosteroids. This level of clinical detail would likely only be available through a prospective study design, as has been suggested by other authors.33 Further research into the use, safety, and effectiveness of PPIs specifically for stress ulcer prophylaxis is warranted.

In conclusion, we found that 73% of nearly 1 million Medicine patients discharged from academic medical centers received a PPI without a valid indication during their hospitalization. The implications of our findings are broad. PPIs are more expensive31 than H2RAs and there is increasing evidence that they have significant side effects. In both databases we examined, the rate of C. difficile infection was higher in patients receiving PPIs than others. The prescribing habits of physicians in these university hospital settings appear to be far out of line with published guidelines and evidence‐based practice. Reducing inappropriate prescribing of PPIs would be an important educational and quality assurance project in most institutions.