The disclosure statement for the following article, Investing in the Future: Building an Academic Hospitalist Faculty Development Program, by Niraj L. Sehgal, MD, MPH, Bradley A. Sharpe, MD, Andrew A. Auerbach, MD, MPH, Robert M. Wachter, MD, that published in Volume 6, Issue 3 pages 161166 of the Journal of Hospital Medicine, was incorrect. The correct disclosure statement is: All authors report no relevant conflicts of interest. The publisher regrets this error.

The disclosure statement for the following article, Investing in the Future: Building an Academic Hospitalist Faculty Development Program, by Niraj L. Sehgal, MD, MPH, Bradley A. Sharpe, MD, Andrew A. Auerbach, MD, MPH, Robert M. Wachter, MD, that published in Volume 6, Issue 3 pages 161166 of the Journal of Hospital Medicine, was incorrect. The correct disclosure statement is: All authors report no relevant conflicts of interest. The publisher regrets this error.

The disclosure statement for the following article, Investing in the Future: Building an Academic Hospitalist Faculty Development Program, by Niraj L. Sehgal, MD, MPH, Bradley A. Sharpe, MD, Andrew A. Auerbach, MD, MPH, Robert M. Wachter, MD, that published in Volume 6, Issue 3 pages 161166 of the Journal of Hospital Medicine, was incorrect. The correct disclosure statement is: All authors report no relevant conflicts of interest. The publisher regrets this error.

Venous thromboembolism (VTE) is a major source of morbidity and mortality for hospitalized patients. Among medical patients at the highest risk, as many as 15% can be expected to develop a VTE during their hospital stay1, 2; however, among general medical patients, the incidence of symptomatic VTE is less than 1%,1 and potentially as low as 0.3%.3 Thromboprophylaxis with subcutaneous heparin reduces the risk of VTE by approximately 50%,4 and is therefore recommended for medical patients at high risk. However, heparin also increases the risk of bleeding and thrombocytopenia and thus should be avoided for patients at low risk of VTE. Consequently, the Joint Commission on Accreditation of Healthcare Organizations (JCAHO) recommends that all hospitalized medical patients receive a risk assessment for VTE.5

Certain disease states, including stroke, acute myocardial infarction, heart failure, respiratory disease, sepsis, and cancer, have been associated with increased risk for VTE, and, based on the inclusion criteria of several randomized trials, current American College of Chest Physicians (ACCP) guidelines recommend thromboprophylaxis for patients hospitalized with these diagnoses.2 However, evidence that these factors actually increase a patient's risk for VTE comes from studies of ambulatory patients and is often weak or conflicting. Existing risk‐stratification tools,6, 7 as well as the ACCP guidelines, have not been validated, and accordingly JCAHO does not specify how risk assessment should be conducted. In order to help clinicians better estimate the risk of VTE in medical patients and therefore to provide more targeted thromboprophylaxis, we examined a large cohort of patients with high‐risk diagnoses and created a risk stratification model.

Methods

Results

Our sample contained 242,738 patients, 194,198 (80%) assigned to the derivation set and 48,540 (20%) to the validation set. Patient characteristics were similar in both sets (Supporting Information Appendix Table 1). Most patients were over age 65, 59% were female, and 64% were white (Table 1). The most common primary diagnoses were pneumonia (33%) and congestive heart failure (19%). The most common comorbidities were hypertension (50%), diabetes (31%), chronic pulmonary disease (30%), and anemia (20%). Most patients were cared for by internists (54%) or family practitioners (21%), and 30% received some form of anticoagulant VTE prophylaxis (Table 2). Of patients with an ICD‐9 code for VTE during hospitalization, just over half lacked either diagnostic testing, treatment, or both, leaving 612 (0.25%) patients who fulfilled our criteria for VTE; an additional 440 (0.18%) were readmitted for VTE, for an overall incidence of 0.43%. Patients with a length of stay 6 days had an incidence of 0.79% vs 0.19% for patients with shorter stays.

Risk factors for VTE

A large number of patient and hospital factors were associated with the development of VTE (Table 1). Due to the large sample size, even weak associations appear highly statistically significant. Compared to patients without VTE, those with VTE were more likely to have received VTE prophylaxis (37% vs 30%, P < 0.001). However, models of patients receiving prophylaxis and of patients not receiving prophylaxis produced similar odds ratios for the various risk factors (Supporting Information Appendix Table 2); therefore, the final model includes both patients who did, and did not, receive VTE prophylaxis. In the multivariable model (Supporting Information Appendix Table 3), age, length of stay, gender, primary diagnosis, cancer, inflammatory bowel disease, obesity, central venous catheter, inherited thrombophilia, steroid use, mechanical ventilation, active chemotherapy, and urinary catheters were all associated with VTE (Table 3). The strongest risk factors were length of stay 6 days (OR 3.22, 95% CI 2.73, 3.79), central venous catheter (OR 1.87, 95% CI 1.52, 2.29), inflammatory bowel disease (OR 3.11, 95% CI 1.59, 6.08), and inherited thrombophilia (OR 4.00, 95% CI 0.98, 16.40). In addition, there were important interactions between age and cancer; cancer was a strong risk factor among younger patients, but is not as strong a risk factor among older patients (OR compared to young patients without cancer was 4.62 (95% CI 2.72, 7.87) for those age 1849 years, and 3.64 (95% CI 2.52, 5.25) for those aged 5064 years).

Table 3.Factors Associated Venous Thromboembolism (VTE) in Multivariable Model

Risk Factor	OR	95% CI
* For patients without cancer. Comparison group is patients aged 18‐49 years without cancer.
Any prophylaxis	0.98	(0.84, 1.14)
Female	0.85	(0.74, 0.98)
Length of stay 6 days	3.22	(2.73, 3.79)
Age*
18‐49 years	1	Referent
50‐64 years	1.15	(0.86, 1.56)
>65 years	1.51	(1.17, 1.96)
Primary diagnosis
Pneumonia	1	Referent
Chronic obstructive pulmonary disease	0.57	(0.44, 0.75)
Stroke	0.84	(0.66, 1.08)
Congestive heart failure	0.86	(0.70, 1.06)
Urinary tract infection	1.19	(0.95, 1.50)
Respiratory failure	1.15	(0.85, 1.55)
Septicemia	1.11	(0.82, 1.50)
Comorbidities
Inflammatory bowel disease	3.11	(1.59, 6.08)
Obesity	1.28	(0.99, 1.66)
Inherited thrombophilia	4.00	(0.98, 16.40)
Cancer
18‐49 years	4.62	(2.72, 7.87)
50‐64 years	3.64	(2.52, 5.25)
>65 years	2.17	(1.61, 2.92)
Treatments
Central venous catheter	1.87	(1.52, 2.29)
Mechanical ventilation	1.61	(1.27, 2.05)
Urinary catheter	1.17	(0.99, 1.38)
Chemotherapy	1.71	(1.03, 2.83)
Steroids	1.22	(1.04, 1.43)

In the derivation set, the multivariable model produced deciles of mean predicted risk from 0.11% to 1.45%, while mean observed risk over the same deciles ranged from 0.12% to 1.42% (Figure 1). Within the validation cohort, the observed rate of VTE was 0.46% (223 cases among 48,543 subjects). The expected rate according to the model was 0.43% (expected/observed ratio: 0.93 [95% CI 0.82, 1.06]). Model discrimination measured by the c‐statistic in the validation set was 0.75 (95% CI 0.71, 0.78). The model produced deciles of mean predicted risk from 0.11% to 1.46%, with mean observed risk over the same deciles from 0.17% to 1.81%. Risk gradient was relatively flat across the first 6 deciles, began to rise at the seventh decile, and rose sharply in the highest one. Using a risk threshold of 1%, the model had a sensitivity of 28% and a specificity of 93%. In the validation set, this translated into a positive predictive value of 2.2% and a negative predictive value of 99.7%. Assuming that VTE prophylaxis has an efficacy of 50%, the number‐needed‐to‐treat to prevent one VTE among high‐risk patients (predicted risk >1%) would be 91. In contrast, providing prophylaxis to the entire validation sample would result in a number‐needed‐to‐treat of 435. Using a lower treatment threshold of 0.4% produced a positive predictive value of 1% and a negative predictive value of 99.8%. At this threshold, the model would detect 73% of patients with VTE and the number‐needed‐to‐treat to prevent one VTE would be 200.

Figure 1

Discussion

In a representative sample of 243,000 hospitalized medical patients with at least one major risk factor for VTE, we found that symptomatic VTE was an uncommon event, occurring in approximately 1 in 231 patients. We identified a number of factors that were associated with an increased risk of VTE, but many previously cited risk factors did not show an association in multivariable models. In particular, patients with a primary diagnosis of COPD appeared not to share the same high risk of VTE as patients with the other diagnoses we examined, a finding reported by others.11 The risk model we developed accurately stratifies patients across a wide range of VTE probabilities, but even among those with the highest predicted rates, symptomatic VTE occurred in less than 2%.

VTE is often described as a frequent complication of hospitalization for medical illness and one of the most common potentially preventable causes of death. Indeed, rates of asymptomatic VTE have been demonstrated to be 3.7% to 26%.12 Although some of these might have fatal consequences, most are distal vein thromboses and their significance is unknown. In contrast, symptomatic events are uncommon, with previous estimates among general medical patients in observational studies in the range of 0.3%3 to 0.8%,12 similar to the rate observed in our study. Symptomatic event rates among control patients in landmark randomized trials have ranged from 0.86%13 to 2.3%,14 but these studies enrolled only very high‐risk patients with more extended hospitalizations, and may involve follow‐up periods of a month or more.

Because it is unlikely that our diagnostic algorithm was 100% sensitive, and because 30% of our patients received chemoprophylaxis, it is probable that we have underestimated the true rate of VTE in our sample. Among the patients who received prophylaxis, the observed rate of VTE was 0.54%. If we assume that prophylaxis is 50% effective, then had these patients not received prophylaxis, their rate of VTE would have been 1.08% (vs 0.39% among those patients who received no prophylaxis) and the overall rate of VTE for the sample would have been 0.60% (1.08 0.30 + 0.39 0.70). If we further assume that our algorithm was only 80% sensitive and 100% specific, the true underlying rate of symptomatic VTE could have been as high as 0.75%, still less than half that seen in randomized trials.

Prophylaxis with heparin has been shown to decrease the rate of both asymptomatic and symptomatic events, but because of the low prevalence, the number‐needed‐to‐treat to prevent one symptomatic pulmonary embolism has been estimated at 345, and prophylaxis has not been shown to affect all‐cause mortality.4, 15 At the same time, prophylaxis costs money, is uncomfortable, and carries a small risk of bleeding and heparin‐induced thrombocytopenia. Given the generally low incidence of symptomatic VTE, it therefore makes sense to reserve prophylaxis for patients at higher risk of thromboembolism.

To decide whether prophylaxis is appropriate for a given patient, it is necessary to quantify the patient's risk and then apply an appropriate threshold for treatment. The National Quality Forum (NQF) recommends,16 and JCAHO has adopted, that a clinician must evaluate each patient upon admission, and regularly thereafter, for the risk of developing DVT [deep vein thrombosis]/VTE. Until now, however, there has been no widely accepted, validated method to risk stratify medical patients. The ACCP recommendations cite just three studies of VTE risk factors in hospitalized medical patients.11, 17, 18 Together they examined 477 cases and 1197 controls, identifying congestive heart failure, pneumonia, cancer, and previous VTE as risk factors. Predictive models based on these factors17, 1921 have not been subjected to validation or have performed poorly.18 Acknowledging this lack of standardized risk assessment, JCAHO leaves the means of assessment to individual hospitals. A quality improvement guide published by the Agency for Healthcare Research and Quality goes one step further, stating that In a typical hospital, it is estimated that fewer than 5% of medical patients could be considered at low risk by most VTE risk stratification methods.22 The guide recommends near universal VTE prophylaxis.

In light of the JCAHO requirements, our model should be welcomed by hospitalists. Rather than assuming that all patients over 40 years of age are at high risk, our model will enable clinicians to risk stratify patients from a low of 0.1% to >1.4% (>10‐fold increase in risk). Moreover, the model was derived from more than 800 episodes of symptomatic VTE among almost 190,000 general medical patients and validated on almost 50,000 more. The observed patients were cared for in clinical practice at a nationally representative group of US hospitals, not in a highly selected clinical trial, increasing the generalizability of our findings. Finally, the model includes ten common risk factors that can easily be entered into decision support software or extracted automatically from the electronic medical record. Electronic reminder systems have already been shown to increase use of VTE prophylaxis, and prevent VTE, especially among cancer patients.23

A more challenging task is defining the appropriate risk threshold to initiate VTE prophylaxis. The Thromboembolic Risk Factors (THRIFT) Consensus Group classified patients according to risk of proximal DVT as low (<1%), moderate (1%‐10%), and high (>10%).21 They recommended heparin prophylaxis for all patients at moderate risk or higher. Although the patients included in our study all had a diagnosis that warranted prophylaxis according to the ACCP guidelines, using the THRIFT threshold for moderate‐to‐high risk, only 7% of our patients should have received prophylaxis. The recommendation not to offer heparin prophylaxis to patients with less than 1% chance of developing symptomatic VTE seems reasonable, given the large number‐needed‐to‐treat, but formal decision analyses should be conducted to better define this threshold. Many hospitalists, however, may feel uncomfortable using the 1% threshold, because our model failed to identify almost three out of four patients who ultimately experienced symptomatic VTE. At that threshold, it would seem that hospital‐acquired VTE is not a preventable complication in most medical patients, as others have pointed out.3, 24 Alternatively, if the threshold were lowered to 0.4%, our model could reduce the use of prophylaxis by 60%, while still identifying three‐fourths of all VTE cases. Further research is needed to know whether such a threshold is reasonable.

Our study has a number of important limitations. First, we relied on claims data, not chart review. We do not know for certain which patients experienced VTE, although our definition of VTE required diagnosis codes plus charges for both diagnosis and treatment. Moreover, our rates are similar to those observed in other trials where symptomatic events were confirmed. Second, about 30% of our patients received at least some VTE prophylaxis, and this may have prevented as many as half of the VTEs in that group. Without prophylaxis, rates might have been 20%30% higher. Similarly, we could not detect patients who were diagnosed after discharge but not admitted to hospital. While we believe this number to be small, it would again increase the rate slightly. Third, we could not assess certain clinical circumstances that are not associated with hospital charges or diagnosis codes, especially prolonged bed rest. Other risk factors, such as the urinary catheter, were probably surrogate markers for immobilization rather than true risk factors. Fourth, we included length of stay in our prediction model. We did this because most randomized trials of VTE prophylaxis included only patients with an expected length of stay 6 days. Physicians' estimates about probable length of stay may be less accurate than actual length of stay as a predictor of VTE. Moreover, the relationship may have been confounded if hospital‐acquired VTE led to longer lengths of stay. We think this unlikely since many of the events were discovered on readmission. Fifth, we studied only patients carrying high‐risk diagnoses, and therefore do not know the baseline risk for patients with less risky conditions, although it should be lower than what we observed. It seems probable that COPD, rather than being protective, as it appears in our model, actually represents the baseline risk for low‐risk diagnoses. It should be noted that we did include a number of other high‐risk diagnoses, such as cancer and inflammatory bowel disease, as secondary diagnoses. A larger, more inclusive study should be conducted to validate our model in other populations. Finally, we cannot know who died of undiagnosed VTE, either in the hospital or after discharge. Such an outcome would be important, but those events are likely to be rare, and VTE prophylaxis has not been shown to affect mortality.

VTE remains a daunting problem in hospitalized medical patients. Although VTE is responsible for a large number of hospital deaths each year, identifying patients at high risk for clinically important VTE is challenging, and may contribute to the persistently low rates of VTE prophylaxis seen in hospitals.25 Current efforts to treat nearly all patients are likely to lead to unnecessary cost, discomfort, and side effects. We present a simple logistic regression model that can easily identify patients at moderate‐to‐high risk (>1%) of developing symptomatic VTE. Future studies should focus on prospectively validating the model in a wider spectrum of medical illness, and better defining the appropriate risk cutoff for general prophylaxis.

Venous thromboembolism (VTE) is a major source of morbidity and mortality for hospitalized patients. Among medical patients at the highest risk, as many as 15% can be expected to develop a VTE during their hospital stay1, 2; however, among general medical patients, the incidence of symptomatic VTE is less than 1%,1 and potentially as low as 0.3%.3 Thromboprophylaxis with subcutaneous heparin reduces the risk of VTE by approximately 50%,4 and is therefore recommended for medical patients at high risk. However, heparin also increases the risk of bleeding and thrombocytopenia and thus should be avoided for patients at low risk of VTE. Consequently, the Joint Commission on Accreditation of Healthcare Organizations (JCAHO) recommends that all hospitalized medical patients receive a risk assessment for VTE.5

Certain disease states, including stroke, acute myocardial infarction, heart failure, respiratory disease, sepsis, and cancer, have been associated with increased risk for VTE, and, based on the inclusion criteria of several randomized trials, current American College of Chest Physicians (ACCP) guidelines recommend thromboprophylaxis for patients hospitalized with these diagnoses.2 However, evidence that these factors actually increase a patient's risk for VTE comes from studies of ambulatory patients and is often weak or conflicting. Existing risk‐stratification tools,6, 7 as well as the ACCP guidelines, have not been validated, and accordingly JCAHO does not specify how risk assessment should be conducted. In order to help clinicians better estimate the risk of VTE in medical patients and therefore to provide more targeted thromboprophylaxis, we examined a large cohort of patients with high‐risk diagnoses and created a risk stratification model.

Methods

Results

Our sample contained 242,738 patients, 194,198 (80%) assigned to the derivation set and 48,540 (20%) to the validation set. Patient characteristics were similar in both sets (Supporting Information Appendix Table 1). Most patients were over age 65, 59% were female, and 64% were white (Table 1). The most common primary diagnoses were pneumonia (33%) and congestive heart failure (19%). The most common comorbidities were hypertension (50%), diabetes (31%), chronic pulmonary disease (30%), and anemia (20%). Most patients were cared for by internists (54%) or family practitioners (21%), and 30% received some form of anticoagulant VTE prophylaxis (Table 2). Of patients with an ICD‐9 code for VTE during hospitalization, just over half lacked either diagnostic testing, treatment, or both, leaving 612 (0.25%) patients who fulfilled our criteria for VTE; an additional 440 (0.18%) were readmitted for VTE, for an overall incidence of 0.43%. Patients with a length of stay 6 days had an incidence of 0.79% vs 0.19% for patients with shorter stays.

Risk factors for VTE

A large number of patient and hospital factors were associated with the development of VTE (Table 1). Due to the large sample size, even weak associations appear highly statistically significant. Compared to patients without VTE, those with VTE were more likely to have received VTE prophylaxis (37% vs 30%, P < 0.001). However, models of patients receiving prophylaxis and of patients not receiving prophylaxis produced similar odds ratios for the various risk factors (Supporting Information Appendix Table 2); therefore, the final model includes both patients who did, and did not, receive VTE prophylaxis. In the multivariable model (Supporting Information Appendix Table 3), age, length of stay, gender, primary diagnosis, cancer, inflammatory bowel disease, obesity, central venous catheter, inherited thrombophilia, steroid use, mechanical ventilation, active chemotherapy, and urinary catheters were all associated with VTE (Table 3). The strongest risk factors were length of stay 6 days (OR 3.22, 95% CI 2.73, 3.79), central venous catheter (OR 1.87, 95% CI 1.52, 2.29), inflammatory bowel disease (OR 3.11, 95% CI 1.59, 6.08), and inherited thrombophilia (OR 4.00, 95% CI 0.98, 16.40). In addition, there were important interactions between age and cancer; cancer was a strong risk factor among younger patients, but is not as strong a risk factor among older patients (OR compared to young patients without cancer was 4.62 (95% CI 2.72, 7.87) for those age 1849 years, and 3.64 (95% CI 2.52, 5.25) for those aged 5064 years).

Table 3.Factors Associated Venous Thromboembolism (VTE) in Multivariable Model

Risk Factor	OR	95% CI
* For patients without cancer. Comparison group is patients aged 18‐49 years without cancer.
Any prophylaxis	0.98	(0.84, 1.14)
Female	0.85	(0.74, 0.98)
Length of stay 6 days	3.22	(2.73, 3.79)
Age*
18‐49 years	1	Referent
50‐64 years	1.15	(0.86, 1.56)
>65 years	1.51	(1.17, 1.96)
Primary diagnosis
Pneumonia	1	Referent
Chronic obstructive pulmonary disease	0.57	(0.44, 0.75)
Stroke	0.84	(0.66, 1.08)
Congestive heart failure	0.86	(0.70, 1.06)
Urinary tract infection	1.19	(0.95, 1.50)
Respiratory failure	1.15	(0.85, 1.55)
Septicemia	1.11	(0.82, 1.50)
Comorbidities
Inflammatory bowel disease	3.11	(1.59, 6.08)
Obesity	1.28	(0.99, 1.66)
Inherited thrombophilia	4.00	(0.98, 16.40)
Cancer
18‐49 years	4.62	(2.72, 7.87)
50‐64 years	3.64	(2.52, 5.25)
>65 years	2.17	(1.61, 2.92)
Treatments
Central venous catheter	1.87	(1.52, 2.29)
Mechanical ventilation	1.61	(1.27, 2.05)
Urinary catheter	1.17	(0.99, 1.38)
Chemotherapy	1.71	(1.03, 2.83)
Steroids	1.22	(1.04, 1.43)

In the derivation set, the multivariable model produced deciles of mean predicted risk from 0.11% to 1.45%, while mean observed risk over the same deciles ranged from 0.12% to 1.42% (Figure 1). Within the validation cohort, the observed rate of VTE was 0.46% (223 cases among 48,543 subjects). The expected rate according to the model was 0.43% (expected/observed ratio: 0.93 [95% CI 0.82, 1.06]). Model discrimination measured by the c‐statistic in the validation set was 0.75 (95% CI 0.71, 0.78). The model produced deciles of mean predicted risk from 0.11% to 1.46%, with mean observed risk over the same deciles from 0.17% to 1.81%. Risk gradient was relatively flat across the first 6 deciles, began to rise at the seventh decile, and rose sharply in the highest one. Using a risk threshold of 1%, the model had a sensitivity of 28% and a specificity of 93%. In the validation set, this translated into a positive predictive value of 2.2% and a negative predictive value of 99.7%. Assuming that VTE prophylaxis has an efficacy of 50%, the number‐needed‐to‐treat to prevent one VTE among high‐risk patients (predicted risk >1%) would be 91. In contrast, providing prophylaxis to the entire validation sample would result in a number‐needed‐to‐treat of 435. Using a lower treatment threshold of 0.4% produced a positive predictive value of 1% and a negative predictive value of 99.8%. At this threshold, the model would detect 73% of patients with VTE and the number‐needed‐to‐treat to prevent one VTE would be 200.

Figure 1

Discussion

In a representative sample of 243,000 hospitalized medical patients with at least one major risk factor for VTE, we found that symptomatic VTE was an uncommon event, occurring in approximately 1 in 231 patients. We identified a number of factors that were associated with an increased risk of VTE, but many previously cited risk factors did not show an association in multivariable models. In particular, patients with a primary diagnosis of COPD appeared not to share the same high risk of VTE as patients with the other diagnoses we examined, a finding reported by others.11 The risk model we developed accurately stratifies patients across a wide range of VTE probabilities, but even among those with the highest predicted rates, symptomatic VTE occurred in less than 2%.

VTE is often described as a frequent complication of hospitalization for medical illness and one of the most common potentially preventable causes of death. Indeed, rates of asymptomatic VTE have been demonstrated to be 3.7% to 26%.12 Although some of these might have fatal consequences, most are distal vein thromboses and their significance is unknown. In contrast, symptomatic events are uncommon, with previous estimates among general medical patients in observational studies in the range of 0.3%3 to 0.8%,12 similar to the rate observed in our study. Symptomatic event rates among control patients in landmark randomized trials have ranged from 0.86%13 to 2.3%,14 but these studies enrolled only very high‐risk patients with more extended hospitalizations, and may involve follow‐up periods of a month or more.

Because it is unlikely that our diagnostic algorithm was 100% sensitive, and because 30% of our patients received chemoprophylaxis, it is probable that we have underestimated the true rate of VTE in our sample. Among the patients who received prophylaxis, the observed rate of VTE was 0.54%. If we assume that prophylaxis is 50% effective, then had these patients not received prophylaxis, their rate of VTE would have been 1.08% (vs 0.39% among those patients who received no prophylaxis) and the overall rate of VTE for the sample would have been 0.60% (1.08 0.30 + 0.39 0.70). If we further assume that our algorithm was only 80% sensitive and 100% specific, the true underlying rate of symptomatic VTE could have been as high as 0.75%, still less than half that seen in randomized trials.

Prophylaxis with heparin has been shown to decrease the rate of both asymptomatic and symptomatic events, but because of the low prevalence, the number‐needed‐to‐treat to prevent one symptomatic pulmonary embolism has been estimated at 345, and prophylaxis has not been shown to affect all‐cause mortality.4, 15 At the same time, prophylaxis costs money, is uncomfortable, and carries a small risk of bleeding and heparin‐induced thrombocytopenia. Given the generally low incidence of symptomatic VTE, it therefore makes sense to reserve prophylaxis for patients at higher risk of thromboembolism.

To decide whether prophylaxis is appropriate for a given patient, it is necessary to quantify the patient's risk and then apply an appropriate threshold for treatment. The National Quality Forum (NQF) recommends,16 and JCAHO has adopted, that a clinician must evaluate each patient upon admission, and regularly thereafter, for the risk of developing DVT [deep vein thrombosis]/VTE. Until now, however, there has been no widely accepted, validated method to risk stratify medical patients. The ACCP recommendations cite just three studies of VTE risk factors in hospitalized medical patients.11, 17, 18 Together they examined 477 cases and 1197 controls, identifying congestive heart failure, pneumonia, cancer, and previous VTE as risk factors. Predictive models based on these factors17, 1921 have not been subjected to validation or have performed poorly.18 Acknowledging this lack of standardized risk assessment, JCAHO leaves the means of assessment to individual hospitals. A quality improvement guide published by the Agency for Healthcare Research and Quality goes one step further, stating that In a typical hospital, it is estimated that fewer than 5% of medical patients could be considered at low risk by most VTE risk stratification methods.22 The guide recommends near universal VTE prophylaxis.

In light of the JCAHO requirements, our model should be welcomed by hospitalists. Rather than assuming that all patients over 40 years of age are at high risk, our model will enable clinicians to risk stratify patients from a low of 0.1% to >1.4% (>10‐fold increase in risk). Moreover, the model was derived from more than 800 episodes of symptomatic VTE among almost 190,000 general medical patients and validated on almost 50,000 more. The observed patients were cared for in clinical practice at a nationally representative group of US hospitals, not in a highly selected clinical trial, increasing the generalizability of our findings. Finally, the model includes ten common risk factors that can easily be entered into decision support software or extracted automatically from the electronic medical record. Electronic reminder systems have already been shown to increase use of VTE prophylaxis, and prevent VTE, especially among cancer patients.23

A more challenging task is defining the appropriate risk threshold to initiate VTE prophylaxis. The Thromboembolic Risk Factors (THRIFT) Consensus Group classified patients according to risk of proximal DVT as low (<1%), moderate (1%‐10%), and high (>10%).21 They recommended heparin prophylaxis for all patients at moderate risk or higher. Although the patients included in our study all had a diagnosis that warranted prophylaxis according to the ACCP guidelines, using the THRIFT threshold for moderate‐to‐high risk, only 7% of our patients should have received prophylaxis. The recommendation not to offer heparin prophylaxis to patients with less than 1% chance of developing symptomatic VTE seems reasonable, given the large number‐needed‐to‐treat, but formal decision analyses should be conducted to better define this threshold. Many hospitalists, however, may feel uncomfortable using the 1% threshold, because our model failed to identify almost three out of four patients who ultimately experienced symptomatic VTE. At that threshold, it would seem that hospital‐acquired VTE is not a preventable complication in most medical patients, as others have pointed out.3, 24 Alternatively, if the threshold were lowered to 0.4%, our model could reduce the use of prophylaxis by 60%, while still identifying three‐fourths of all VTE cases. Further research is needed to know whether such a threshold is reasonable.

Our study has a number of important limitations. First, we relied on claims data, not chart review. We do not know for certain which patients experienced VTE, although our definition of VTE required diagnosis codes plus charges for both diagnosis and treatment. Moreover, our rates are similar to those observed in other trials where symptomatic events were confirmed. Second, about 30% of our patients received at least some VTE prophylaxis, and this may have prevented as many as half of the VTEs in that group. Without prophylaxis, rates might have been 20%30% higher. Similarly, we could not detect patients who were diagnosed after discharge but not admitted to hospital. While we believe this number to be small, it would again increase the rate slightly. Third, we could not assess certain clinical circumstances that are not associated with hospital charges or diagnosis codes, especially prolonged bed rest. Other risk factors, such as the urinary catheter, were probably surrogate markers for immobilization rather than true risk factors. Fourth, we included length of stay in our prediction model. We did this because most randomized trials of VTE prophylaxis included only patients with an expected length of stay 6 days. Physicians' estimates about probable length of stay may be less accurate than actual length of stay as a predictor of VTE. Moreover, the relationship may have been confounded if hospital‐acquired VTE led to longer lengths of stay. We think this unlikely since many of the events were discovered on readmission. Fifth, we studied only patients carrying high‐risk diagnoses, and therefore do not know the baseline risk for patients with less risky conditions, although it should be lower than what we observed. It seems probable that COPD, rather than being protective, as it appears in our model, actually represents the baseline risk for low‐risk diagnoses. It should be noted that we did include a number of other high‐risk diagnoses, such as cancer and inflammatory bowel disease, as secondary diagnoses. A larger, more inclusive study should be conducted to validate our model in other populations. Finally, we cannot know who died of undiagnosed VTE, either in the hospital or after discharge. Such an outcome would be important, but those events are likely to be rare, and VTE prophylaxis has not been shown to affect mortality.

VTE remains a daunting problem in hospitalized medical patients. Although VTE is responsible for a large number of hospital deaths each year, identifying patients at high risk for clinically important VTE is challenging, and may contribute to the persistently low rates of VTE prophylaxis seen in hospitals.25 Current efforts to treat nearly all patients are likely to lead to unnecessary cost, discomfort, and side effects. We present a simple logistic regression model that can easily identify patients at moderate‐to‐high risk (>1%) of developing symptomatic VTE. Future studies should focus on prospectively validating the model in a wider spectrum of medical illness, and better defining the appropriate risk cutoff for general prophylaxis.

Venous thromboembolism (VTE) is a major source of morbidity and mortality for hospitalized patients. Among medical patients at the highest risk, as many as 15% can be expected to develop a VTE during their hospital stay1, 2; however, among general medical patients, the incidence of symptomatic VTE is less than 1%,1 and potentially as low as 0.3%.3 Thromboprophylaxis with subcutaneous heparin reduces the risk of VTE by approximately 50%,4 and is therefore recommended for medical patients at high risk. However, heparin also increases the risk of bleeding and thrombocytopenia and thus should be avoided for patients at low risk of VTE. Consequently, the Joint Commission on Accreditation of Healthcare Organizations (JCAHO) recommends that all hospitalized medical patients receive a risk assessment for VTE.5

Certain disease states, including stroke, acute myocardial infarction, heart failure, respiratory disease, sepsis, and cancer, have been associated with increased risk for VTE, and, based on the inclusion criteria of several randomized trials, current American College of Chest Physicians (ACCP) guidelines recommend thromboprophylaxis for patients hospitalized with these diagnoses.2 However, evidence that these factors actually increase a patient's risk for VTE comes from studies of ambulatory patients and is often weak or conflicting. Existing risk‐stratification tools,6, 7 as well as the ACCP guidelines, have not been validated, and accordingly JCAHO does not specify how risk assessment should be conducted. In order to help clinicians better estimate the risk of VTE in medical patients and therefore to provide more targeted thromboprophylaxis, we examined a large cohort of patients with high‐risk diagnoses and created a risk stratification model.

Methods

Results

Our sample contained 242,738 patients, 194,198 (80%) assigned to the derivation set and 48,540 (20%) to the validation set. Patient characteristics were similar in both sets (Supporting Information Appendix Table 1). Most patients were over age 65, 59% were female, and 64% were white (Table 1). The most common primary diagnoses were pneumonia (33%) and congestive heart failure (19%). The most common comorbidities were hypertension (50%), diabetes (31%), chronic pulmonary disease (30%), and anemia (20%). Most patients were cared for by internists (54%) or family practitioners (21%), and 30% received some form of anticoagulant VTE prophylaxis (Table 2). Of patients with an ICD‐9 code for VTE during hospitalization, just over half lacked either diagnostic testing, treatment, or both, leaving 612 (0.25%) patients who fulfilled our criteria for VTE; an additional 440 (0.18%) were readmitted for VTE, for an overall incidence of 0.43%. Patients with a length of stay 6 days had an incidence of 0.79% vs 0.19% for patients with shorter stays.

Risk factors for VTE

A large number of patient and hospital factors were associated with the development of VTE (Table 1). Due to the large sample size, even weak associations appear highly statistically significant. Compared to patients without VTE, those with VTE were more likely to have received VTE prophylaxis (37% vs 30%, P < 0.001). However, models of patients receiving prophylaxis and of patients not receiving prophylaxis produced similar odds ratios for the various risk factors (Supporting Information Appendix Table 2); therefore, the final model includes both patients who did, and did not, receive VTE prophylaxis. In the multivariable model (Supporting Information Appendix Table 3), age, length of stay, gender, primary diagnosis, cancer, inflammatory bowel disease, obesity, central venous catheter, inherited thrombophilia, steroid use, mechanical ventilation, active chemotherapy, and urinary catheters were all associated with VTE (Table 3). The strongest risk factors were length of stay 6 days (OR 3.22, 95% CI 2.73, 3.79), central venous catheter (OR 1.87, 95% CI 1.52, 2.29), inflammatory bowel disease (OR 3.11, 95% CI 1.59, 6.08), and inherited thrombophilia (OR 4.00, 95% CI 0.98, 16.40). In addition, there were important interactions between age and cancer; cancer was a strong risk factor among younger patients, but is not as strong a risk factor among older patients (OR compared to young patients without cancer was 4.62 (95% CI 2.72, 7.87) for those age 1849 years, and 3.64 (95% CI 2.52, 5.25) for those aged 5064 years).

Table 3.Factors Associated Venous Thromboembolism (VTE) in Multivariable Model

Risk Factor	OR	95% CI
* For patients without cancer. Comparison group is patients aged 18‐49 years without cancer.
Any prophylaxis	0.98	(0.84, 1.14)
Female	0.85	(0.74, 0.98)
Length of stay 6 days	3.22	(2.73, 3.79)
Age*
18‐49 years	1	Referent
50‐64 years	1.15	(0.86, 1.56)
>65 years	1.51	(1.17, 1.96)
Primary diagnosis
Pneumonia	1	Referent
Chronic obstructive pulmonary disease	0.57	(0.44, 0.75)
Stroke	0.84	(0.66, 1.08)
Congestive heart failure	0.86	(0.70, 1.06)
Urinary tract infection	1.19	(0.95, 1.50)
Respiratory failure	1.15	(0.85, 1.55)
Septicemia	1.11	(0.82, 1.50)
Comorbidities
Inflammatory bowel disease	3.11	(1.59, 6.08)
Obesity	1.28	(0.99, 1.66)
Inherited thrombophilia	4.00	(0.98, 16.40)
Cancer
18‐49 years	4.62	(2.72, 7.87)
50‐64 years	3.64	(2.52, 5.25)
>65 years	2.17	(1.61, 2.92)
Treatments
Central venous catheter	1.87	(1.52, 2.29)
Mechanical ventilation	1.61	(1.27, 2.05)
Urinary catheter	1.17	(0.99, 1.38)
Chemotherapy	1.71	(1.03, 2.83)
Steroids	1.22	(1.04, 1.43)

In the derivation set, the multivariable model produced deciles of mean predicted risk from 0.11% to 1.45%, while mean observed risk over the same deciles ranged from 0.12% to 1.42% (Figure 1). Within the validation cohort, the observed rate of VTE was 0.46% (223 cases among 48,543 subjects). The expected rate according to the model was 0.43% (expected/observed ratio: 0.93 [95% CI 0.82, 1.06]). Model discrimination measured by the c‐statistic in the validation set was 0.75 (95% CI 0.71, 0.78). The model produced deciles of mean predicted risk from 0.11% to 1.46%, with mean observed risk over the same deciles from 0.17% to 1.81%. Risk gradient was relatively flat across the first 6 deciles, began to rise at the seventh decile, and rose sharply in the highest one. Using a risk threshold of 1%, the model had a sensitivity of 28% and a specificity of 93%. In the validation set, this translated into a positive predictive value of 2.2% and a negative predictive value of 99.7%. Assuming that VTE prophylaxis has an efficacy of 50%, the number‐needed‐to‐treat to prevent one VTE among high‐risk patients (predicted risk >1%) would be 91. In contrast, providing prophylaxis to the entire validation sample would result in a number‐needed‐to‐treat of 435. Using a lower treatment threshold of 0.4% produced a positive predictive value of 1% and a negative predictive value of 99.8%. At this threshold, the model would detect 73% of patients with VTE and the number‐needed‐to‐treat to prevent one VTE would be 200.

Figure 1

Discussion

In a representative sample of 243,000 hospitalized medical patients with at least one major risk factor for VTE, we found that symptomatic VTE was an uncommon event, occurring in approximately 1 in 231 patients. We identified a number of factors that were associated with an increased risk of VTE, but many previously cited risk factors did not show an association in multivariable models. In particular, patients with a primary diagnosis of COPD appeared not to share the same high risk of VTE as patients with the other diagnoses we examined, a finding reported by others.11 The risk model we developed accurately stratifies patients across a wide range of VTE probabilities, but even among those with the highest predicted rates, symptomatic VTE occurred in less than 2%.

VTE is often described as a frequent complication of hospitalization for medical illness and one of the most common potentially preventable causes of death. Indeed, rates of asymptomatic VTE have been demonstrated to be 3.7% to 26%.12 Although some of these might have fatal consequences, most are distal vein thromboses and their significance is unknown. In contrast, symptomatic events are uncommon, with previous estimates among general medical patients in observational studies in the range of 0.3%3 to 0.8%,12 similar to the rate observed in our study. Symptomatic event rates among control patients in landmark randomized trials have ranged from 0.86%13 to 2.3%,14 but these studies enrolled only very high‐risk patients with more extended hospitalizations, and may involve follow‐up periods of a month or more.

Because it is unlikely that our diagnostic algorithm was 100% sensitive, and because 30% of our patients received chemoprophylaxis, it is probable that we have underestimated the true rate of VTE in our sample. Among the patients who received prophylaxis, the observed rate of VTE was 0.54%. If we assume that prophylaxis is 50% effective, then had these patients not received prophylaxis, their rate of VTE would have been 1.08% (vs 0.39% among those patients who received no prophylaxis) and the overall rate of VTE for the sample would have been 0.60% (1.08 0.30 + 0.39 0.70). If we further assume that our algorithm was only 80% sensitive and 100% specific, the true underlying rate of symptomatic VTE could have been as high as 0.75%, still less than half that seen in randomized trials.

Prophylaxis with heparin has been shown to decrease the rate of both asymptomatic and symptomatic events, but because of the low prevalence, the number‐needed‐to‐treat to prevent one symptomatic pulmonary embolism has been estimated at 345, and prophylaxis has not been shown to affect all‐cause mortality.4, 15 At the same time, prophylaxis costs money, is uncomfortable, and carries a small risk of bleeding and heparin‐induced thrombocytopenia. Given the generally low incidence of symptomatic VTE, it therefore makes sense to reserve prophylaxis for patients at higher risk of thromboembolism.

To decide whether prophylaxis is appropriate for a given patient, it is necessary to quantify the patient's risk and then apply an appropriate threshold for treatment. The National Quality Forum (NQF) recommends,16 and JCAHO has adopted, that a clinician must evaluate each patient upon admission, and regularly thereafter, for the risk of developing DVT [deep vein thrombosis]/VTE. Until now, however, there has been no widely accepted, validated method to risk stratify medical patients. The ACCP recommendations cite just three studies of VTE risk factors in hospitalized medical patients.11, 17, 18 Together they examined 477 cases and 1197 controls, identifying congestive heart failure, pneumonia, cancer, and previous VTE as risk factors. Predictive models based on these factors17, 1921 have not been subjected to validation or have performed poorly.18 Acknowledging this lack of standardized risk assessment, JCAHO leaves the means of assessment to individual hospitals. A quality improvement guide published by the Agency for Healthcare Research and Quality goes one step further, stating that In a typical hospital, it is estimated that fewer than 5% of medical patients could be considered at low risk by most VTE risk stratification methods.22 The guide recommends near universal VTE prophylaxis.

In light of the JCAHO requirements, our model should be welcomed by hospitalists. Rather than assuming that all patients over 40 years of age are at high risk, our model will enable clinicians to risk stratify patients from a low of 0.1% to >1.4% (>10‐fold increase in risk). Moreover, the model was derived from more than 800 episodes of symptomatic VTE among almost 190,000 general medical patients and validated on almost 50,000 more. The observed patients were cared for in clinical practice at a nationally representative group of US hospitals, not in a highly selected clinical trial, increasing the generalizability of our findings. Finally, the model includes ten common risk factors that can easily be entered into decision support software or extracted automatically from the electronic medical record. Electronic reminder systems have already been shown to increase use of VTE prophylaxis, and prevent VTE, especially among cancer patients.23

A more challenging task is defining the appropriate risk threshold to initiate VTE prophylaxis. The Thromboembolic Risk Factors (THRIFT) Consensus Group classified patients according to risk of proximal DVT as low (<1%), moderate (1%‐10%), and high (>10%).21 They recommended heparin prophylaxis for all patients at moderate risk or higher. Although the patients included in our study all had a diagnosis that warranted prophylaxis according to the ACCP guidelines, using the THRIFT threshold for moderate‐to‐high risk, only 7% of our patients should have received prophylaxis. The recommendation not to offer heparin prophylaxis to patients with less than 1% chance of developing symptomatic VTE seems reasonable, given the large number‐needed‐to‐treat, but formal decision analyses should be conducted to better define this threshold. Many hospitalists, however, may feel uncomfortable using the 1% threshold, because our model failed to identify almost three out of four patients who ultimately experienced symptomatic VTE. At that threshold, it would seem that hospital‐acquired VTE is not a preventable complication in most medical patients, as others have pointed out.3, 24 Alternatively, if the threshold were lowered to 0.4%, our model could reduce the use of prophylaxis by 60%, while still identifying three‐fourths of all VTE cases. Further research is needed to know whether such a threshold is reasonable.

Our study has a number of important limitations. First, we relied on claims data, not chart review. We do not know for certain which patients experienced VTE, although our definition of VTE required diagnosis codes plus charges for both diagnosis and treatment. Moreover, our rates are similar to those observed in other trials where symptomatic events were confirmed. Second, about 30% of our patients received at least some VTE prophylaxis, and this may have prevented as many as half of the VTEs in that group. Without prophylaxis, rates might have been 20%30% higher. Similarly, we could not detect patients who were diagnosed after discharge but not admitted to hospital. While we believe this number to be small, it would again increase the rate slightly. Third, we could not assess certain clinical circumstances that are not associated with hospital charges or diagnosis codes, especially prolonged bed rest. Other risk factors, such as the urinary catheter, were probably surrogate markers for immobilization rather than true risk factors. Fourth, we included length of stay in our prediction model. We did this because most randomized trials of VTE prophylaxis included only patients with an expected length of stay 6 days. Physicians' estimates about probable length of stay may be less accurate than actual length of stay as a predictor of VTE. Moreover, the relationship may have been confounded if hospital‐acquired VTE led to longer lengths of stay. We think this unlikely since many of the events were discovered on readmission. Fifth, we studied only patients carrying high‐risk diagnoses, and therefore do not know the baseline risk for patients with less risky conditions, although it should be lower than what we observed. It seems probable that COPD, rather than being protective, as it appears in our model, actually represents the baseline risk for low‐risk diagnoses. It should be noted that we did include a number of other high‐risk diagnoses, such as cancer and inflammatory bowel disease, as secondary diagnoses. A larger, more inclusive study should be conducted to validate our model in other populations. Finally, we cannot know who died of undiagnosed VTE, either in the hospital or after discharge. Such an outcome would be important, but those events are likely to be rare, and VTE prophylaxis has not been shown to affect mortality.

VTE remains a daunting problem in hospitalized medical patients. Although VTE is responsible for a large number of hospital deaths each year, identifying patients at high risk for clinically important VTE is challenging, and may contribute to the persistently low rates of VTE prophylaxis seen in hospitals.25 Current efforts to treat nearly all patients are likely to lead to unnecessary cost, discomfort, and side effects. We present a simple logistic regression model that can easily identify patients at moderate‐to‐high risk (>1%) of developing symptomatic VTE. Future studies should focus on prospectively validating the model in a wider spectrum of medical illness, and better defining the appropriate risk cutoff for general prophylaxis.

Symptoms, signs, chest radiograms, electrocardiograms and laboratory data have a low specificity for the diagnosis of pulmonary embolism (PE) when used in isolation, but when used in combination they can accurately identify patients with an increased likelihood of having a PE.17 The Wells score combines multiple variables into a prediction tool (Table 1). The original model identified three categories of patients with increasing likelihoods of having a PE,6 but a simpler, dichotomous version was subsequently proposed.7 A sequential diagnostic strategy combining the dichotomous Wells rule with a serum d‐dimer test has been validated against contrast‐enhanced spiral computed tomography (CTPE) on cohorts comprised largely of ambulatory outpatient and emergency room patients.815 This method, however, has never been tested in hospitalized patients who were receiving heparin in doses designed to prevent the development of venous thromboembolism (VTE). The purpose of this study was to evaluate the utility of the modified Wells score to predict the presence or absence of PE in hospitalized patients who were receiving prophylactic heparin.

Methods

We screened consecutive patients who underwent CTPE studies from January 2006 through December 2007 at Denver Health, a university‐affiliated public hospital. Inclusion criteria were patients between 18 and 89 years of age who underwent CTPE imaging 2 or more days after being hospitalized, and had been receiving fractionated or unfractionated heparin in doses appropriate for preventing the development of deep venous thrombosis from the time of admission. Patients were excluded if they had signs or symptoms that were consistent with a diagnosis of PE at the time of admission, if they had a contraindication to prophylactic anticoagulation or if their prophylactic heparin therapy had been interrupted for any reason from the time prior to when the CTPE was ordered.

Patients were grouped depending on the service or location of their admission (ie, Medicine, Surgery, Orthopedics, Medical or Surgical Intensive Care Units). The objective elements of the Wells score were obtained by reviewing each patient's history and physical examination, progress notes and discharge summary. Patients were considered to have an alternate diagnosis of equal or greater likelihood than a PE if a d‐dimer was ordered, or if such a possibility was suggested by the treating clinician in the computerized order for the CTPE. The modified Wells score was used to classify patients into PE‐likely (total score 4) or PE‐unlikely (total score <4).7 Fisher's exact test was used to analyze the 2 2 table. P< 0.05 was taken to represent significance.

The Colorado Multiple Institutional Review Board approved this study with a waiver of informed consent.

Results

Of 446 patients who had CTPEs during the study period 286 (64%) met the inclusion criteria (Figure 1). Those who were excluded included 131 who did not receive continuous prophylactic anticoagulation from the time they were admitted to the time of the CT, 18 who had preexisting signs or symptoms and signs consistent with a diagnosis of PE at the time of admission, and 11 who were receiving therapeutic anticoagulation. The patients were hospitalized on different units and on a number of different services (Table 2).

Figure 1

Low molecular weight heparin was given to 165 patients (dalteparin, 5000 units, once daily), unfractionated heparin to 120 patients (104 receiving 5000 units twice daily and 16 receiving 5000 units 3 times a day) and 1 patient was given a Factor Xa inhibitor (fondaparinux 2.5 mg once daily) due to a history of heparin induced thrombocytopenia.

Hypoxia and tachycardia were the most common reasons for requesting a CTPE in instances in which an indication for CT imaging was documented. In almost 28% of patients, however, the reason for suspecting PE was not apparent on chart review (Table 3).

The prevalence of PE was 20/286 (7.0%, 95% CI (confidence interval): 4.0‐10.0). On the basis of the Wells score 212 patients (74%) were classified as PE‐likely and 74 (26%) as PE‐unlikely. Immobility or recent surgery, tachycardia and the absence of a more plausible diagnosis were the most common contributors to the final score (Table 4).

Nineteen of the 20 patients (95%) who had PE diagnosed on the basis of a positive CTPE were risk‐stratified on the basis of the Wells score into the PE‐likely category, and 1 (5%) was classified as PE‐unlikely. Of the 266 patients whose CTPEs were negative 193 (73%) were classified as PE‐likely and 73 (27%) as PE‐unlikely (P < 0.03). Accordingly, the modified Wells score was 95% sensitive for having a diagnosis of PE confirmed on CTPE, the specificity was only 27%, the positive predictive value was only 9% and the negative predictive value was 99%(Table 5) with negative likelihood ratio of 0.19.

A d‐dimer was ordered for 70 of the 74 patients (95%) who were classified as PE‐unlikely. In 67 of these (96%) the test was positive, and in all but 1 the result was falsely positive. D‐dimer testing was also obtained in 8 of 212 (4%) of patients classified as PE‐likely and was positive in all 8.

Discussion

This retrospective cohort study demonstrated that in hospitalized patients who were receiving prophylactic doses of fractionated or unfractionated heparin and underwent CTPE studies for the clinical suspicion of PE, the prevalence of PE was very low, the modified Wells rule classified 26% of the patients as PE‐unlikely, and the PE‐unlikely category was associated with an extremely high negative predictive value and low negative likelihood ratio for PE. We also confirmed that the prevalence of a positive d‐dimer was so high in this population that the test did not add to the ability to risk‐stratify patients for the likelihood of having a PE. These findings lead to the conclusion that CTPE studies were performed excessively in this cohort of patients.

Previous studies validating the Wells score enrolled combinations of inpatients and outpatients813 or outpatients exclusively.14, 15 To our knowledge the present study is the first to validate the utility of the scoring system in inpatients receiving prophylactic anticoagulation. As would be expected, the prevalence of PE in our population was lower than the 9% to 30% that has previously been reported in patients not receiving prophylactic anticoagulation,815 consistent with the 68% to 76% reduction in the risk of deep venous thrombosis that occurs with use of low‐dose heparin or low molecular weight heparin.16

Similar to the findings of Arnason et al.17 a large proportion of this inpatient cohort was classified as PE‐likely on the basis of only 3 of the 7 variablestachycardia, immobility or previous surgery, and the absence of a more likely competing diagnosis.

The d‐dimer was elevated above the upper limit of normal in nearly all the cases in which it was tested (96%). Bounameaux et al.18 first suggested that conditions other than VTE could increase the plasma d‐dimer level. D‐dimer levels above the cutoff that excludes thrombosis have been documented in absence of thrombosis in the elderly and in patients with numerous other conditions including infections, cancer, coronary, cerebral and peripheral arterial vascular disease, heart failure, rheumatologic diseases, surgery, trauma burns, and pregnancy.1821 Van Beek et al.22 and Miron et al.23 demonstrated that d‐dimer testing was not useful in hospitalized patients. Kabrhel et al.24 reported similar results in an Emergency Department cohort and concluded that d‐dimer testing increased the percent of patients who were investigated for PE and the percent that were sent for pulmonary vascular imaging without increasing the percent of patients diagnosed as having a PE. In our cohort, 74 patients (26%) were classified as PE‐unlikely, and we theorize that 67 (90%) of these underwent CTPE studies solely on the basis of having a positive d‐dimer. All but one of the CTPEs in the patients with positive d‐dimers were negative for PE confirming the that the low specificity of d‐dimer testing in hospitalized patients also applies to those receiving prophylactic anticoagulation.

The Wells rule was associated with a high negative predictive value (99%) and a corresponding low negative likelihood ratio of 0.19, with both of these parameters likely being strongly influenced by the low prevalence of PE in this cohort.

In most longitudinal controlled studies of heparin‐based prophylaxis the incidence of VTE in all medical and most surgical patients approximates 5%.25^,26 If this were taken to represent the pre‐test probability of VTE in patients on prophylaxis in whom the question of PE arises, then according to Bayesian theory, a PE‐unlikely classification with a negative likelihood ratio of 0.19 would result in a post‐test probability of less than 1%. This is well below the threshold at which diagnostic imaging delivers no benefit and in fact, may cause harm. Accordingly, PE can be safely excluded in those who are risk‐stratified to PE‐unlikely, with or without an accompanying negative d‐dimer. The average charge for a CTPE at our institution is $1800 and the 2009 cost/charge ratio was 54%. Accordingly, the cost savings to our hospital if CTPEs were not done on the 74 patients classified as PE‐unlikely would exceed $66,000/year.

Our study has a number of potential limitations. Because the data came from a single university‐affiliated public hospital the results might not generalize to other hospitals (teaching or nonteaching). Despite finding a very low prevalence of PE in patients receiving prophylactic heparin, the true prevalence of PE might have been overestimated since our sample size was small and Denver Health is a regional level I trauma center and has a busy joint arthroplasty service, i.e., services known to have an increased prevalence of venous thrombosis.16 If the prevalence of PE were indeed lower than what we observed, however, it would decrease the number of true positive and false negative CTPEs which would, in turn, further strengthen the conclusion that CTPEs are being overused in hospitalized patients receiving prophylactic heparin who are risk‐stratified to a PE‐unlikely category. Similarly, because our sample size was small we may have underestimated the prevalence of PE. Our narrow CIs, and the fact that the prevalence we observed is consistent with the effect of prophylaxic heparin on the incidence of VTE suggest that, if an error were made, it would not be large enough to alter our conclusions.

Our analysis did not include patients in whom PE was excluded without performing CTPE testing. If these patients had CTPEs the large majority would be negative because of a very low pretest probability and risk‐stratification would have placed them in a PE‐unlikely category (ie, true negatives), thereby also increasing the negative predictive value of the Wells score used in this setting.

We calculated the Wells score retrospectively as was previously done in studies by Chagnon et al.,11 Righini et al.,14 and Ranji et al.27 (although the methods used in these studies were not described in detail). We assumed that whenever a d‐dimer test was ordered the treating physician thought that PE was less likely than an alternate diagnosis reasoning that, if they thought PE were the most likely diagnosis, d‐dimers should not have been obtained as, in this circumstance, they are not recommended as part of the diagnostic algorithm.8 Conversely, we assumed that for patients who did not get d‐dimer testing, the treating physician thought that PE was the most likely diagnosis. Alternatively, the physicians might not have ordered a d‐dimer because they recognized that the test is of limited clinical utility in hospitalized patients. In this latter circumstance, the number of PE‐likely patients would be overestimated and the number of PE‐unlikely would be underestimated, reducing the strength of our conclusions or potentially invalidating them. Since the accuracy of prediction rules mirrors that of implicit clinical judgment, however, we suggest that, for most of the patients who had CTPEs performed without d‐dimers, the ordering physician had a high suspicion of PE28, 29 and that the large majority of PE‐likely patients were correctly classified.

In summary, we found that CTPE testing is frequently performed in hospitalized patients receiving prophylactic heparin despite there being a very low prevalence of PE in this cohort, and that risk‐stratifying patients into the PE‐unlikely category using the modified Wells score accurately excludes the diagnosis of PE. The problem of overuse of CTPEs is compounded by the well‐recognized misuse of d‐dimer testing in hospitalized patients. On the basis of our findings we recommend that, when hospitalized patients who are receiving heparin prophylaxis to prevent VTE develop signs or symptoms suggestive of PE they should be risk‐stratified using the modified Wells criteria. In those classified as PE‐unlikely PE can be safely excluded without further testing. Using this approach 26% of CTPEs done on the cohort of hospitalized patients we studied, and all d‐dimers could have been avoided. If the results of our study are duplicated in other centers these recommendations should be included in future guidelines summarizing the most cost‐effective ways to evaluate patients for possible PE.

Symptoms, signs, chest radiograms, electrocardiograms and laboratory data have a low specificity for the diagnosis of pulmonary embolism (PE) when used in isolation, but when used in combination they can accurately identify patients with an increased likelihood of having a PE.17 The Wells score combines multiple variables into a prediction tool (Table 1). The original model identified three categories of patients with increasing likelihoods of having a PE,6 but a simpler, dichotomous version was subsequently proposed.7 A sequential diagnostic strategy combining the dichotomous Wells rule with a serum d‐dimer test has been validated against contrast‐enhanced spiral computed tomography (CTPE) on cohorts comprised largely of ambulatory outpatient and emergency room patients.815 This method, however, has never been tested in hospitalized patients who were receiving heparin in doses designed to prevent the development of venous thromboembolism (VTE). The purpose of this study was to evaluate the utility of the modified Wells score to predict the presence or absence of PE in hospitalized patients who were receiving prophylactic heparin.

Methods

We screened consecutive patients who underwent CTPE studies from January 2006 through December 2007 at Denver Health, a university‐affiliated public hospital. Inclusion criteria were patients between 18 and 89 years of age who underwent CTPE imaging 2 or more days after being hospitalized, and had been receiving fractionated or unfractionated heparin in doses appropriate for preventing the development of deep venous thrombosis from the time of admission. Patients were excluded if they had signs or symptoms that were consistent with a diagnosis of PE at the time of admission, if they had a contraindication to prophylactic anticoagulation or if their prophylactic heparin therapy had been interrupted for any reason from the time prior to when the CTPE was ordered.

Patients were grouped depending on the service or location of their admission (ie, Medicine, Surgery, Orthopedics, Medical or Surgical Intensive Care Units). The objective elements of the Wells score were obtained by reviewing each patient's history and physical examination, progress notes and discharge summary. Patients were considered to have an alternate diagnosis of equal or greater likelihood than a PE if a d‐dimer was ordered, or if such a possibility was suggested by the treating clinician in the computerized order for the CTPE. The modified Wells score was used to classify patients into PE‐likely (total score 4) or PE‐unlikely (total score <4).7 Fisher's exact test was used to analyze the 2 2 table. P< 0.05 was taken to represent significance.

The Colorado Multiple Institutional Review Board approved this study with a waiver of informed consent.

Results

Of 446 patients who had CTPEs during the study period 286 (64%) met the inclusion criteria (Figure 1). Those who were excluded included 131 who did not receive continuous prophylactic anticoagulation from the time they were admitted to the time of the CT, 18 who had preexisting signs or symptoms and signs consistent with a diagnosis of PE at the time of admission, and 11 who were receiving therapeutic anticoagulation. The patients were hospitalized on different units and on a number of different services (Table 2).

Figure 1

Low molecular weight heparin was given to 165 patients (dalteparin, 5000 units, once daily), unfractionated heparin to 120 patients (104 receiving 5000 units twice daily and 16 receiving 5000 units 3 times a day) and 1 patient was given a Factor Xa inhibitor (fondaparinux 2.5 mg once daily) due to a history of heparin induced thrombocytopenia.

Hypoxia and tachycardia were the most common reasons for requesting a CTPE in instances in which an indication for CT imaging was documented. In almost 28% of patients, however, the reason for suspecting PE was not apparent on chart review (Table 3).

The prevalence of PE was 20/286 (7.0%, 95% CI (confidence interval): 4.0‐10.0). On the basis of the Wells score 212 patients (74%) were classified as PE‐likely and 74 (26%) as PE‐unlikely. Immobility or recent surgery, tachycardia and the absence of a more plausible diagnosis were the most common contributors to the final score (Table 4).

Nineteen of the 20 patients (95%) who had PE diagnosed on the basis of a positive CTPE were risk‐stratified on the basis of the Wells score into the PE‐likely category, and 1 (5%) was classified as PE‐unlikely. Of the 266 patients whose CTPEs were negative 193 (73%) were classified as PE‐likely and 73 (27%) as PE‐unlikely (P < 0.03). Accordingly, the modified Wells score was 95% sensitive for having a diagnosis of PE confirmed on CTPE, the specificity was only 27%, the positive predictive value was only 9% and the negative predictive value was 99%(Table 5) with negative likelihood ratio of 0.19.

A d‐dimer was ordered for 70 of the 74 patients (95%) who were classified as PE‐unlikely. In 67 of these (96%) the test was positive, and in all but 1 the result was falsely positive. D‐dimer testing was also obtained in 8 of 212 (4%) of patients classified as PE‐likely and was positive in all 8.

Discussion

This retrospective cohort study demonstrated that in hospitalized patients who were receiving prophylactic doses of fractionated or unfractionated heparin and underwent CTPE studies for the clinical suspicion of PE, the prevalence of PE was very low, the modified Wells rule classified 26% of the patients as PE‐unlikely, and the PE‐unlikely category was associated with an extremely high negative predictive value and low negative likelihood ratio for PE. We also confirmed that the prevalence of a positive d‐dimer was so high in this population that the test did not add to the ability to risk‐stratify patients for the likelihood of having a PE. These findings lead to the conclusion that CTPE studies were performed excessively in this cohort of patients.

Previous studies validating the Wells score enrolled combinations of inpatients and outpatients813 or outpatients exclusively.14, 15 To our knowledge the present study is the first to validate the utility of the scoring system in inpatients receiving prophylactic anticoagulation. As would be expected, the prevalence of PE in our population was lower than the 9% to 30% that has previously been reported in patients not receiving prophylactic anticoagulation,815 consistent with the 68% to 76% reduction in the risk of deep venous thrombosis that occurs with use of low‐dose heparin or low molecular weight heparin.16

Similar to the findings of Arnason et al.17 a large proportion of this inpatient cohort was classified as PE‐likely on the basis of only 3 of the 7 variablestachycardia, immobility or previous surgery, and the absence of a more likely competing diagnosis.

The d‐dimer was elevated above the upper limit of normal in nearly all the cases in which it was tested (96%). Bounameaux et al.18 first suggested that conditions other than VTE could increase the plasma d‐dimer level. D‐dimer levels above the cutoff that excludes thrombosis have been documented in absence of thrombosis in the elderly and in patients with numerous other conditions including infections, cancer, coronary, cerebral and peripheral arterial vascular disease, heart failure, rheumatologic diseases, surgery, trauma burns, and pregnancy.1821 Van Beek et al.22 and Miron et al.23 demonstrated that d‐dimer testing was not useful in hospitalized patients. Kabrhel et al.24 reported similar results in an Emergency Department cohort and concluded that d‐dimer testing increased the percent of patients who were investigated for PE and the percent that were sent for pulmonary vascular imaging without increasing the percent of patients diagnosed as having a PE. In our cohort, 74 patients (26%) were classified as PE‐unlikely, and we theorize that 67 (90%) of these underwent CTPE studies solely on the basis of having a positive d‐dimer. All but one of the CTPEs in the patients with positive d‐dimers were negative for PE confirming the that the low specificity of d‐dimer testing in hospitalized patients also applies to those receiving prophylactic anticoagulation.

The Wells rule was associated with a high negative predictive value (99%) and a corresponding low negative likelihood ratio of 0.19, with both of these parameters likely being strongly influenced by the low prevalence of PE in this cohort.

In most longitudinal controlled studies of heparin‐based prophylaxis the incidence of VTE in all medical and most surgical patients approximates 5%.25^,26 If this were taken to represent the pre‐test probability of VTE in patients on prophylaxis in whom the question of PE arises, then according to Bayesian theory, a PE‐unlikely classification with a negative likelihood ratio of 0.19 would result in a post‐test probability of less than 1%. This is well below the threshold at which diagnostic imaging delivers no benefit and in fact, may cause harm. Accordingly, PE can be safely excluded in those who are risk‐stratified to PE‐unlikely, with or without an accompanying negative d‐dimer. The average charge for a CTPE at our institution is $1800 and the 2009 cost/charge ratio was 54%. Accordingly, the cost savings to our hospital if CTPEs were not done on the 74 patients classified as PE‐unlikely would exceed $66,000/year.

Our study has a number of potential limitations. Because the data came from a single university‐affiliated public hospital the results might not generalize to other hospitals (teaching or nonteaching). Despite finding a very low prevalence of PE in patients receiving prophylactic heparin, the true prevalence of PE might have been overestimated since our sample size was small and Denver Health is a regional level I trauma center and has a busy joint arthroplasty service, i.e., services known to have an increased prevalence of venous thrombosis.16 If the prevalence of PE were indeed lower than what we observed, however, it would decrease the number of true positive and false negative CTPEs which would, in turn, further strengthen the conclusion that CTPEs are being overused in hospitalized patients receiving prophylactic heparin who are risk‐stratified to a PE‐unlikely category. Similarly, because our sample size was small we may have underestimated the prevalence of PE. Our narrow CIs, and the fact that the prevalence we observed is consistent with the effect of prophylaxic heparin on the incidence of VTE suggest that, if an error were made, it would not be large enough to alter our conclusions.

Our analysis did not include patients in whom PE was excluded without performing CTPE testing. If these patients had CTPEs the large majority would be negative because of a very low pretest probability and risk‐stratification would have placed them in a PE‐unlikely category (ie, true negatives), thereby also increasing the negative predictive value of the Wells score used in this setting.

We calculated the Wells score retrospectively as was previously done in studies by Chagnon et al.,11 Righini et al.,14 and Ranji et al.27 (although the methods used in these studies were not described in detail). We assumed that whenever a d‐dimer test was ordered the treating physician thought that PE was less likely than an alternate diagnosis reasoning that, if they thought PE were the most likely diagnosis, d‐dimers should not have been obtained as, in this circumstance, they are not recommended as part of the diagnostic algorithm.8 Conversely, we assumed that for patients who did not get d‐dimer testing, the treating physician thought that PE was the most likely diagnosis. Alternatively, the physicians might not have ordered a d‐dimer because they recognized that the test is of limited clinical utility in hospitalized patients. In this latter circumstance, the number of PE‐likely patients would be overestimated and the number of PE‐unlikely would be underestimated, reducing the strength of our conclusions or potentially invalidating them. Since the accuracy of prediction rules mirrors that of implicit clinical judgment, however, we suggest that, for most of the patients who had CTPEs performed without d‐dimers, the ordering physician had a high suspicion of PE28, 29 and that the large majority of PE‐likely patients were correctly classified.

In summary, we found that CTPE testing is frequently performed in hospitalized patients receiving prophylactic heparin despite there being a very low prevalence of PE in this cohort, and that risk‐stratifying patients into the PE‐unlikely category using the modified Wells score accurately excludes the diagnosis of PE. The problem of overuse of CTPEs is compounded by the well‐recognized misuse of d‐dimer testing in hospitalized patients. On the basis of our findings we recommend that, when hospitalized patients who are receiving heparin prophylaxis to prevent VTE develop signs or symptoms suggestive of PE they should be risk‐stratified using the modified Wells criteria. In those classified as PE‐unlikely PE can be safely excluded without further testing. Using this approach 26% of CTPEs done on the cohort of hospitalized patients we studied, and all d‐dimers could have been avoided. If the results of our study are duplicated in other centers these recommendations should be included in future guidelines summarizing the most cost‐effective ways to evaluate patients for possible PE.

Symptoms, signs, chest radiograms, electrocardiograms and laboratory data have a low specificity for the diagnosis of pulmonary embolism (PE) when used in isolation, but when used in combination they can accurately identify patients with an increased likelihood of having a PE.17 The Wells score combines multiple variables into a prediction tool (Table 1). The original model identified three categories of patients with increasing likelihoods of having a PE,6 but a simpler, dichotomous version was subsequently proposed.7 A sequential diagnostic strategy combining the dichotomous Wells rule with a serum d‐dimer test has been validated against contrast‐enhanced spiral computed tomography (CTPE) on cohorts comprised largely of ambulatory outpatient and emergency room patients.815 This method, however, has never been tested in hospitalized patients who were receiving heparin in doses designed to prevent the development of venous thromboembolism (VTE). The purpose of this study was to evaluate the utility of the modified Wells score to predict the presence or absence of PE in hospitalized patients who were receiving prophylactic heparin.

Methods

We screened consecutive patients who underwent CTPE studies from January 2006 through December 2007 at Denver Health, a university‐affiliated public hospital. Inclusion criteria were patients between 18 and 89 years of age who underwent CTPE imaging 2 or more days after being hospitalized, and had been receiving fractionated or unfractionated heparin in doses appropriate for preventing the development of deep venous thrombosis from the time of admission. Patients were excluded if they had signs or symptoms that were consistent with a diagnosis of PE at the time of admission, if they had a contraindication to prophylactic anticoagulation or if their prophylactic heparin therapy had been interrupted for any reason from the time prior to when the CTPE was ordered.

Patients were grouped depending on the service or location of their admission (ie, Medicine, Surgery, Orthopedics, Medical or Surgical Intensive Care Units). The objective elements of the Wells score were obtained by reviewing each patient's history and physical examination, progress notes and discharge summary. Patients were considered to have an alternate diagnosis of equal or greater likelihood than a PE if a d‐dimer was ordered, or if such a possibility was suggested by the treating clinician in the computerized order for the CTPE. The modified Wells score was used to classify patients into PE‐likely (total score 4) or PE‐unlikely (total score <4).7 Fisher's exact test was used to analyze the 2 2 table. P< 0.05 was taken to represent significance.

The Colorado Multiple Institutional Review Board approved this study with a waiver of informed consent.

Results

Of 446 patients who had CTPEs during the study period 286 (64%) met the inclusion criteria (Figure 1). Those who were excluded included 131 who did not receive continuous prophylactic anticoagulation from the time they were admitted to the time of the CT, 18 who had preexisting signs or symptoms and signs consistent with a diagnosis of PE at the time of admission, and 11 who were receiving therapeutic anticoagulation. The patients were hospitalized on different units and on a number of different services (Table 2).

Figure 1

Low molecular weight heparin was given to 165 patients (dalteparin, 5000 units, once daily), unfractionated heparin to 120 patients (104 receiving 5000 units twice daily and 16 receiving 5000 units 3 times a day) and 1 patient was given a Factor Xa inhibitor (fondaparinux 2.5 mg once daily) due to a history of heparin induced thrombocytopenia.

Hypoxia and tachycardia were the most common reasons for requesting a CTPE in instances in which an indication for CT imaging was documented. In almost 28% of patients, however, the reason for suspecting PE was not apparent on chart review (Table 3).

The prevalence of PE was 20/286 (7.0%, 95% CI (confidence interval): 4.0‐10.0). On the basis of the Wells score 212 patients (74%) were classified as PE‐likely and 74 (26%) as PE‐unlikely. Immobility or recent surgery, tachycardia and the absence of a more plausible diagnosis were the most common contributors to the final score (Table 4).

Nineteen of the 20 patients (95%) who had PE diagnosed on the basis of a positive CTPE were risk‐stratified on the basis of the Wells score into the PE‐likely category, and 1 (5%) was classified as PE‐unlikely. Of the 266 patients whose CTPEs were negative 193 (73%) were classified as PE‐likely and 73 (27%) as PE‐unlikely (P < 0.03). Accordingly, the modified Wells score was 95% sensitive for having a diagnosis of PE confirmed on CTPE, the specificity was only 27%, the positive predictive value was only 9% and the negative predictive value was 99%(Table 5) with negative likelihood ratio of 0.19.

A d‐dimer was ordered for 70 of the 74 patients (95%) who were classified as PE‐unlikely. In 67 of these (96%) the test was positive, and in all but 1 the result was falsely positive. D‐dimer testing was also obtained in 8 of 212 (4%) of patients classified as PE‐likely and was positive in all 8.

Discussion

This retrospective cohort study demonstrated that in hospitalized patients who were receiving prophylactic doses of fractionated or unfractionated heparin and underwent CTPE studies for the clinical suspicion of PE, the prevalence of PE was very low, the modified Wells rule classified 26% of the patients as PE‐unlikely, and the PE‐unlikely category was associated with an extremely high negative predictive value and low negative likelihood ratio for PE. We also confirmed that the prevalence of a positive d‐dimer was so high in this population that the test did not add to the ability to risk‐stratify patients for the likelihood of having a PE. These findings lead to the conclusion that CTPE studies were performed excessively in this cohort of patients.

Previous studies validating the Wells score enrolled combinations of inpatients and outpatients813 or outpatients exclusively.14, 15 To our knowledge the present study is the first to validate the utility of the scoring system in inpatients receiving prophylactic anticoagulation. As would be expected, the prevalence of PE in our population was lower than the 9% to 30% that has previously been reported in patients not receiving prophylactic anticoagulation,815 consistent with the 68% to 76% reduction in the risk of deep venous thrombosis that occurs with use of low‐dose heparin or low molecular weight heparin.16

Similar to the findings of Arnason et al.17 a large proportion of this inpatient cohort was classified as PE‐likely on the basis of only 3 of the 7 variablestachycardia, immobility or previous surgery, and the absence of a more likely competing diagnosis.

The d‐dimer was elevated above the upper limit of normal in nearly all the cases in which it was tested (96%). Bounameaux et al.18 first suggested that conditions other than VTE could increase the plasma d‐dimer level. D‐dimer levels above the cutoff that excludes thrombosis have been documented in absence of thrombosis in the elderly and in patients with numerous other conditions including infections, cancer, coronary, cerebral and peripheral arterial vascular disease, heart failure, rheumatologic diseases, surgery, trauma burns, and pregnancy.1821 Van Beek et al.22 and Miron et al.23 demonstrated that d‐dimer testing was not useful in hospitalized patients. Kabrhel et al.24 reported similar results in an Emergency Department cohort and concluded that d‐dimer testing increased the percent of patients who were investigated for PE and the percent that were sent for pulmonary vascular imaging without increasing the percent of patients diagnosed as having a PE. In our cohort, 74 patients (26%) were classified as PE‐unlikely, and we theorize that 67 (90%) of these underwent CTPE studies solely on the basis of having a positive d‐dimer. All but one of the CTPEs in the patients with positive d‐dimers were negative for PE confirming the that the low specificity of d‐dimer testing in hospitalized patients also applies to those receiving prophylactic anticoagulation.

The Wells rule was associated with a high negative predictive value (99%) and a corresponding low negative likelihood ratio of 0.19, with both of these parameters likely being strongly influenced by the low prevalence of PE in this cohort.

In most longitudinal controlled studies of heparin‐based prophylaxis the incidence of VTE in all medical and most surgical patients approximates 5%.25^,26 If this were taken to represent the pre‐test probability of VTE in patients on prophylaxis in whom the question of PE arises, then according to Bayesian theory, a PE‐unlikely classification with a negative likelihood ratio of 0.19 would result in a post‐test probability of less than 1%. This is well below the threshold at which diagnostic imaging delivers no benefit and in fact, may cause harm. Accordingly, PE can be safely excluded in those who are risk‐stratified to PE‐unlikely, with or without an accompanying negative d‐dimer. The average charge for a CTPE at our institution is $1800 and the 2009 cost/charge ratio was 54%. Accordingly, the cost savings to our hospital if CTPEs were not done on the 74 patients classified as PE‐unlikely would exceed $66,000/year.

Our study has a number of potential limitations. Because the data came from a single university‐affiliated public hospital the results might not generalize to other hospitals (teaching or nonteaching). Despite finding a very low prevalence of PE in patients receiving prophylactic heparin, the true prevalence of PE might have been overestimated since our sample size was small and Denver Health is a regional level I trauma center and has a busy joint arthroplasty service, i.e., services known to have an increased prevalence of venous thrombosis.16 If the prevalence of PE were indeed lower than what we observed, however, it would decrease the number of true positive and false negative CTPEs which would, in turn, further strengthen the conclusion that CTPEs are being overused in hospitalized patients receiving prophylactic heparin who are risk‐stratified to a PE‐unlikely category. Similarly, because our sample size was small we may have underestimated the prevalence of PE. Our narrow CIs, and the fact that the prevalence we observed is consistent with the effect of prophylaxic heparin on the incidence of VTE suggest that, if an error were made, it would not be large enough to alter our conclusions.

Our analysis did not include patients in whom PE was excluded without performing CTPE testing. If these patients had CTPEs the large majority would be negative because of a very low pretest probability and risk‐stratification would have placed them in a PE‐unlikely category (ie, true negatives), thereby also increasing the negative predictive value of the Wells score used in this setting.

We calculated the Wells score retrospectively as was previously done in studies by Chagnon et al.,11 Righini et al.,14 and Ranji et al.27 (although the methods used in these studies were not described in detail). We assumed that whenever a d‐dimer test was ordered the treating physician thought that PE was less likely than an alternate diagnosis reasoning that, if they thought PE were the most likely diagnosis, d‐dimers should not have been obtained as, in this circumstance, they are not recommended as part of the diagnostic algorithm.8 Conversely, we assumed that for patients who did not get d‐dimer testing, the treating physician thought that PE was the most likely diagnosis. Alternatively, the physicians might not have ordered a d‐dimer because they recognized that the test is of limited clinical utility in hospitalized patients. In this latter circumstance, the number of PE‐likely patients would be overestimated and the number of PE‐unlikely would be underestimated, reducing the strength of our conclusions or potentially invalidating them. Since the accuracy of prediction rules mirrors that of implicit clinical judgment, however, we suggest that, for most of the patients who had CTPEs performed without d‐dimers, the ordering physician had a high suspicion of PE28, 29 and that the large majority of PE‐likely patients were correctly classified.

In summary, we found that CTPE testing is frequently performed in hospitalized patients receiving prophylactic heparin despite there being a very low prevalence of PE in this cohort, and that risk‐stratifying patients into the PE‐unlikely category using the modified Wells score accurately excludes the diagnosis of PE. The problem of overuse of CTPEs is compounded by the well‐recognized misuse of d‐dimer testing in hospitalized patients. On the basis of our findings we recommend that, when hospitalized patients who are receiving heparin prophylaxis to prevent VTE develop signs or symptoms suggestive of PE they should be risk‐stratified using the modified Wells criteria. In those classified as PE‐unlikely PE can be safely excluded without further testing. Using this approach 26% of CTPEs done on the cohort of hospitalized patients we studied, and all d‐dimers could have been avoided. If the results of our study are duplicated in other centers these recommendations should be included in future guidelines summarizing the most cost‐effective ways to evaluate patients for possible PE.

In order to improve resident supervision and timeliness of invasive bedside procedures such as paracentesis, thoracentesis, and lumbar puncture, some academic medical centers have implemented procedure services that focus on providing high‐quality procedural care.1, 2

Procedure services have the potential to affect patient satisfaction, a key indicator in quality of care measurment.3 Having senior physicians present increases patient comfort during outpatient case presentations4 and improves patient satisfaction with explanations of tests and medications.5 However, we had concerns that teaching during a procedure may heighten patient anxiety. Patients are reluctant to be the first patient of a resident or medical student for a procedure,68 and patients are more likely to refuse consent to have a resident perform complex procedures.8 In previous studies, patient satisfaction with gynecological exams and flexible sigmoidoscopy performed by residents was comparable to satisfaction with those performed by staff physicians,9, 10 though in the case of flexible sigmoidoscopy, procedure duration was slightly longer.10 Few, if any, data describe bedside teaching or patient impressions of physician communication during procedures.

We carried out a prospective study of patient perceptions of the University of California San Francisco (UCSF) Hospitalist Procedure Service (HPS). Our study had the primary goal of understanding how our modelwhich involves bedside procedural teaching and feedback in real time (eg, as the procedure is performed)is perceived by patients.

Patients and Methods

Results

Discussion

Even though novice interns performed procedures and simultaneous bedside teaching, patient satisfaction with a teaching procedure service was high, and reported complication rates were low. In addition, a majority of patients found discussions related to teaching activities reassuring and potentially important to their perception of care quality. Analogous studies examining patient satisfaction with endoscopic care found similar rates of patient satisfaction with endoscopists' bedside manner, technical skills, and pain control, but these studies included sedated patients.21 Our results are unique, as we evaluated awake patients with attention to perception of bedside teaching with novice interns.

Our findings offer an alternative strategy for bedside procedural teaching that employs transparency in the use of an expert and a trainee to introduce patients to bedside teaching by experts, which is not common at many academic medical centers.28 Patients may have been reassured by a clear explanation of the role of the service and the providers involved as well as an assurance of expertise and attention to patient comfort and safety. In addition to patient satisfaction, this model has the potential to impact both the safety of bedside procedures and housestaff education around procedure performance. For example, pneumothorax rates using our procedure service model are lower than those published (0% vs. 4% for ultrasound‐guided thoracentesis and 8.5% for thoracentesis by less experienced clinicians).29

Providers may be reluctant to teach at the bedside of awake patients for fear of heightening patient anxiety over trainee inexperience. In the 1960s similar fears were raised over the concern for patient anxiety with bedside rounding,30 but later studies revealed these concerns to be largely unfounded. Instead, bedside rounds have been shown to positively influence patients' feelings about their hospital experience and their relationships with their physicians compared with patients whose case presentations were made in a conference room.31, 32 Given the opportunity to comment on areas for improvement, patients in our study specifically elaborated regarding pain control, communication, and efficiency problems. Although 16% of patients did not find the communication of physicians reassuring, none of the negative comments reflected problems with bedside teaching, but rather concepts such as desiring a better explanation of steps throughout the procedure. Specifically, patients desire better communication for unanticipated pain.

There are several limitations to this study. Lack of patient satisfaction data from a control group of patients whose procedures were performed by attendings or housestaff alone limits our ability to draw conclusions about our satisfaction scores. The scarce applicable literature offers only imperfect comparison data. Because hospitalists were not blinded to the survey, attending behavior may have been subject to a Hawthorne effect.33 Consenting patients after the procedure could have provided hospitalists with an opportunity to exclude patients who appeared less satisfied with their procedure; however, attempts were made to prevent this behavior by requiring strict accounting of why a patient was not consented for the study. Use of alternative personnel for consent such as nurses was explored, but was found not to be feasible due to limited resources. These data are only applicable to English‐speaking patients who are literate and well enough to complete a survey. It is not clear whether the experience for other patients would reflect the same outcomes. It is plausible that non‐English‐speaking patients might have more concerns about incomprehensible conversations taking place during their procedure. Although the surveys were anonymous and patients were told that the proceduralists would not see individual responses, responses may have been biased out of patient concern that their response might affect their care. Hospitalists obtaining consent, however, were careful to stress anonymity and the distinction between the primary team and the procedure team.

Academic hospitals are struggling with providing quality procedural care while balancing housestaff education and experience.28 With hospitalists playing an increasingly prominent role in housestaff education and patient satisfaction initiatives, the supervision of housestaff by trained hospitalist faculty may help meet both aims in the performance of invasive bedside procedures, particularly at institutions where simulation training resources are limited. Although concern may exist for potential patient anxiety with bedside teaching, our data demonstrate high levels of patient satisfaction with a hospitalist procedure service despite novice procedure performers and an emphasis on teaching during the procedure.

In order to improve resident supervision and timeliness of invasive bedside procedures such as paracentesis, thoracentesis, and lumbar puncture, some academic medical centers have implemented procedure services that focus on providing high‐quality procedural care.1, 2

Procedure services have the potential to affect patient satisfaction, a key indicator in quality of care measurment.3 Having senior physicians present increases patient comfort during outpatient case presentations4 and improves patient satisfaction with explanations of tests and medications.5 However, we had concerns that teaching during a procedure may heighten patient anxiety. Patients are reluctant to be the first patient of a resident or medical student for a procedure,68 and patients are more likely to refuse consent to have a resident perform complex procedures.8 In previous studies, patient satisfaction with gynecological exams and flexible sigmoidoscopy performed by residents was comparable to satisfaction with those performed by staff physicians,9, 10 though in the case of flexible sigmoidoscopy, procedure duration was slightly longer.10 Few, if any, data describe bedside teaching or patient impressions of physician communication during procedures.

We carried out a prospective study of patient perceptions of the University of California San Francisco (UCSF) Hospitalist Procedure Service (HPS). Our study had the primary goal of understanding how our modelwhich involves bedside procedural teaching and feedback in real time (eg, as the procedure is performed)is perceived by patients.

Patients and Methods

Results

Discussion

Even though novice interns performed procedures and simultaneous bedside teaching, patient satisfaction with a teaching procedure service was high, and reported complication rates were low. In addition, a majority of patients found discussions related to teaching activities reassuring and potentially important to their perception of care quality. Analogous studies examining patient satisfaction with endoscopic care found similar rates of patient satisfaction with endoscopists' bedside manner, technical skills, and pain control, but these studies included sedated patients.21 Our results are unique, as we evaluated awake patients with attention to perception of bedside teaching with novice interns.

Our findings offer an alternative strategy for bedside procedural teaching that employs transparency in the use of an expert and a trainee to introduce patients to bedside teaching by experts, which is not common at many academic medical centers.28 Patients may have been reassured by a clear explanation of the role of the service and the providers involved as well as an assurance of expertise and attention to patient comfort and safety. In addition to patient satisfaction, this model has the potential to impact both the safety of bedside procedures and housestaff education around procedure performance. For example, pneumothorax rates using our procedure service model are lower than those published (0% vs. 4% for ultrasound‐guided thoracentesis and 8.5% for thoracentesis by less experienced clinicians).29

Providers may be reluctant to teach at the bedside of awake patients for fear of heightening patient anxiety over trainee inexperience. In the 1960s similar fears were raised over the concern for patient anxiety with bedside rounding,30 but later studies revealed these concerns to be largely unfounded. Instead, bedside rounds have been shown to positively influence patients' feelings about their hospital experience and their relationships with their physicians compared with patients whose case presentations were made in a conference room.31, 32 Given the opportunity to comment on areas for improvement, patients in our study specifically elaborated regarding pain control, communication, and efficiency problems. Although 16% of patients did not find the communication of physicians reassuring, none of the negative comments reflected problems with bedside teaching, but rather concepts such as desiring a better explanation of steps throughout the procedure. Specifically, patients desire better communication for unanticipated pain.

There are several limitations to this study. Lack of patient satisfaction data from a control group of patients whose procedures were performed by attendings or housestaff alone limits our ability to draw conclusions about our satisfaction scores. The scarce applicable literature offers only imperfect comparison data. Because hospitalists were not blinded to the survey, attending behavior may have been subject to a Hawthorne effect.33 Consenting patients after the procedure could have provided hospitalists with an opportunity to exclude patients who appeared less satisfied with their procedure; however, attempts were made to prevent this behavior by requiring strict accounting of why a patient was not consented for the study. Use of alternative personnel for consent such as nurses was explored, but was found not to be feasible due to limited resources. These data are only applicable to English‐speaking patients who are literate and well enough to complete a survey. It is not clear whether the experience for other patients would reflect the same outcomes. It is plausible that non‐English‐speaking patients might have more concerns about incomprehensible conversations taking place during their procedure. Although the surveys were anonymous and patients were told that the proceduralists would not see individual responses, responses may have been biased out of patient concern that their response might affect their care. Hospitalists obtaining consent, however, were careful to stress anonymity and the distinction between the primary team and the procedure team.

Academic hospitals are struggling with providing quality procedural care while balancing housestaff education and experience.28 With hospitalists playing an increasingly prominent role in housestaff education and patient satisfaction initiatives, the supervision of housestaff by trained hospitalist faculty may help meet both aims in the performance of invasive bedside procedures, particularly at institutions where simulation training resources are limited. Although concern may exist for potential patient anxiety with bedside teaching, our data demonstrate high levels of patient satisfaction with a hospitalist procedure service despite novice procedure performers and an emphasis on teaching during the procedure.

In order to improve resident supervision and timeliness of invasive bedside procedures such as paracentesis, thoracentesis, and lumbar puncture, some academic medical centers have implemented procedure services that focus on providing high‐quality procedural care.1, 2

Procedure services have the potential to affect patient satisfaction, a key indicator in quality of care measurment.3 Having senior physicians present increases patient comfort during outpatient case presentations4 and improves patient satisfaction with explanations of tests and medications.5 However, we had concerns that teaching during a procedure may heighten patient anxiety. Patients are reluctant to be the first patient of a resident or medical student for a procedure,68 and patients are more likely to refuse consent to have a resident perform complex procedures.8 In previous studies, patient satisfaction with gynecological exams and flexible sigmoidoscopy performed by residents was comparable to satisfaction with those performed by staff physicians,9, 10 though in the case of flexible sigmoidoscopy, procedure duration was slightly longer.10 Few, if any, data describe bedside teaching or patient impressions of physician communication during procedures.

We carried out a prospective study of patient perceptions of the University of California San Francisco (UCSF) Hospitalist Procedure Service (HPS). Our study had the primary goal of understanding how our modelwhich involves bedside procedural teaching and feedback in real time (eg, as the procedure is performed)is perceived by patients.

Patients and Methods