Accuracy of Hospitalist‐Performed HCUE

Hand‐carried ultrasound echocardiography (HCUE) can help noncardiologists answer well‐defined questions at patients' bedsides in less than 10 minutes.1, 2 Indeed, intensivists3 and emergency department physicians4 already use HCUE to make rapid, point‐of‐care assessments. Since cardiovascular diagnoses are common among general medicine inpatients, HCUE may become an important skill for hospitalists to learn.5

However, uncertainty exists about the duration of HCUE training for hospitalists. In 2002, experts from the American Society of Echocardiography (ASE) published recommendations on training requirements for HCUE.6 With limited data on the safety or performance of HCUE training programs, which had just begun to emerge, the ASE borrowed from the proven training recommendations for standard echocardiography (SE). They recommended that all HCUE trainees, cardiologist and noncardiologist alike, complete level 1 SE training: 75 personally‐performed and 150 personally‐interpreted echocardiographic examinations. Since then, however, several HCUE training programs designed for noncardiologists have emerged.2, 5, 710 These alternative programs suggest that the ASE's recommended duration of training may be too long, particularly for focused HCUE that is limited to a few relatively simple assessments. It is important not to overshoot the requirements of HCUE training, because doing so may discourage groups of noncardiologists, like hospitalists, who may derive great benefits from HCUE.11

To address this uncertainty for hospitalists, we first developed a brief HCUE training program to assess 6 important cardiac abnormalities. We then studied the diagnostic accuracy of HCUE by hospitalists as a test of these 6 cardiac abnormalities assessed by SE.

Patients and Methods

Setting and Subjects

This prospective cohort study was performed at Stroger Hospital of Cook County, a 500‐bed public teaching hospital in Chicago, IL, from March through May of 2007. The cohort was adult inpatients who were referred for SE on weekdays from 3 distinct patient care units (Figure 1). We used 2 sampling modes to balance practical constraints (short‐stay unit [SSU] patients were more localized and, therefore, easier to study) with clinical diversity. We consecutively sampled patients from our SSU, where adults with provisional cardiovascular diagnoses are admitted if they might be eligible for discharge with in 3 days.12 But we used random number tables with a daily unique starting point to randomly sample patients from the general medical wards and the coronary care unit (CCU). Patients were excluded if repositioning them for HCUE was potentially harmful. The study was approved by our hospital's institutional review board, and we obtained written informed consent from all enrolled patients.

Figure 1

Results

During the 3 month study period, 654 patients were referred for SE from the 3 participating patient care units (Figure 1). Among these, 65 patients were ineligible because their SE was performed on the weekend and 178 other patients were not randomized from the general medical wards and CCU. From the remaining eligible patients, 322 underwent HCUE and 314 (98% of 322) underwent both SE and HCUE. Individual SE assessments were not interpretable (and therefore excluded) due to poor image quality for LA enlargement in 1 patient and IVC dilatation in 30 patients. Eighty‐three percent of patients who underwent SE (260/314) were referred to assess LV function (Table 3). The prevalence of the 6 clinically pertinent cardiac abnormalities under study ranged from 1% for moderate or large pericardial effusion to 25% for LV systolic dysfunction. Overall, 40% of patients had at least 1 out of 6 cardiac abnormalities.

Each hospitalist performed a similar total number of HCUE examinations (range, 3447). The median time difference between performance of SE and HCUE was 2.8 hours (25th75th percentiles, 1.45.1). Despite the high prevalence of chronic obstructive pulmonary disease and obesity, hospitalists considered HCUE assessments indeterminate in only 2% to 6% of the 6 assessments made for each patient (Table 4). Among the 38 patients (12% of 322) with any indeterminate HCUE assessment, 24 patients had only 1 out of 6 possible. Hospitalists completed HCUE in a median time of 28 minutes (25th‐75th percentiles, 2035), which included the time to record 7 best‐quality moving images and to fill out the research data collection form.

When HCUE results were analyzed as dichotomous values, positive likelihood ratios ranged from 2.5 to 21, and negative likelihood ratios ranged from 0 to 0.4 (Table 5). Positive and negative likelihood ratios were both sufficiency high and low to respectively increase and decrease by 5‐fold the prior odds of 3 out of 6 cardiac abnormalities: LV systolic dysfunction, moderate or severe MR regurgitation, and moderate or large pericardial effusion. Considering HCUE results as ordinal values for ROC analysis yielded additional diagnostic information (Figure 2). For example, the likelihood ratio of 1.0 (95% CI, 0.42.0) for borderline positive moderate or severe LA enlargement increased to 29 (range, 1362) for extreme positive results. Areas under the ROC curves were 0.9 for 4 out of 6 cardiac abnormalities.

Figure 2

LV systolic dysfunction and IVC dilatation were both prevalent enough to meet our criterion to examine interobserver variability; the mean number of abnormal patients per hospitalist was 10 patients for LV systolic dysfunction and 6 patients for IVC dilatation. For LV systolic dysfunction, SDs around mean sensitivity (84%) and specificity (87%) were 12% and 6%, respectively. For IVC dilatation, SDs around mean sensitivity (58%) and specificity (86%) were 24% and 7%, respectively.

Discussion

We found that, after a 27‐hour training program, hospitalists performed HCUE with moderate to excellent diagnostic accuracy for 6 important cardiac abnormalities. For example, hospitalists' assessments of LV systolic function yielded positive and negative likelihood ratios of 6.9 (95% CI, 4.99.8) and 0.2 (95% CI, 0.10.3), respectively. At the bedsides of patients with acute heart failure, therefore, hospitalists could use HCUE to lower or raise the 50:50 chance of LV systolic dysfunction30 to 15% or 85%, respectively. Whether or not these posttest likelihoods are extreme enough to cross important thresholds will depend on the clinical context. Yet these findings demonstrate how HCUE has the potential to provide hospitalists with valuable point‐of‐care data that are otherwise unavailableeither because routine clinical assessments are unreliable31 or because echocardiographic services are not immediately accessible.1

In fact, recent data from the Joint Commission on Accreditation of Healthcare Organizations shows how inaccessible SE may be. Approximately one‐quarter of hospitals in the United States send home about 10% of patients with acute heart failure without echocardiographic assessment of LV systolic function before, during, or immediately after hospitalization.32 In doing so, these hospitals leave unmet the 2002 National Quality Improvement Goal of universal assessment of LV systolic function for all heart failure patients. Hospitalists could close this quality gap with routine, 10‐minute HCUE assessments in all patients admitted with acute heart failure. (Our research HCUE protocol required a median time of 28 minutes, but this included time to assess 5 other cardiac abnormalities and collect data for research purposes). Until the clinical consequences of introducing hospitalist‐performed HCUE are studied, potential benefits like this are tentative. But our findings suggest that training hospitalists to accurately perform HCUE can be successfully accomplished in just 27 hours.

Other studies of HCUE training programs for noncardiologists have also challenged the opinion that learning to perform HCUE requires more than 100 hours of training.2, 711 Yet only 1 prior study has examined an HCUE training program for hospitalists.5 In this study by Martin et al.,5 hospitalists completed 5 supervised HCUE examinations and 6 hours of interpretation training before investigators scored their image acquisition and interpretation skills from 30 unsupervised HCUE examinations. To estimate their final skill levels at the completion of all 35 examinations by accounting for an initially steep learning curve, investigators then adjusted these scores with regression models. Despite these upward adjustments, hospitalists' image acquisition and interpretation scores were low in comparison to echocardiographic technicians and cardiology fellows. Besides these adjusted measurements of hospitalists' skills, however, Martin et al.5 unfortunately did not also report standard measures of diagnostic accuracy, like those proposed by the Standards for Reporting of Diagnostic Accuracy (STARD) initiative.33 Therefore, direct comparisons to the present study are difficult. Nevertheless, their findings suggest that a training program limited to 5 supervised HCUE examinations may be inadequate for hospitalists. In fact, the same group's earlier study of medical trainees suggested a minimum of 30 supervised HCUE examinations.9 We chose to design our hospitalist training program based on this minimum, though they surprisingly did not.5 As others continue to refine the components of hospitalist HCUE training programs, such as the optimal number of supervised examinations, our program could serve as a reasonable comparative example: more rigorous than the program designed by Martin et al.5 but more feasible than ASE level 1 training.

The number and complexity of assessments taught in HCUE training programs will determine their duration. With ongoing advancements in HCUE technology, there is a growing list of potential assessments to choose from. Although HCUE training programs ought to include assessments with proven clinical applications, there are no trials of HCUE‐directed care to inform such decisions. In their absence, therefore, we chose 6 assessments based on the following 3 criteria. First, our assessments were otherwise not reliably available from routine clinical data, such as the physical examination. Second, our assessments were straightforward: easy to learn and simple to perform. Here, we based our reasoning on an expectation that the value of HCUE lies not in highly complex, state‐of‐the‐art assessmentswhich are best left to echocardiographers equipped with SEbut in simple, routine assessments made with highly portable machines that grant noncardiologists newfound access to point‐of‐care data.34 Third, our assessments were clinically pertinent and, where appropriate, defined by cut‐points at levels of severity that often lead to changes in management. We suspect that setting high cut‐points has the salutary effects of making assessments easier to learn and more accurate, because distinguishing mild abnormalities is likely the most challenging aspect of echocardiographic interpretation.35 Whether or not our choices of assessments, and their cut‐points, are optimal has yet to be determined by future research designed to study how they affect patient outcomes. Given our hospitalists' performance in the present study, these assessments seem worthy of such future research.

Our study had several limitations. We studied physicians and patients from only 1 hospital; similar studies performed in different settings, particularly among patients with different proportions and manifestations of disease, may find different results. Nevertheless, our sampling method of prospectively enrolling consecutive patients strengthens our findings. Some echocardiographic measurement methods used by our hospitalists differed in subtle ways from echocardiography guideline recommendations.35 We chose our methods (Table 2) for 2 reasons. First, whenever possible, we chose methods of interpretation that coincided with our local cardiologists'. Second, we chose simplicity over precision. For example, the biplane method of disks, or modified Simpson's rule, is the preferred volumetric method of calculating LA size.35 This method requires tracing the contours of the LA in 2 planes and then dividing the LA volume into stacked oval disks for calculation. We chose instead to train our hospitalists in a simpler method based on 2 linear measurements. Any loss of precision, however, was balanced by a large gain in simplicity. Regardless, minor variations in LA size are not likely to affect hospitalists' bedside evaluations. Finally, we did not validate the results of our reference standard (SE) by documenting interobserver reliability. Yet, because SE is generally accurate for the 6 cardiac abnormalities under study, the effect of this bias should be small.

These limitations can be addressed best by controlled trials of HCUE‐directed care. These trials will determine the clinical impact of hospitalist‐performed HCUE and, in turn, inform our design of HCUE training programs. As the current study shows, training hospitalists to participate in such trials is feasible: like other groups of noncardiologists, hospitalists can accurately perform HCUE after a brief training program. Whether or not hospitalists should perform HCUE requires further study.

Hand‐carried ultrasound echocardiography (HCUE) can help noncardiologists answer well‐defined questions at patients' bedsides in less than 10 minutes.1, 2 Indeed, intensivists3 and emergency department physicians4 already use HCUE to make rapid, point‐of‐care assessments. Since cardiovascular diagnoses are common among general medicine inpatients, HCUE may become an important skill for hospitalists to learn.5

However, uncertainty exists about the duration of HCUE training for hospitalists. In 2002, experts from the American Society of Echocardiography (ASE) published recommendations on training requirements for HCUE.6 With limited data on the safety or performance of HCUE training programs, which had just begun to emerge, the ASE borrowed from the proven training recommendations for standard echocardiography (SE). They recommended that all HCUE trainees, cardiologist and noncardiologist alike, complete level 1 SE training: 75 personally‐performed and 150 personally‐interpreted echocardiographic examinations. Since then, however, several HCUE training programs designed for noncardiologists have emerged.2, 5, 710 These alternative programs suggest that the ASE's recommended duration of training may be too long, particularly for focused HCUE that is limited to a few relatively simple assessments. It is important not to overshoot the requirements of HCUE training, because doing so may discourage groups of noncardiologists, like hospitalists, who may derive great benefits from HCUE.11

To address this uncertainty for hospitalists, we first developed a brief HCUE training program to assess 6 important cardiac abnormalities. We then studied the diagnostic accuracy of HCUE by hospitalists as a test of these 6 cardiac abnormalities assessed by SE.

Patients and Methods

Setting and Subjects

This prospective cohort study was performed at Stroger Hospital of Cook County, a 500‐bed public teaching hospital in Chicago, IL, from March through May of 2007. The cohort was adult inpatients who were referred for SE on weekdays from 3 distinct patient care units (Figure 1). We used 2 sampling modes to balance practical constraints (short‐stay unit [SSU] patients were more localized and, therefore, easier to study) with clinical diversity. We consecutively sampled patients from our SSU, where adults with provisional cardiovascular diagnoses are admitted if they might be eligible for discharge with in 3 days.12 But we used random number tables with a daily unique starting point to randomly sample patients from the general medical wards and the coronary care unit (CCU). Patients were excluded if repositioning them for HCUE was potentially harmful. The study was approved by our hospital's institutional review board, and we obtained written informed consent from all enrolled patients.

Figure 1

Results

During the 3 month study period, 654 patients were referred for SE from the 3 participating patient care units (Figure 1). Among these, 65 patients were ineligible because their SE was performed on the weekend and 178 other patients were not randomized from the general medical wards and CCU. From the remaining eligible patients, 322 underwent HCUE and 314 (98% of 322) underwent both SE and HCUE. Individual SE assessments were not interpretable (and therefore excluded) due to poor image quality for LA enlargement in 1 patient and IVC dilatation in 30 patients. Eighty‐three percent of patients who underwent SE (260/314) were referred to assess LV function (Table 3). The prevalence of the 6 clinically pertinent cardiac abnormalities under study ranged from 1% for moderate or large pericardial effusion to 25% for LV systolic dysfunction. Overall, 40% of patients had at least 1 out of 6 cardiac abnormalities.

Each hospitalist performed a similar total number of HCUE examinations (range, 3447). The median time difference between performance of SE and HCUE was 2.8 hours (25th75th percentiles, 1.45.1). Despite the high prevalence of chronic obstructive pulmonary disease and obesity, hospitalists considered HCUE assessments indeterminate in only 2% to 6% of the 6 assessments made for each patient (Table 4). Among the 38 patients (12% of 322) with any indeterminate HCUE assessment, 24 patients had only 1 out of 6 possible. Hospitalists completed HCUE in a median time of 28 minutes (25th‐75th percentiles, 2035), which included the time to record 7 best‐quality moving images and to fill out the research data collection form.

When HCUE results were analyzed as dichotomous values, positive likelihood ratios ranged from 2.5 to 21, and negative likelihood ratios ranged from 0 to 0.4 (Table 5). Positive and negative likelihood ratios were both sufficiency high and low to respectively increase and decrease by 5‐fold the prior odds of 3 out of 6 cardiac abnormalities: LV systolic dysfunction, moderate or severe MR regurgitation, and moderate or large pericardial effusion. Considering HCUE results as ordinal values for ROC analysis yielded additional diagnostic information (Figure 2). For example, the likelihood ratio of 1.0 (95% CI, 0.42.0) for borderline positive moderate or severe LA enlargement increased to 29 (range, 1362) for extreme positive results. Areas under the ROC curves were 0.9 for 4 out of 6 cardiac abnormalities.

Figure 2

LV systolic dysfunction and IVC dilatation were both prevalent enough to meet our criterion to examine interobserver variability; the mean number of abnormal patients per hospitalist was 10 patients for LV systolic dysfunction and 6 patients for IVC dilatation. For LV systolic dysfunction, SDs around mean sensitivity (84%) and specificity (87%) were 12% and 6%, respectively. For IVC dilatation, SDs around mean sensitivity (58%) and specificity (86%) were 24% and 7%, respectively.

Discussion

We found that, after a 27‐hour training program, hospitalists performed HCUE with moderate to excellent diagnostic accuracy for 6 important cardiac abnormalities. For example, hospitalists' assessments of LV systolic function yielded positive and negative likelihood ratios of 6.9 (95% CI, 4.99.8) and 0.2 (95% CI, 0.10.3), respectively. At the bedsides of patients with acute heart failure, therefore, hospitalists could use HCUE to lower or raise the 50:50 chance of LV systolic dysfunction30 to 15% or 85%, respectively. Whether or not these posttest likelihoods are extreme enough to cross important thresholds will depend on the clinical context. Yet these findings demonstrate how HCUE has the potential to provide hospitalists with valuable point‐of‐care data that are otherwise unavailableeither because routine clinical assessments are unreliable31 or because echocardiographic services are not immediately accessible.1

In fact, recent data from the Joint Commission on Accreditation of Healthcare Organizations shows how inaccessible SE may be. Approximately one‐quarter of hospitals in the United States send home about 10% of patients with acute heart failure without echocardiographic assessment of LV systolic function before, during, or immediately after hospitalization.32 In doing so, these hospitals leave unmet the 2002 National Quality Improvement Goal of universal assessment of LV systolic function for all heart failure patients. Hospitalists could close this quality gap with routine, 10‐minute HCUE assessments in all patients admitted with acute heart failure. (Our research HCUE protocol required a median time of 28 minutes, but this included time to assess 5 other cardiac abnormalities and collect data for research purposes). Until the clinical consequences of introducing hospitalist‐performed HCUE are studied, potential benefits like this are tentative. But our findings suggest that training hospitalists to accurately perform HCUE can be successfully accomplished in just 27 hours.

Other studies of HCUE training programs for noncardiologists have also challenged the opinion that learning to perform HCUE requires more than 100 hours of training.2, 711 Yet only 1 prior study has examined an HCUE training program for hospitalists.5 In this study by Martin et al.,5 hospitalists completed 5 supervised HCUE examinations and 6 hours of interpretation training before investigators scored their image acquisition and interpretation skills from 30 unsupervised HCUE examinations. To estimate their final skill levels at the completion of all 35 examinations by accounting for an initially steep learning curve, investigators then adjusted these scores with regression models. Despite these upward adjustments, hospitalists' image acquisition and interpretation scores were low in comparison to echocardiographic technicians and cardiology fellows. Besides these adjusted measurements of hospitalists' skills, however, Martin et al.5 unfortunately did not also report standard measures of diagnostic accuracy, like those proposed by the Standards for Reporting of Diagnostic Accuracy (STARD) initiative.33 Therefore, direct comparisons to the present study are difficult. Nevertheless, their findings suggest that a training program limited to 5 supervised HCUE examinations may be inadequate for hospitalists. In fact, the same group's earlier study of medical trainees suggested a minimum of 30 supervised HCUE examinations.9 We chose to design our hospitalist training program based on this minimum, though they surprisingly did not.5 As others continue to refine the components of hospitalist HCUE training programs, such as the optimal number of supervised examinations, our program could serve as a reasonable comparative example: more rigorous than the program designed by Martin et al.5 but more feasible than ASE level 1 training.

The number and complexity of assessments taught in HCUE training programs will determine their duration. With ongoing advancements in HCUE technology, there is a growing list of potential assessments to choose from. Although HCUE training programs ought to include assessments with proven clinical applications, there are no trials of HCUE‐directed care to inform such decisions. In their absence, therefore, we chose 6 assessments based on the following 3 criteria. First, our assessments were otherwise not reliably available from routine clinical data, such as the physical examination. Second, our assessments were straightforward: easy to learn and simple to perform. Here, we based our reasoning on an expectation that the value of HCUE lies not in highly complex, state‐of‐the‐art assessmentswhich are best left to echocardiographers equipped with SEbut in simple, routine assessments made with highly portable machines that grant noncardiologists newfound access to point‐of‐care data.34 Third, our assessments were clinically pertinent and, where appropriate, defined by cut‐points at levels of severity that often lead to changes in management. We suspect that setting high cut‐points has the salutary effects of making assessments easier to learn and more accurate, because distinguishing mild abnormalities is likely the most challenging aspect of echocardiographic interpretation.35 Whether or not our choices of assessments, and their cut‐points, are optimal has yet to be determined by future research designed to study how they affect patient outcomes. Given our hospitalists' performance in the present study, these assessments seem worthy of such future research.

Our study had several limitations. We studied physicians and patients from only 1 hospital; similar studies performed in different settings, particularly among patients with different proportions and manifestations of disease, may find different results. Nevertheless, our sampling method of prospectively enrolling consecutive patients strengthens our findings. Some echocardiographic measurement methods used by our hospitalists differed in subtle ways from echocardiography guideline recommendations.35 We chose our methods (Table 2) for 2 reasons. First, whenever possible, we chose methods of interpretation that coincided with our local cardiologists'. Second, we chose simplicity over precision. For example, the biplane method of disks, or modified Simpson's rule, is the preferred volumetric method of calculating LA size.35 This method requires tracing the contours of the LA in 2 planes and then dividing the LA volume into stacked oval disks for calculation. We chose instead to train our hospitalists in a simpler method based on 2 linear measurements. Any loss of precision, however, was balanced by a large gain in simplicity. Regardless, minor variations in LA size are not likely to affect hospitalists' bedside evaluations. Finally, we did not validate the results of our reference standard (SE) by documenting interobserver reliability. Yet, because SE is generally accurate for the 6 cardiac abnormalities under study, the effect of this bias should be small.

These limitations can be addressed best by controlled trials of HCUE‐directed care. These trials will determine the clinical impact of hospitalist‐performed HCUE and, in turn, inform our design of HCUE training programs. As the current study shows, training hospitalists to participate in such trials is feasible: like other groups of noncardiologists, hospitalists can accurately perform HCUE after a brief training program. Whether or not hospitalists should perform HCUE requires further study.

Hand‐carried ultrasound echocardiography (HCUE) can help noncardiologists answer well‐defined questions at patients' bedsides in less than 10 minutes.1, 2 Indeed, intensivists3 and emergency department physicians4 already use HCUE to make rapid, point‐of‐care assessments. Since cardiovascular diagnoses are common among general medicine inpatients, HCUE may become an important skill for hospitalists to learn.5

However, uncertainty exists about the duration of HCUE training for hospitalists. In 2002, experts from the American Society of Echocardiography (ASE) published recommendations on training requirements for HCUE.6 With limited data on the safety or performance of HCUE training programs, which had just begun to emerge, the ASE borrowed from the proven training recommendations for standard echocardiography (SE). They recommended that all HCUE trainees, cardiologist and noncardiologist alike, complete level 1 SE training: 75 personally‐performed and 150 personally‐interpreted echocardiographic examinations. Since then, however, several HCUE training programs designed for noncardiologists have emerged.2, 5, 710 These alternative programs suggest that the ASE's recommended duration of training may be too long, particularly for focused HCUE that is limited to a few relatively simple assessments. It is important not to overshoot the requirements of HCUE training, because doing so may discourage groups of noncardiologists, like hospitalists, who may derive great benefits from HCUE.11

To address this uncertainty for hospitalists, we first developed a brief HCUE training program to assess 6 important cardiac abnormalities. We then studied the diagnostic accuracy of HCUE by hospitalists as a test of these 6 cardiac abnormalities assessed by SE.

Patients and Methods

Setting and Subjects

This prospective cohort study was performed at Stroger Hospital of Cook County, a 500‐bed public teaching hospital in Chicago, IL, from March through May of 2007. The cohort was adult inpatients who were referred for SE on weekdays from 3 distinct patient care units (Figure 1). We used 2 sampling modes to balance practical constraints (short‐stay unit [SSU] patients were more localized and, therefore, easier to study) with clinical diversity. We consecutively sampled patients from our SSU, where adults with provisional cardiovascular diagnoses are admitted if they might be eligible for discharge with in 3 days.12 But we used random number tables with a daily unique starting point to randomly sample patients from the general medical wards and the coronary care unit (CCU). Patients were excluded if repositioning them for HCUE was potentially harmful. The study was approved by our hospital's institutional review board, and we obtained written informed consent from all enrolled patients.

Figure 1

Results

During the 3 month study period, 654 patients were referred for SE from the 3 participating patient care units (Figure 1). Among these, 65 patients were ineligible because their SE was performed on the weekend and 178 other patients were not randomized from the general medical wards and CCU. From the remaining eligible patients, 322 underwent HCUE and 314 (98% of 322) underwent both SE and HCUE. Individual SE assessments were not interpretable (and therefore excluded) due to poor image quality for LA enlargement in 1 patient and IVC dilatation in 30 patients. Eighty‐three percent of patients who underwent SE (260/314) were referred to assess LV function (Table 3). The prevalence of the 6 clinically pertinent cardiac abnormalities under study ranged from 1% for moderate or large pericardial effusion to 25% for LV systolic dysfunction. Overall, 40% of patients had at least 1 out of 6 cardiac abnormalities.

Each hospitalist performed a similar total number of HCUE examinations (range, 3447). The median time difference between performance of SE and HCUE was 2.8 hours (25th75th percentiles, 1.45.1). Despite the high prevalence of chronic obstructive pulmonary disease and obesity, hospitalists considered HCUE assessments indeterminate in only 2% to 6% of the 6 assessments made for each patient (Table 4). Among the 38 patients (12% of 322) with any indeterminate HCUE assessment, 24 patients had only 1 out of 6 possible. Hospitalists completed HCUE in a median time of 28 minutes (25th‐75th percentiles, 2035), which included the time to record 7 best‐quality moving images and to fill out the research data collection form.

When HCUE results were analyzed as dichotomous values, positive likelihood ratios ranged from 2.5 to 21, and negative likelihood ratios ranged from 0 to 0.4 (Table 5). Positive and negative likelihood ratios were both sufficiency high and low to respectively increase and decrease by 5‐fold the prior odds of 3 out of 6 cardiac abnormalities: LV systolic dysfunction, moderate or severe MR regurgitation, and moderate or large pericardial effusion. Considering HCUE results as ordinal values for ROC analysis yielded additional diagnostic information (Figure 2). For example, the likelihood ratio of 1.0 (95% CI, 0.42.0) for borderline positive moderate or severe LA enlargement increased to 29 (range, 1362) for extreme positive results. Areas under the ROC curves were 0.9 for 4 out of 6 cardiac abnormalities.

Figure 2

LV systolic dysfunction and IVC dilatation were both prevalent enough to meet our criterion to examine interobserver variability; the mean number of abnormal patients per hospitalist was 10 patients for LV systolic dysfunction and 6 patients for IVC dilatation. For LV systolic dysfunction, SDs around mean sensitivity (84%) and specificity (87%) were 12% and 6%, respectively. For IVC dilatation, SDs around mean sensitivity (58%) and specificity (86%) were 24% and 7%, respectively.

Discussion

We found that, after a 27‐hour training program, hospitalists performed HCUE with moderate to excellent diagnostic accuracy for 6 important cardiac abnormalities. For example, hospitalists' assessments of LV systolic function yielded positive and negative likelihood ratios of 6.9 (95% CI, 4.99.8) and 0.2 (95% CI, 0.10.3), respectively. At the bedsides of patients with acute heart failure, therefore, hospitalists could use HCUE to lower or raise the 50:50 chance of LV systolic dysfunction30 to 15% or 85%, respectively. Whether or not these posttest likelihoods are extreme enough to cross important thresholds will depend on the clinical context. Yet these findings demonstrate how HCUE has the potential to provide hospitalists with valuable point‐of‐care data that are otherwise unavailableeither because routine clinical assessments are unreliable31 or because echocardiographic services are not immediately accessible.1

In fact, recent data from the Joint Commission on Accreditation of Healthcare Organizations shows how inaccessible SE may be. Approximately one‐quarter of hospitals in the United States send home about 10% of patients with acute heart failure without echocardiographic assessment of LV systolic function before, during, or immediately after hospitalization.32 In doing so, these hospitals leave unmet the 2002 National Quality Improvement Goal of universal assessment of LV systolic function for all heart failure patients. Hospitalists could close this quality gap with routine, 10‐minute HCUE assessments in all patients admitted with acute heart failure. (Our research HCUE protocol required a median time of 28 minutes, but this included time to assess 5 other cardiac abnormalities and collect data for research purposes). Until the clinical consequences of introducing hospitalist‐performed HCUE are studied, potential benefits like this are tentative. But our findings suggest that training hospitalists to accurately perform HCUE can be successfully accomplished in just 27 hours.

Other studies of HCUE training programs for noncardiologists have also challenged the opinion that learning to perform HCUE requires more than 100 hours of training.2, 711 Yet only 1 prior study has examined an HCUE training program for hospitalists.5 In this study by Martin et al.,5 hospitalists completed 5 supervised HCUE examinations and 6 hours of interpretation training before investigators scored their image acquisition and interpretation skills from 30 unsupervised HCUE examinations. To estimate their final skill levels at the completion of all 35 examinations by accounting for an initially steep learning curve, investigators then adjusted these scores with regression models. Despite these upward adjustments, hospitalists' image acquisition and interpretation scores were low in comparison to echocardiographic technicians and cardiology fellows. Besides these adjusted measurements of hospitalists' skills, however, Martin et al.5 unfortunately did not also report standard measures of diagnostic accuracy, like those proposed by the Standards for Reporting of Diagnostic Accuracy (STARD) initiative.33 Therefore, direct comparisons to the present study are difficult. Nevertheless, their findings suggest that a training program limited to 5 supervised HCUE examinations may be inadequate for hospitalists. In fact, the same group's earlier study of medical trainees suggested a minimum of 30 supervised HCUE examinations.9 We chose to design our hospitalist training program based on this minimum, though they surprisingly did not.5 As others continue to refine the components of hospitalist HCUE training programs, such as the optimal number of supervised examinations, our program could serve as a reasonable comparative example: more rigorous than the program designed by Martin et al.5 but more feasible than ASE level 1 training.

The number and complexity of assessments taught in HCUE training programs will determine their duration. With ongoing advancements in HCUE technology, there is a growing list of potential assessments to choose from. Although HCUE training programs ought to include assessments with proven clinical applications, there are no trials of HCUE‐directed care to inform such decisions. In their absence, therefore, we chose 6 assessments based on the following 3 criteria. First, our assessments were otherwise not reliably available from routine clinical data, such as the physical examination. Second, our assessments were straightforward: easy to learn and simple to perform. Here, we based our reasoning on an expectation that the value of HCUE lies not in highly complex, state‐of‐the‐art assessmentswhich are best left to echocardiographers equipped with SEbut in simple, routine assessments made with highly portable machines that grant noncardiologists newfound access to point‐of‐care data.34 Third, our assessments were clinically pertinent and, where appropriate, defined by cut‐points at levels of severity that often lead to changes in management. We suspect that setting high cut‐points has the salutary effects of making assessments easier to learn and more accurate, because distinguishing mild abnormalities is likely the most challenging aspect of echocardiographic interpretation.35 Whether or not our choices of assessments, and their cut‐points, are optimal has yet to be determined by future research designed to study how they affect patient outcomes. Given our hospitalists' performance in the present study, these assessments seem worthy of such future research.

Our study had several limitations. We studied physicians and patients from only 1 hospital; similar studies performed in different settings, particularly among patients with different proportions and manifestations of disease, may find different results. Nevertheless, our sampling method of prospectively enrolling consecutive patients strengthens our findings. Some echocardiographic measurement methods used by our hospitalists differed in subtle ways from echocardiography guideline recommendations.35 We chose our methods (Table 2) for 2 reasons. First, whenever possible, we chose methods of interpretation that coincided with our local cardiologists'. Second, we chose simplicity over precision. For example, the biplane method of disks, or modified Simpson's rule, is the preferred volumetric method of calculating LA size.35 This method requires tracing the contours of the LA in 2 planes and then dividing the LA volume into stacked oval disks for calculation. We chose instead to train our hospitalists in a simpler method based on 2 linear measurements. Any loss of precision, however, was balanced by a large gain in simplicity. Regardless, minor variations in LA size are not likely to affect hospitalists' bedside evaluations. Finally, we did not validate the results of our reference standard (SE) by documenting interobserver reliability. Yet, because SE is generally accurate for the 6 cardiac abnormalities under study, the effect of this bias should be small.

These limitations can be addressed best by controlled trials of HCUE‐directed care. These trials will determine the clinical impact of hospitalist‐performed HCUE and, in turn, inform our design of HCUE training programs. As the current study shows, training hospitalists to participate in such trials is feasible: like other groups of noncardiologists, hospitalists can accurately perform HCUE after a brief training program. Whether or not hospitalists should perform HCUE requires further study.