Rudolf Kumapley, MBChB

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

SE Protocol

As part of enrolled patients' routine clinical care, SE images were acquired and interpreted in the usual fashion in our hospital's echocardiography laboratory, which performs SE on over 7,000 patients per year. Echocardiographic technicians acquired images with a General Electric Vivid 7 cardiac ultrasound machine (General Electric, Milwaukee, WI) equipped with a GE M4S 1.8 to 3.4 MHz cardiac transducer (General Electric). Technicians followed the standard adult transthoracic echocardiography scanning protocol to acquire 40 to 100 images on every patient using all available echocardiographic modalities: 2‐dimensional, M‐mode, color Doppler, continuous‐wave Doppler, pulse‐wave Doppler, and tissue Doppler.13 Blinded to HCUE results, attending physician cardiologist echocardiographers then interpreted archived images using computer software (Centricity System; General Electric) to generate final reports that were entered into patients' medical records. This software ensured that final reports were standardized, because echocardiographers' final qualitative assessments were limited to short lists of standard options; for example, in reporting left atrium (LA) size, echocardiographers chose from only 5 standard options: normal, mildly dilated, moderately dilated, severely dilated, and not interpretable. Investigators, who were also blinded to HCUE results, later abstracted SE results from these standardized report forms in patients' medical records. All echocardiographers fulfilled ASE training guidelines to independently interpret SE: a minimum of 150 personally‐performed and 300 personally‐interpreted echocardiographic examinations (training level 2).14

HCUE Training

Based on the recommendations of our cardiologist investigator (B.M.), we developed a training program for 1 hospitalist to become an HCUE instructor. Our instructor trainee (C.C.) was board‐eligible in internal medicine but had no previous formal training in cardiology or echocardiography. We a priori established that her training would continue until our cardiologist investigator determined that she was ready to train other hospitalists; this determination occurred after 5 weeks. She learned image acquisition by performing focused SE on 30 patients under the direct supervision of an echocardiographic technician. She also performed focused HCUE on 65 inpatients without direct supervision but with ongoing access to consult the technician to review archived images and troubleshoot difficulties with acquisition. She learned image interpretation by reading relevant chapters from a SE textbook15 and by participating in daily didactic sessions in which attending cardiologist echocardiographers train cardiology fellows in SE interpretation.

This hospitalist then served as the HCUE instructor for 8 other attending physician hospitalists who were board‐certified internists with no previous formal training in cardiology or echocardiography. The training program was limited to acquisition and interpretation of 2‐dimensional grayscale and color Doppler images for the 6 cardiac assessments under study (Table 1). The instructor marshaled pairs of hospitalists through the 3 components of the training program, which lasted a total of 27 hours.

First, hospitalists attended a 2‐hour lecture on the basic principles of HCUE. Slides from this lecture and additional images of normal and abnormal findings were provided to each hospitalist on a digital video disc. Second, each hospitalist underwent 20 hours of hands‐on training in 2‐hour sessions scheduled over 2 weeks. Willing inpatients from our hospital's emergency department were used as volunteers for these hand‐on training sessions. During these sessions the instructor provided practical suggestions to optimize image quality, such as transducer location and patient positioning. In the first 3 sessions, the minimum pace was 1 patient per hour; thereafter, the pace was increased to 1 patient per half‐hour. We chose 20 hours of hands‐on training and these minimum paces because they allowed each hospitalist to attain a cumulative experience of no less than 30 patientsan amount that heralds a flattening of the HCUE learning curve among medical trainees.9 Third, each pair of hospitalists received feedback from a cardiologist investigator (B.M.) who critiqued the quality and interpretation of images acquired by hospitalists during hands‐on training sessions. Since image quality varies by patient,16 hospitalists' images were compared side‐by‐side to images recorded by the instructor on the same patients. The cardiologist also critiqued hospitalists' interpretations of both their own images and additional sets of archived images from patients with abnormal findings.

HCUE Protocol

After completing the training program and blinded to the results of SE, the 8 hospitalists performed HCUE on enrolled patients within hours of SE. We limited the time interval between tests to minimize the effect that changes in physiologic variables, such as blood pressure and intravascular volume, have on the reliability of serial echocardiographic measurements.16 Hospitalists performed HCUE with a MicroMaxx 3.4 hand‐carried ultrasound machine equipped with a cardiology software package and a 1 to 5 MHz P17 cardiac transducer (Sonosite, Inc., Bothell, WA); simultaneous electrocardiographic recording, though available, was not used. While patients laid on their own standard hospital beds or on a standard hospital gurney in a room adjacent to the SE waiting room, hospitalists positioned them without assistance from nursing staff and recorded 7 best‐quality images per patient. Patients were first positioned in a partial (3045 degrees) left lateral decubitus position to record 4 grayscale images of the short‐axis and long‐axis parasternal and 2‐chamber and 4‐chamber apical views; 2 color Doppler images of the mitral inflow were also recorded from the long‐axis parasternal and the 4‐chamber apical views. Patients were then positioned supine to record 1 grayscale image of the inferior vena cava (IVC) from the transhepatic view. Hospitalists did not perform a history or physical exam on enrolled patients, nor did they review patients' medical records.

Immediately following the HCUE, hospitalists replayed the recorded images as often as needed and entered final interpretations on data collection forms. Linear measurements were made manually with a caliper held directly to the hand‐carried ultrasound monitor. These measurements were then translated into qualitative assessments based on standard values used by our hospital's echocardiographers (Table 2).17 When a hospitalist could not confidently assess a cardiac abnormality, the final HCUE assessment was recorded as indeterminate. Hospitalists also recorded the time to perform each HCUE, which included the time to record 7 best‐quality images, to interpret the findings, and to fill out the data collection form.

Data Analysis

We based our sample size calculations on earlier reports of HCUE by noncardiologist trainees for assessment of left ventricular (LV) systolic function.7, 10 From these reports, we estimated a negative likelihood ratio of 0.3. In addition, we expected about a quarter of our patients to have LV systolic dysfunction (B.M., personal communication). Therefore, to achieve 95% confidence intervals (CIs) around the point estimate of a negative likelihood ratio that excluded 0.50, our upper bound for a clinically meaningful result, we needed a sample size of approximately 300 patients.18

We defined threshold levels of ordinal severity for the 6 cardiac abnormalities under study based on their clinical pertinence to hospitalists (Table 2). Here, we reasoned that abnormalities at or above these levels would likely lead to important changes in hospitalists' management of inpatients; abnormalities below these levels rarely represent cardiac disease that is worthy of an immediate change in management. Since even mild degrees of LV dysfunction have important diagnostic and therapeutic implications for most general medicine inpatients, particularly those presenting with heart failure,19 we set our threshold for LV dysfunction at mild or greater. In contrast, since neither mild nor moderate mitral regurgitation (MR) has immediate implications for medical or surgical therapy even if symptoms or LV dysfunction are present,20 we set our threshold for MR at severe. Similarly, though mild LA enlargement21 and mild LV hypertrophy22 have clear prognostic implications for patients' chronic medical conditions, we reasoned that only moderate or severe versions likely reflect underlying abnormalities that affect hospitalists' point‐of‐care decision‐making. Since cardiac tamponade is rarely both subclinical23 and due to a small pericardial effusion,24 we set our threshold for pericardial effusion size at moderate or large. Finally, we set our threshold IVC diameter, a marker of central venous volume status,25 at dilated, because volume overload is an important consideration in hospitalized cardiac patients.

Using these thresholds, investigators dichotomized echocardiographers' SE readings as normal or abnormal for each of the 6 cardiac abnormalities under study to serve as the reference standards. Hospitalists' HCUE results were then compared to the reference standards in 2 different ways. We first analyzed HCUE results as dichotomous values to calculate conventional sensitivity, specificity, and positive and negative likelihood ratios. Here we considered indeterminate HCUE results positive in a clinically conservative tradeoff that neither ignores indeterminate results nor risks falsely classifying them as negative.26 We then analyzed hospitalists' HCUE results as ordinal values for receiver operating characteristic (ROC) curve analysis. Here we considered an indeterminate result as 1 possible test result.27

To examine interobserver variability of HCUE, we first chose from the 6 possible assessments only those with a mean number of abnormal patients per hospitalist greater than 5. We reasoned that variability among assessments with lower prevalence would be predictably wide and inconclusive. We then expressed variability as standard deviations (SDs) around mean sensitivity and specificity for the 8 hospitalists.

The CIs for likelihood ratios were constructed using the likelihood‐based approach to binomial proportions of Koopman.28 The areas under ROC curves were computed using the trapezoidal rule, and the CIs for these areas were constructed using the algorithm described by DeLong et al.29 All analyses were conducted with Stata Statistical Software, Release 10 (StataCorp, College Station, TX).

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

SE Protocol

As part of enrolled patients' routine clinical care, SE images were acquired and interpreted in the usual fashion in our hospital's echocardiography laboratory, which performs SE on over 7,000 patients per year. Echocardiographic technicians acquired images with a General Electric Vivid 7 cardiac ultrasound machine (General Electric, Milwaukee, WI) equipped with a GE M4S 1.8 to 3.4 MHz cardiac transducer (General Electric). Technicians followed the standard adult transthoracic echocardiography scanning protocol to acquire 40 to 100 images on every patient using all available echocardiographic modalities: 2‐dimensional, M‐mode, color Doppler, continuous‐wave Doppler, pulse‐wave Doppler, and tissue Doppler.13 Blinded to HCUE results, attending physician cardiologist echocardiographers then interpreted archived images using computer software (Centricity System; General Electric) to generate final reports that were entered into patients' medical records. This software ensured that final reports were standardized, because echocardiographers' final qualitative assessments were limited to short lists of standard options; for example, in reporting left atrium (LA) size, echocardiographers chose from only 5 standard options: normal, mildly dilated, moderately dilated, severely dilated, and not interpretable. Investigators, who were also blinded to HCUE results, later abstracted SE results from these standardized report forms in patients' medical records. All echocardiographers fulfilled ASE training guidelines to independently interpret SE: a minimum of 150 personally‐performed and 300 personally‐interpreted echocardiographic examinations (training level 2).14

HCUE Training

Based on the recommendations of our cardiologist investigator (B.M.), we developed a training program for 1 hospitalist to become an HCUE instructor. Our instructor trainee (C.C.) was board‐eligible in internal medicine but had no previous formal training in cardiology or echocardiography. We a priori established that her training would continue until our cardiologist investigator determined that she was ready to train other hospitalists; this determination occurred after 5 weeks. She learned image acquisition by performing focused SE on 30 patients under the direct supervision of an echocardiographic technician. She also performed focused HCUE on 65 inpatients without direct supervision but with ongoing access to consult the technician to review archived images and troubleshoot difficulties with acquisition. She learned image interpretation by reading relevant chapters from a SE textbook15 and by participating in daily didactic sessions in which attending cardiologist echocardiographers train cardiology fellows in SE interpretation.

This hospitalist then served as the HCUE instructor for 8 other attending physician hospitalists who were board‐certified internists with no previous formal training in cardiology or echocardiography. The training program was limited to acquisition and interpretation of 2‐dimensional grayscale and color Doppler images for the 6 cardiac assessments under study (Table 1). The instructor marshaled pairs of hospitalists through the 3 components of the training program, which lasted a total of 27 hours.

First, hospitalists attended a 2‐hour lecture on the basic principles of HCUE. Slides from this lecture and additional images of normal and abnormal findings were provided to each hospitalist on a digital video disc. Second, each hospitalist underwent 20 hours of hands‐on training in 2‐hour sessions scheduled over 2 weeks. Willing inpatients from our hospital's emergency department were used as volunteers for these hand‐on training sessions. During these sessions the instructor provided practical suggestions to optimize image quality, such as transducer location and patient positioning. In the first 3 sessions, the minimum pace was 1 patient per hour; thereafter, the pace was increased to 1 patient per half‐hour. We chose 20 hours of hands‐on training and these minimum paces because they allowed each hospitalist to attain a cumulative experience of no less than 30 patientsan amount that heralds a flattening of the HCUE learning curve among medical trainees.9 Third, each pair of hospitalists received feedback from a cardiologist investigator (B.M.) who critiqued the quality and interpretation of images acquired by hospitalists during hands‐on training sessions. Since image quality varies by patient,16 hospitalists' images were compared side‐by‐side to images recorded by the instructor on the same patients. The cardiologist also critiqued hospitalists' interpretations of both their own images and additional sets of archived images from patients with abnormal findings.

HCUE Protocol

After completing the training program and blinded to the results of SE, the 8 hospitalists performed HCUE on enrolled patients within hours of SE. We limited the time interval between tests to minimize the effect that changes in physiologic variables, such as blood pressure and intravascular volume, have on the reliability of serial echocardiographic measurements.16 Hospitalists performed HCUE with a MicroMaxx 3.4 hand‐carried ultrasound machine equipped with a cardiology software package and a 1 to 5 MHz P17 cardiac transducer (Sonosite, Inc., Bothell, WA); simultaneous electrocardiographic recording, though available, was not used. While patients laid on their own standard hospital beds or on a standard hospital gurney in a room adjacent to the SE waiting room, hospitalists positioned them without assistance from nursing staff and recorded 7 best‐quality images per patient. Patients were first positioned in a partial (3045 degrees) left lateral decubitus position to record 4 grayscale images of the short‐axis and long‐axis parasternal and 2‐chamber and 4‐chamber apical views; 2 color Doppler images of the mitral inflow were also recorded from the long‐axis parasternal and the 4‐chamber apical views. Patients were then positioned supine to record 1 grayscale image of the inferior vena cava (IVC) from the transhepatic view. Hospitalists did not perform a history or physical exam on enrolled patients, nor did they review patients' medical records.

Immediately following the HCUE, hospitalists replayed the recorded images as often as needed and entered final interpretations on data collection forms. Linear measurements were made manually with a caliper held directly to the hand‐carried ultrasound monitor. These measurements were then translated into qualitative assessments based on standard values used by our hospital's echocardiographers (Table 2).17 When a hospitalist could not confidently assess a cardiac abnormality, the final HCUE assessment was recorded as indeterminate. Hospitalists also recorded the time to perform each HCUE, which included the time to record 7 best‐quality images, to interpret the findings, and to fill out the data collection form.

Data Analysis

We based our sample size calculations on earlier reports of HCUE by noncardiologist trainees for assessment of left ventricular (LV) systolic function.7, 10 From these reports, we estimated a negative likelihood ratio of 0.3. In addition, we expected about a quarter of our patients to have LV systolic dysfunction (B.M., personal communication). Therefore, to achieve 95% confidence intervals (CIs) around the point estimate of a negative likelihood ratio that excluded 0.50, our upper bound for a clinically meaningful result, we needed a sample size of approximately 300 patients.18

We defined threshold levels of ordinal severity for the 6 cardiac abnormalities under study based on their clinical pertinence to hospitalists (Table 2). Here, we reasoned that abnormalities at or above these levels would likely lead to important changes in hospitalists' management of inpatients; abnormalities below these levels rarely represent cardiac disease that is worthy of an immediate change in management. Since even mild degrees of LV dysfunction have important diagnostic and therapeutic implications for most general medicine inpatients, particularly those presenting with heart failure,19 we set our threshold for LV dysfunction at mild or greater. In contrast, since neither mild nor moderate mitral regurgitation (MR) has immediate implications for medical or surgical therapy even if symptoms or LV dysfunction are present,20 we set our threshold for MR at severe. Similarly, though mild LA enlargement21 and mild LV hypertrophy22 have clear prognostic implications for patients' chronic medical conditions, we reasoned that only moderate or severe versions likely reflect underlying abnormalities that affect hospitalists' point‐of‐care decision‐making. Since cardiac tamponade is rarely both subclinical23 and due to a small pericardial effusion,24 we set our threshold for pericardial effusion size at moderate or large. Finally, we set our threshold IVC diameter, a marker of central venous volume status,25 at dilated, because volume overload is an important consideration in hospitalized cardiac patients.

Using these thresholds, investigators dichotomized echocardiographers' SE readings as normal or abnormal for each of the 6 cardiac abnormalities under study to serve as the reference standards. Hospitalists' HCUE results were then compared to the reference standards in 2 different ways. We first analyzed HCUE results as dichotomous values to calculate conventional sensitivity, specificity, and positive and negative likelihood ratios. Here we considered indeterminate HCUE results positive in a clinically conservative tradeoff that neither ignores indeterminate results nor risks falsely classifying them as negative.26 We then analyzed hospitalists' HCUE results as ordinal values for receiver operating characteristic (ROC) curve analysis. Here we considered an indeterminate result as 1 possible test result.27

To examine interobserver variability of HCUE, we first chose from the 6 possible assessments only those with a mean number of abnormal patients per hospitalist greater than 5. We reasoned that variability among assessments with lower prevalence would be predictably wide and inconclusive. We then expressed variability as standard deviations (SDs) around mean sensitivity and specificity for the 8 hospitalists.

The CIs for likelihood ratios were constructed using the likelihood‐based approach to binomial proportions of Koopman.28 The areas under ROC curves were computed using the trapezoidal rule, and the CIs for these areas were constructed using the algorithm described by DeLong et al.29 All analyses were conducted with Stata Statistical Software, Release 10 (StataCorp, College Station, TX).

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

SE Protocol

As part of enrolled patients' routine clinical care, SE images were acquired and interpreted in the usual fashion in our hospital's echocardiography laboratory, which performs SE on over 7,000 patients per year. Echocardiographic technicians acquired images with a General Electric Vivid 7 cardiac ultrasound machine (General Electric, Milwaukee, WI) equipped with a GE M4S 1.8 to 3.4 MHz cardiac transducer (General Electric). Technicians followed the standard adult transthoracic echocardiography scanning protocol to acquire 40 to 100 images on every patient using all available echocardiographic modalities: 2‐dimensional, M‐mode, color Doppler, continuous‐wave Doppler, pulse‐wave Doppler, and tissue Doppler.13 Blinded to HCUE results, attending physician cardiologist echocardiographers then interpreted archived images using computer software (Centricity System; General Electric) to generate final reports that were entered into patients' medical records. This software ensured that final reports were standardized, because echocardiographers' final qualitative assessments were limited to short lists of standard options; for example, in reporting left atrium (LA) size, echocardiographers chose from only 5 standard options: normal, mildly dilated, moderately dilated, severely dilated, and not interpretable. Investigators, who were also blinded to HCUE results, later abstracted SE results from these standardized report forms in patients' medical records. All echocardiographers fulfilled ASE training guidelines to independently interpret SE: a minimum of 150 personally‐performed and 300 personally‐interpreted echocardiographic examinations (training level 2).14

HCUE Training

Based on the recommendations of our cardiologist investigator (B.M.), we developed a training program for 1 hospitalist to become an HCUE instructor. Our instructor trainee (C.C.) was board‐eligible in internal medicine but had no previous formal training in cardiology or echocardiography. We a priori established that her training would continue until our cardiologist investigator determined that she was ready to train other hospitalists; this determination occurred after 5 weeks. She learned image acquisition by performing focused SE on 30 patients under the direct supervision of an echocardiographic technician. She also performed focused HCUE on 65 inpatients without direct supervision but with ongoing access to consult the technician to review archived images and troubleshoot difficulties with acquisition. She learned image interpretation by reading relevant chapters from a SE textbook15 and by participating in daily didactic sessions in which attending cardiologist echocardiographers train cardiology fellows in SE interpretation.

This hospitalist then served as the HCUE instructor for 8 other attending physician hospitalists who were board‐certified internists with no previous formal training in cardiology or echocardiography. The training program was limited to acquisition and interpretation of 2‐dimensional grayscale and color Doppler images for the 6 cardiac assessments under study (Table 1). The instructor marshaled pairs of hospitalists through the 3 components of the training program, which lasted a total of 27 hours.

First, hospitalists attended a 2‐hour lecture on the basic principles of HCUE. Slides from this lecture and additional images of normal and abnormal findings were provided to each hospitalist on a digital video disc. Second, each hospitalist underwent 20 hours of hands‐on training in 2‐hour sessions scheduled over 2 weeks. Willing inpatients from our hospital's emergency department were used as volunteers for these hand‐on training sessions. During these sessions the instructor provided practical suggestions to optimize image quality, such as transducer location and patient positioning. In the first 3 sessions, the minimum pace was 1 patient per hour; thereafter, the pace was increased to 1 patient per half‐hour. We chose 20 hours of hands‐on training and these minimum paces because they allowed each hospitalist to attain a cumulative experience of no less than 30 patientsan amount that heralds a flattening of the HCUE learning curve among medical trainees.9 Third, each pair of hospitalists received feedback from a cardiologist investigator (B.M.) who critiqued the quality and interpretation of images acquired by hospitalists during hands‐on training sessions. Since image quality varies by patient,16 hospitalists' images were compared side‐by‐side to images recorded by the instructor on the same patients. The cardiologist also critiqued hospitalists' interpretations of both their own images and additional sets of archived images from patients with abnormal findings.

HCUE Protocol

After completing the training program and blinded to the results of SE, the 8 hospitalists performed HCUE on enrolled patients within hours of SE. We limited the time interval between tests to minimize the effect that changes in physiologic variables, such as blood pressure and intravascular volume, have on the reliability of serial echocardiographic measurements.16 Hospitalists performed HCUE with a MicroMaxx 3.4 hand‐carried ultrasound machine equipped with a cardiology software package and a 1 to 5 MHz P17 cardiac transducer (Sonosite, Inc., Bothell, WA); simultaneous electrocardiographic recording, though available, was not used. While patients laid on their own standard hospital beds or on a standard hospital gurney in a room adjacent to the SE waiting room, hospitalists positioned them without assistance from nursing staff and recorded 7 best‐quality images per patient. Patients were first positioned in a partial (3045 degrees) left lateral decubitus position to record 4 grayscale images of the short‐axis and long‐axis parasternal and 2‐chamber and 4‐chamber apical views; 2 color Doppler images of the mitral inflow were also recorded from the long‐axis parasternal and the 4‐chamber apical views. Patients were then positioned supine to record 1 grayscale image of the inferior vena cava (IVC) from the transhepatic view. Hospitalists did not perform a history or physical exam on enrolled patients, nor did they review patients' medical records.

Immediately following the HCUE, hospitalists replayed the recorded images as often as needed and entered final interpretations on data collection forms. Linear measurements were made manually with a caliper held directly to the hand‐carried ultrasound monitor. These measurements were then translated into qualitative assessments based on standard values used by our hospital's echocardiographers (Table 2).17 When a hospitalist could not confidently assess a cardiac abnormality, the final HCUE assessment was recorded as indeterminate. Hospitalists also recorded the time to perform each HCUE, which included the time to record 7 best‐quality images, to interpret the findings, and to fill out the data collection form.

Data Analysis

We based our sample size calculations on earlier reports of HCUE by noncardiologist trainees for assessment of left ventricular (LV) systolic function.7, 10 From these reports, we estimated a negative likelihood ratio of 0.3. In addition, we expected about a quarter of our patients to have LV systolic dysfunction (B.M., personal communication). Therefore, to achieve 95% confidence intervals (CIs) around the point estimate of a negative likelihood ratio that excluded 0.50, our upper bound for a clinically meaningful result, we needed a sample size of approximately 300 patients.18

We defined threshold levels of ordinal severity for the 6 cardiac abnormalities under study based on their clinical pertinence to hospitalists (Table 2). Here, we reasoned that abnormalities at or above these levels would likely lead to important changes in hospitalists' management of inpatients; abnormalities below these levels rarely represent cardiac disease that is worthy of an immediate change in management. Since even mild degrees of LV dysfunction have important diagnostic and therapeutic implications for most general medicine inpatients, particularly those presenting with heart failure,19 we set our threshold for LV dysfunction at mild or greater. In contrast, since neither mild nor moderate mitral regurgitation (MR) has immediate implications for medical or surgical therapy even if symptoms or LV dysfunction are present,20 we set our threshold for MR at severe. Similarly, though mild LA enlargement21 and mild LV hypertrophy22 have clear prognostic implications for patients' chronic medical conditions, we reasoned that only moderate or severe versions likely reflect underlying abnormalities that affect hospitalists' point‐of‐care decision‐making. Since cardiac tamponade is rarely both subclinical23 and due to a small pericardial effusion,24 we set our threshold for pericardial effusion size at moderate or large. Finally, we set our threshold IVC diameter, a marker of central venous volume status,25 at dilated, because volume overload is an important consideration in hospitalized cardiac patients.

Using these thresholds, investigators dichotomized echocardiographers' SE readings as normal or abnormal for each of the 6 cardiac abnormalities under study to serve as the reference standards. Hospitalists' HCUE results were then compared to the reference standards in 2 different ways. We first analyzed HCUE results as dichotomous values to calculate conventional sensitivity, specificity, and positive and negative likelihood ratios. Here we considered indeterminate HCUE results positive in a clinically conservative tradeoff that neither ignores indeterminate results nor risks falsely classifying them as negative.26 We then analyzed hospitalists' HCUE results as ordinal values for receiver operating characteristic (ROC) curve analysis. Here we considered an indeterminate result as 1 possible test result.27

To examine interobserver variability of HCUE, we first chose from the 6 possible assessments only those with a mean number of abnormal patients per hospitalist greater than 5. We reasoned that variability among assessments with lower prevalence would be predictably wide and inconclusive. We then expressed variability as standard deviations (SDs) around mean sensitivity and specificity for the 8 hospitalists.

The CIs for likelihood ratios were constructed using the likelihood‐based approach to binomial proportions of Koopman.28 The areas under ROC curves were computed using the trapezoidal rule, and the CIs for these areas were constructed using the algorithm described by DeLong et al.29 All analyses were conducted with Stata Statistical Software, Release 10 (StataCorp, College Station, TX).

Design and Setting

In this prospective cohort study, we included all patients admitted to the hospitalist‐run SSU of Cook County Hospital, a 500‐bed public teaching hospital in Chicago, Illinois, from January through April of 2006. Our 14‐bed SSU opened in 2002 to reduce overcrowding on the traditional inpatient wards by admitting adult patients who require inpatient care but might be eligible for discharge within 3 days. The unit is geographically part of the ED but is staffed by resident physicians and a rotating group of hospitalist attending physicians from the Department of Medicine. At least 1 attending and resident physician are available throughout the day, including weekend days and holidays; evenings are covered by a resident who presents overnight admissions to an attending physician the following morning.

ED physicians admit general medical patients to the SSU 24 hours per day, 7 days per week. Though admissions do not require prior approval from SSU physicians, the Departments of Medicine and Emergency Medicine have collaboratively promoted 5 suggested admission‐location guidelines to admitting ED physicians (Figure 1). For candidate SSU patients, these 5 guidelines are not intended to be restrictive but to provide a framework for the complex decision‐making process that our ED physicians encounter, particularly during periods of extreme overcrowding.7 First, patients should have an anticipated stay shorter than 72 hours. Second, patients should not have an eventual need for admission to traditional inpatient services such as the general medicine wards or intensive care units; this guideline is intended to improve patient safety by reducing unnecessary handoffs between physicians.8 Third, patients with provisional cardiovascular diagnoses should be preferentially admitted to the SSU over the general medical wards; this guideline is intended to improve hospital‐wide efficiency because our SSU is equipped with continuous telemetry monitors, an exercise treadmill testing (ETT) laboratory, and other reserved cardiac tests (see Admission Characteristics and Services Received section, below). Fourth, patients' risk should be no higher than intermediate level. Admitting ED physicians are encouraged to use posted risk estimators for patients with provisional diagnoses of possible acute coronary syndrome (ACS), decompensated heart failure, asthma exacerbation, and out‐of‐control diabetes. Finally, patients should not need advanced ancillary services; these include bedside procedures (eg, central venous catheter insertions), time‐intensive nursing (eg, regular dressing changes), and complex social‐services (eg, long‐term care facility placements).

Figure 1

Subjects

The study subjects were all patients admitted to the SSU during the 4‐month study period. Patients were excluded from the entire study if they refused verbal consent to participate. All patients who consented were included in the description of patient admission characteristics. Thirteen patients who prematurely left the SSU against medical advice, however, were neither included in the descriptions of services received nor in the analyses of predictors of successful SSU stays. We excluded these patients because they needed services that they did not receiveincluding these patients in our analysis would tend to overestimate the efficiency of our SSU by shortening the length‐of‐stay (LOS) without adding diagnostic tests or treatments.

Data Collection

After receiving approval from the institutional review board, attending physician investigators conducted an interview, physical examination, and review of medical records for each enrolled patient within 12 hours of admission to the SSU. When ED attending physicians' provisional primary diagnoses included possible ACS or decompensated heart failure, which we knew from earlier pilot data were our 2 most common provisional diagnoses, investigators gathered patient data to be applied in validated models of risk after the study period (Figure 2).9, 10 Some of the clinical predictors required for these models are based on patients' findings on presentation to the ED. For example, Goldman's risk model for major cardiac events uses patients' initial systolic blood pressures on presentation to the ED.9 In such cases, investigators gathered needed data from electronic and paper charts generated in the ED. Upon discharge from the SSU, investigators reviewed patients' medical records a second time. All data were entered by investigators and instantly committed into an online database.

Figure 2

Admission Characteristics and Services Received

Patients were grouped according to the provisional diagnoses of ED attending physicians upon admission to the SSU (Figure 2). We chose to group patients by the provisional diagnoses of EDnot SSUattending physicians to better understand how ED physicians, the physicians who make the admission‐location decisions in our hospital, were using the SSU. Patients were first grouped as having possible ACS or heart failure, because patients with these provisional diagnoses were preferentially admitted to the SSU (Figure 1). When neither diagnosis was listed, patients were grouped according to ED attending physicians' first‐listed diagnoses. At the end of the study period, relevant risk models were applied to patients with possible ACS or heart failure and stratified as very low, low, intermediate, or high risk.9, 10 Patients with both possible ACS and heart failure were grouped according to the diagnosis with the highest corresponding risk assessment; if both risk assessments were the same, then the first‐listed diagnosis was used. Though developed to predict different clinical outcomes during different time periods, risk strata from the corresponding risk models were pooled across both diagnoses to develop a risk summary.

Upon discharge, investigators recorded which advanced diagnostic tests, specialty consultations, and acute care treatments patients received while in the SSU. Diagnostic tests were considered advanced if they were not routinely performed within 2 hours of being ordered. Advanced diagnostic tests were grouped into 2 types by their accessibility to ordering SSU physicians. Open access tests included echocardiograms and ETTs, which were reserved for SSU patients 6 days per week. Though the availability of open access tests was not unlimited, ordering physicians' needs for them rarely exceeded the immediate supply. On the other hand, limited access tests included both cardiac stress imaging studies, which were reserved for SSU patients on a very limited basis 4 or 5 days per week, and other tests that were not reserved for SSU patients, such as endoscopy, magnetic resonance imaging, or ultrasonography. Ordering physicians' needs for limited access tests often exceeded their immediate supply; in such cases, SSU patients were placed without priority into queues that included patients from the entire hospital.

Investigators recorded when patients received advanced diagnostic tests that were ordered by specialists. These tests, however, were not included in analyses of how services received by SSU patients affected SSU success, because SSU attending physicians were only indirectly involved in whether or not patients received these tests. Treatments were considered acute care treatments if they were commonly administered only in acute care settings, such as heparin for unstable angina or intravenous furosemide for pulmonary edema.

SSU Success

The SSU was designed to care for patients during brief stays and without eventual admission to traditional inpatient services. Therefore, we used patients' LOS and whether or not patients were admitted to traditional inpatient services as measures of SSU success. LOS was calculated from the time patients arrived in the SSU until the time they left. Therefore, neither time spent in the ED before admission to the SSU nor time spent on traditional inpatient services (if needed) contributed to our definition of LOS. Individual SSU patients were considered successfully cared for in the SSU if their LOS was less than 72 hours and they were discharged directly home from the SSU. We explored associations between these outcomes and provisional diagnoses, risk assessments, and services received.

Data Analysis

LOS data were right‐skewed; therefore, we used the Mann‐Whitney test for comparisons between 2 groups and the Kruskal‐Wallis test for comparisons among 3 or more groups. To test for trends of median LOS among ordered groupings, we used the method of Cuzick.11 We used Pearson's chi‐square test to compare proportions of patients grouped into categories and the chi‐square test for trends with equal scoring to test for trends among ordered groupings.

We performed multiple logistic regression to explore which variables were associated with SSU success. The following 5 demographic variables from Table 1 were insignificant in all single‐variable and multiple‐variable regression models that we tested and were, therefore, removed from further analyses to create more parsimonious models: gender, language, ethnicity, race, and whether or not patients had a primary care provider. Our multiple logistic regression models were fitted by maximum likelihood methods. In all of these models, odds ratios (ORs) were adjusted for patient characteristics that included age (in years), insulin‐requiring diabetes mellitus (yes or no), SSU attending physician, day of the week of SSU admission (weekday or weekend), and hospitalization during the preceding year. Confidence intervals (CIs) for predicted probabilities were computed using the delta method. All analyses were conducted with Stata Statistical Software, Release 9 (StataCorp, College Station, TX).

Subjects

During the 4‐month study period, 755 patients were admitted to the SSU. Among these patients, 4 were excluded from our study because they refused verbal consent. In the remaining study sample of 751 patients, all were included in the descriptions of patients' admission characteristics (Table 1), but 13 patients who left prematurely were excluded in both the descriptions of services received (Table 2) and the analyses of SSU success (Tables 3 and 4).

Admission Characteristics and Services Received

A narrow range of provisional diagnoses were listed by ED attending physicians and 641 patients (85% of 751) were grouped as having possible ACS or heart failure (Figure 2). Patients with these diagnoses were later risk‐stratified and, when pooled across risk strata, only 14 patients (2% of 641) exceeded the suggested admission‐location criterion for the SSU of lower than high risk. Despite the array and frequency of diagnostic and treatment services that patients received, SSU physicians worked mostly independently, requesting specialty consultations for only 19% of patients (141/738; Table 2).

SSU Success

The median LOS for all patients was 42 hours (interquartile range [IRQ] 22‐63) and 156 patients (21% of 738) had unsuccessful SSU stays (Table 3). The most common reason for an unsuccessful stay was a stay longer than 72 hours (71% of 156). Among the 66 patients who required admission to traditional inpatient services, nearly one‐half (48%) were admitted expressly to receive treatments not available in the SSU after having a specialty consult.

Patients' provisional diagnoses were associated with unsuccessful stays in bivariate analyses (Table 3). In addition, when patients were grouped into 3 risk stratums (very‐low, low, and intermediate‐and‐high), unsuccessful stays increased with increasing risk. For example, in patients with possible ACS, the proportion of unsuccessful stays increased from 17% of 306 very‐low risk patients to 27% of 55 intermediate‐and‐high risk patients (P value for trend = 0.012. Similarly, in patients with heart failure, the proportion of unsuccessful stays increased from 25% of 181 very‐low risk patients to 100% of 3 intermediate‐and‐high risk patients (P value for trend = 0.004).

However, in multiple variable models that simultaneously included patients' characteristics upon admission with services received during their SSU stay, only the provisional diagnosis of heart failure was associated with unsuccessful stays (OR, 1.9; 95% CI, 1.12‐3.18); risk assessments for possible ACS (P = 0.29) and heart failure (P = 0.32) were unimportant predictors of unsuccessful stays (Table 4). On the other hand, whether or not patients received diagnostic tests, acute care treatments, or specialty consultations were important predictors. In particular, patients who received specialty consultations were much more likely to require admission to traditional inpatient services than those patients who did not (OR, 13.1; 95% CI, 6.9‐24.9) and had a 52% chance of having an unsuccessful stay (95% CI, 42‐61%; Figure 3). In addition, the accessibility of a diagnostic test was inversely proportional to the chance of having a long stay; patients who received an open access test had a 12% chance of a long stay (95% CI, 8‐16%) whereas those who received a limited access test had a 29% chance of a long stay (95% CI, 20‐39%). Receiving acute care treatments was also a significant, though less important, predictor of an unsuccessful stay (Table 4).

Figure 3

Design and Setting

In this prospective cohort study, we included all patients admitted to the hospitalist‐run SSU of Cook County Hospital, a 500‐bed public teaching hospital in Chicago, Illinois, from January through April of 2006. Our 14‐bed SSU opened in 2002 to reduce overcrowding on the traditional inpatient wards by admitting adult patients who require inpatient care but might be eligible for discharge within 3 days. The unit is geographically part of the ED but is staffed by resident physicians and a rotating group of hospitalist attending physicians from the Department of Medicine. At least 1 attending and resident physician are available throughout the day, including weekend days and holidays; evenings are covered by a resident who presents overnight admissions to an attending physician the following morning.

ED physicians admit general medical patients to the SSU 24 hours per day, 7 days per week. Though admissions do not require prior approval from SSU physicians, the Departments of Medicine and Emergency Medicine have collaboratively promoted 5 suggested admission‐location guidelines to admitting ED physicians (Figure 1). For candidate SSU patients, these 5 guidelines are not intended to be restrictive but to provide a framework for the complex decision‐making process that our ED physicians encounter, particularly during periods of extreme overcrowding.7 First, patients should have an anticipated stay shorter than 72 hours. Second, patients should not have an eventual need for admission to traditional inpatient services such as the general medicine wards or intensive care units; this guideline is intended to improve patient safety by reducing unnecessary handoffs between physicians.8 Third, patients with provisional cardiovascular diagnoses should be preferentially admitted to the SSU over the general medical wards; this guideline is intended to improve hospital‐wide efficiency because our SSU is equipped with continuous telemetry monitors, an exercise treadmill testing (ETT) laboratory, and other reserved cardiac tests (see Admission Characteristics and Services Received section, below). Fourth, patients' risk should be no higher than intermediate level. Admitting ED physicians are encouraged to use posted risk estimators for patients with provisional diagnoses of possible acute coronary syndrome (ACS), decompensated heart failure, asthma exacerbation, and out‐of‐control diabetes. Finally, patients should not need advanced ancillary services; these include bedside procedures (eg, central venous catheter insertions), time‐intensive nursing (eg, regular dressing changes), and complex social‐services (eg, long‐term care facility placements).

Figure 1

Subjects

The study subjects were all patients admitted to the SSU during the 4‐month study period. Patients were excluded from the entire study if they refused verbal consent to participate. All patients who consented were included in the description of patient admission characteristics. Thirteen patients who prematurely left the SSU against medical advice, however, were neither included in the descriptions of services received nor in the analyses of predictors of successful SSU stays. We excluded these patients because they needed services that they did not receiveincluding these patients in our analysis would tend to overestimate the efficiency of our SSU by shortening the length‐of‐stay (LOS) without adding diagnostic tests or treatments.

Data Collection

After receiving approval from the institutional review board, attending physician investigators conducted an interview, physical examination, and review of medical records for each enrolled patient within 12 hours of admission to the SSU. When ED attending physicians' provisional primary diagnoses included possible ACS or decompensated heart failure, which we knew from earlier pilot data were our 2 most common provisional diagnoses, investigators gathered patient data to be applied in validated models of risk after the study period (Figure 2).9, 10 Some of the clinical predictors required for these models are based on patients' findings on presentation to the ED. For example, Goldman's risk model for major cardiac events uses patients' initial systolic blood pressures on presentation to the ED.9 In such cases, investigators gathered needed data from electronic and paper charts generated in the ED. Upon discharge from the SSU, investigators reviewed patients' medical records a second time. All data were entered by investigators and instantly committed into an online database.

Figure 2

Admission Characteristics and Services Received

Patients were grouped according to the provisional diagnoses of ED attending physicians upon admission to the SSU (Figure 2). We chose to group patients by the provisional diagnoses of EDnot SSUattending physicians to better understand how ED physicians, the physicians who make the admission‐location decisions in our hospital, were using the SSU. Patients were first grouped as having possible ACS or heart failure, because patients with these provisional diagnoses were preferentially admitted to the SSU (Figure 1). When neither diagnosis was listed, patients were grouped according to ED attending physicians' first‐listed diagnoses. At the end of the study period, relevant risk models were applied to patients with possible ACS or heart failure and stratified as very low, low, intermediate, or high risk.9, 10 Patients with both possible ACS and heart failure were grouped according to the diagnosis with the highest corresponding risk assessment; if both risk assessments were the same, then the first‐listed diagnosis was used. Though developed to predict different clinical outcomes during different time periods, risk strata from the corresponding risk models were pooled across both diagnoses to develop a risk summary.

Upon discharge, investigators recorded which advanced diagnostic tests, specialty consultations, and acute care treatments patients received while in the SSU. Diagnostic tests were considered advanced if they were not routinely performed within 2 hours of being ordered. Advanced diagnostic tests were grouped into 2 types by their accessibility to ordering SSU physicians. Open access tests included echocardiograms and ETTs, which were reserved for SSU patients 6 days per week. Though the availability of open access tests was not unlimited, ordering physicians' needs for them rarely exceeded the immediate supply. On the other hand, limited access tests included both cardiac stress imaging studies, which were reserved for SSU patients on a very limited basis 4 or 5 days per week, and other tests that were not reserved for SSU patients, such as endoscopy, magnetic resonance imaging, or ultrasonography. Ordering physicians' needs for limited access tests often exceeded their immediate supply; in such cases, SSU patients were placed without priority into queues that included patients from the entire hospital.

Investigators recorded when patients received advanced diagnostic tests that were ordered by specialists. These tests, however, were not included in analyses of how services received by SSU patients affected SSU success, because SSU attending physicians were only indirectly involved in whether or not patients received these tests. Treatments were considered acute care treatments if they were commonly administered only in acute care settings, such as heparin for unstable angina or intravenous furosemide for pulmonary edema.

SSU Success

The SSU was designed to care for patients during brief stays and without eventual admission to traditional inpatient services. Therefore, we used patients' LOS and whether or not patients were admitted to traditional inpatient services as measures of SSU success. LOS was calculated from the time patients arrived in the SSU until the time they left. Therefore, neither time spent in the ED before admission to the SSU nor time spent on traditional inpatient services (if needed) contributed to our definition of LOS. Individual SSU patients were considered successfully cared for in the SSU if their LOS was less than 72 hours and they were discharged directly home from the SSU. We explored associations between these outcomes and provisional diagnoses, risk assessments, and services received.

Data Analysis

LOS data were right‐skewed; therefore, we used the Mann‐Whitney test for comparisons between 2 groups and the Kruskal‐Wallis test for comparisons among 3 or more groups. To test for trends of median LOS among ordered groupings, we used the method of Cuzick.11 We used Pearson's chi‐square test to compare proportions of patients grouped into categories and the chi‐square test for trends with equal scoring to test for trends among ordered groupings.

We performed multiple logistic regression to explore which variables were associated with SSU success. The following 5 demographic variables from Table 1 were insignificant in all single‐variable and multiple‐variable regression models that we tested and were, therefore, removed from further analyses to create more parsimonious models: gender, language, ethnicity, race, and whether or not patients had a primary care provider. Our multiple logistic regression models were fitted by maximum likelihood methods. In all of these models, odds ratios (ORs) were adjusted for patient characteristics that included age (in years), insulin‐requiring diabetes mellitus (yes or no), SSU attending physician, day of the week of SSU admission (weekday or weekend), and hospitalization during the preceding year. Confidence intervals (CIs) for predicted probabilities were computed using the delta method. All analyses were conducted with Stata Statistical Software, Release 9 (StataCorp, College Station, TX).

Subjects

During the 4‐month study period, 755 patients were admitted to the SSU. Among these patients, 4 were excluded from our study because they refused verbal consent. In the remaining study sample of 751 patients, all were included in the descriptions of patients' admission characteristics (Table 1), but 13 patients who left prematurely were excluded in both the descriptions of services received (Table 2) and the analyses of SSU success (Tables 3 and 4).

Admission Characteristics and Services Received

A narrow range of provisional diagnoses were listed by ED attending physicians and 641 patients (85% of 751) were grouped as having possible ACS or heart failure (Figure 2). Patients with these diagnoses were later risk‐stratified and, when pooled across risk strata, only 14 patients (2% of 641) exceeded the suggested admission‐location criterion for the SSU of lower than high risk. Despite the array and frequency of diagnostic and treatment services that patients received, SSU physicians worked mostly independently, requesting specialty consultations for only 19% of patients (141/738; Table 2).

SSU Success

The median LOS for all patients was 42 hours (interquartile range [IRQ] 22‐63) and 156 patients (21% of 738) had unsuccessful SSU stays (Table 3). The most common reason for an unsuccessful stay was a stay longer than 72 hours (71% of 156). Among the 66 patients who required admission to traditional inpatient services, nearly one‐half (48%) were admitted expressly to receive treatments not available in the SSU after having a specialty consult.

Patients' provisional diagnoses were associated with unsuccessful stays in bivariate analyses (Table 3). In addition, when patients were grouped into 3 risk stratums (very‐low, low, and intermediate‐and‐high), unsuccessful stays increased with increasing risk. For example, in patients with possible ACS, the proportion of unsuccessful stays increased from 17% of 306 very‐low risk patients to 27% of 55 intermediate‐and‐high risk patients (P value for trend = 0.012. Similarly, in patients with heart failure, the proportion of unsuccessful stays increased from 25% of 181 very‐low risk patients to 100% of 3 intermediate‐and‐high risk patients (P value for trend = 0.004).

However, in multiple variable models that simultaneously included patients' characteristics upon admission with services received during their SSU stay, only the provisional diagnosis of heart failure was associated with unsuccessful stays (OR, 1.9; 95% CI, 1.12‐3.18); risk assessments for possible ACS (P = 0.29) and heart failure (P = 0.32) were unimportant predictors of unsuccessful stays (Table 4). On the other hand, whether or not patients received diagnostic tests, acute care treatments, or specialty consultations were important predictors. In particular, patients who received specialty consultations were much more likely to require admission to traditional inpatient services than those patients who did not (OR, 13.1; 95% CI, 6.9‐24.9) and had a 52% chance of having an unsuccessful stay (95% CI, 42‐61%; Figure 3). In addition, the accessibility of a diagnostic test was inversely proportional to the chance of having a long stay; patients who received an open access test had a 12% chance of a long stay (95% CI, 8‐16%) whereas those who received a limited access test had a 29% chance of a long stay (95% CI, 20‐39%). Receiving acute care treatments was also a significant, though less important, predictor of an unsuccessful stay (Table 4).

Figure 3

Design and Setting

In this prospective cohort study, we included all patients admitted to the hospitalist‐run SSU of Cook County Hospital, a 500‐bed public teaching hospital in Chicago, Illinois, from January through April of 2006. Our 14‐bed SSU opened in 2002 to reduce overcrowding on the traditional inpatient wards by admitting adult patients who require inpatient care but might be eligible for discharge within 3 days. The unit is geographically part of the ED but is staffed by resident physicians and a rotating group of hospitalist attending physicians from the Department of Medicine. At least 1 attending and resident physician are available throughout the day, including weekend days and holidays; evenings are covered by a resident who presents overnight admissions to an attending physician the following morning.

ED physicians admit general medical patients to the SSU 24 hours per day, 7 days per week. Though admissions do not require prior approval from SSU physicians, the Departments of Medicine and Emergency Medicine have collaboratively promoted 5 suggested admission‐location guidelines to admitting ED physicians (Figure 1). For candidate SSU patients, these 5 guidelines are not intended to be restrictive but to provide a framework for the complex decision‐making process that our ED physicians encounter, particularly during periods of extreme overcrowding.7 First, patients should have an anticipated stay shorter than 72 hours. Second, patients should not have an eventual need for admission to traditional inpatient services such as the general medicine wards or intensive care units; this guideline is intended to improve patient safety by reducing unnecessary handoffs between physicians.8 Third, patients with provisional cardiovascular diagnoses should be preferentially admitted to the SSU over the general medical wards; this guideline is intended to improve hospital‐wide efficiency because our SSU is equipped with continuous telemetry monitors, an exercise treadmill testing (ETT) laboratory, and other reserved cardiac tests (see Admission Characteristics and Services Received section, below). Fourth, patients' risk should be no higher than intermediate level. Admitting ED physicians are encouraged to use posted risk estimators for patients with provisional diagnoses of possible acute coronary syndrome (ACS), decompensated heart failure, asthma exacerbation, and out‐of‐control diabetes. Finally, patients should not need advanced ancillary services; these include bedside procedures (eg, central venous catheter insertions), time‐intensive nursing (eg, regular dressing changes), and complex social‐services (eg, long‐term care facility placements).

Figure 1

Subjects

The study subjects were all patients admitted to the SSU during the 4‐month study period. Patients were excluded from the entire study if they refused verbal consent to participate. All patients who consented were included in the description of patient admission characteristics. Thirteen patients who prematurely left the SSU against medical advice, however, were neither included in the descriptions of services received nor in the analyses of predictors of successful SSU stays. We excluded these patients because they needed services that they did not receiveincluding these patients in our analysis would tend to overestimate the efficiency of our SSU by shortening the length‐of‐stay (LOS) without adding diagnostic tests or treatments.

Data Collection

After receiving approval from the institutional review board, attending physician investigators conducted an interview, physical examination, and review of medical records for each enrolled patient within 12 hours of admission to the SSU. When ED attending physicians' provisional primary diagnoses included possible ACS or decompensated heart failure, which we knew from earlier pilot data were our 2 most common provisional diagnoses, investigators gathered patient data to be applied in validated models of risk after the study period (Figure 2).9, 10 Some of the clinical predictors required for these models are based on patients' findings on presentation to the ED. For example, Goldman's risk model for major cardiac events uses patients' initial systolic blood pressures on presentation to the ED.9 In such cases, investigators gathered needed data from electronic and paper charts generated in the ED. Upon discharge from the SSU, investigators reviewed patients' medical records a second time. All data were entered by investigators and instantly committed into an online database.

Figure 2

Admission Characteristics and Services Received

Patients were grouped according to the provisional diagnoses of ED attending physicians upon admission to the SSU (Figure 2). We chose to group patients by the provisional diagnoses of EDnot SSUattending physicians to better understand how ED physicians, the physicians who make the admission‐location decisions in our hospital, were using the SSU. Patients were first grouped as having possible ACS or heart failure, because patients with these provisional diagnoses were preferentially admitted to the SSU (Figure 1). When neither diagnosis was listed, patients were grouped according to ED attending physicians' first‐listed diagnoses. At the end of the study period, relevant risk models were applied to patients with possible ACS or heart failure and stratified as very low, low, intermediate, or high risk.9, 10 Patients with both possible ACS and heart failure were grouped according to the diagnosis with the highest corresponding risk assessment; if both risk assessments were the same, then the first‐listed diagnosis was used. Though developed to predict different clinical outcomes during different time periods, risk strata from the corresponding risk models were pooled across both diagnoses to develop a risk summary.

Upon discharge, investigators recorded which advanced diagnostic tests, specialty consultations, and acute care treatments patients received while in the SSU. Diagnostic tests were considered advanced if they were not routinely performed within 2 hours of being ordered. Advanced diagnostic tests were grouped into 2 types by their accessibility to ordering SSU physicians. Open access tests included echocardiograms and ETTs, which were reserved for SSU patients 6 days per week. Though the availability of open access tests was not unlimited, ordering physicians' needs for them rarely exceeded the immediate supply. On the other hand, limited access tests included both cardiac stress imaging studies, which were reserved for SSU patients on a very limited basis 4 or 5 days per week, and other tests that were not reserved for SSU patients, such as endoscopy, magnetic resonance imaging, or ultrasonography. Ordering physicians' needs for limited access tests often exceeded their immediate supply; in such cases, SSU patients were placed without priority into queues that included patients from the entire hospital.

Investigators recorded when patients received advanced diagnostic tests that were ordered by specialists. These tests, however, were not included in analyses of how services received by SSU patients affected SSU success, because SSU attending physicians were only indirectly involved in whether or not patients received these tests. Treatments were considered acute care treatments if they were commonly administered only in acute care settings, such as heparin for unstable angina or intravenous furosemide for pulmonary edema.

SSU Success

The SSU was designed to care for patients during brief stays and without eventual admission to traditional inpatient services. Therefore, we used patients' LOS and whether or not patients were admitted to traditional inpatient services as measures of SSU success. LOS was calculated from the time patients arrived in the SSU until the time they left. Therefore, neither time spent in the ED before admission to the SSU nor time spent on traditional inpatient services (if needed) contributed to our definition of LOS. Individual SSU patients were considered successfully cared for in the SSU if their LOS was less than 72 hours and they were discharged directly home from the SSU. We explored associations between these outcomes and provisional diagnoses, risk assessments, and services received.

Data Analysis

LOS data were right‐skewed; therefore, we used the Mann‐Whitney test for comparisons between 2 groups and the Kruskal‐Wallis test for comparisons among 3 or more groups. To test for trends of median LOS among ordered groupings, we used the method of Cuzick.11 We used Pearson's chi‐square test to compare proportions of patients grouped into categories and the chi‐square test for trends with equal scoring to test for trends among ordered groupings.

We performed multiple logistic regression to explore which variables were associated with SSU success. The following 5 demographic variables from Table 1 were insignificant in all single‐variable and multiple‐variable regression models that we tested and were, therefore, removed from further analyses to create more parsimonious models: gender, language, ethnicity, race, and whether or not patients had a primary care provider. Our multiple logistic regression models were fitted by maximum likelihood methods. In all of these models, odds ratios (ORs) were adjusted for patient characteristics that included age (in years), insulin‐requiring diabetes mellitus (yes or no), SSU attending physician, day of the week of SSU admission (weekday or weekend), and hospitalization during the preceding year. Confidence intervals (CIs) for predicted probabilities were computed using the delta method. All analyses were conducted with Stata Statistical Software, Release 9 (StataCorp, College Station, TX).

Subjects

During the 4‐month study period, 755 patients were admitted to the SSU. Among these patients, 4 were excluded from our study because they refused verbal consent. In the remaining study sample of 751 patients, all were included in the descriptions of patients' admission characteristics (Table 1), but 13 patients who left prematurely were excluded in both the descriptions of services received (Table 2) and the analyses of SSU success (Tables 3 and 4).

Admission Characteristics and Services Received

A narrow range of provisional diagnoses were listed by ED attending physicians and 641 patients (85% of 751) were grouped as having possible ACS or heart failure (Figure 2). Patients with these diagnoses were later risk‐stratified and, when pooled across risk strata, only 14 patients (2% of 641) exceeded the suggested admission‐location criterion for the SSU of lower than high risk. Despite the array and frequency of diagnostic and treatment services that patients received, SSU physicians worked mostly independently, requesting specialty consultations for only 19% of patients (141/738; Table 2).

SSU Success

The median LOS for all patients was 42 hours (interquartile range [IRQ] 22‐63) and 156 patients (21% of 738) had unsuccessful SSU stays (Table 3). The most common reason for an unsuccessful stay was a stay longer than 72 hours (71% of 156). Among the 66 patients who required admission to traditional inpatient services, nearly one‐half (48%) were admitted expressly to receive treatments not available in the SSU after having a specialty consult.

Patients' provisional diagnoses were associated with unsuccessful stays in bivariate analyses (Table 3). In addition, when patients were grouped into 3 risk stratums (very‐low, low, and intermediate‐and‐high), unsuccessful stays increased with increasing risk. For example, in patients with possible ACS, the proportion of unsuccessful stays increased from 17% of 306 very‐low risk patients to 27% of 55 intermediate‐and‐high risk patients (P value for trend = 0.012. Similarly, in patients with heart failure, the proportion of unsuccessful stays increased from 25% of 181 very‐low risk patients to 100% of 3 intermediate‐and‐high risk patients (P value for trend = 0.004).

However, in multiple variable models that simultaneously included patients' characteristics upon admission with services received during their SSU stay, only the provisional diagnosis of heart failure was associated with unsuccessful stays (OR, 1.9; 95% CI, 1.12‐3.18); risk assessments for possible ACS (P = 0.29) and heart failure (P = 0.32) were unimportant predictors of unsuccessful stays (Table 4). On the other hand, whether or not patients received diagnostic tests, acute care treatments, or specialty consultations were important predictors. In particular, patients who received specialty consultations were much more likely to require admission to traditional inpatient services than those patients who did not (OR, 13.1; 95% CI, 6.9‐24.9) and had a 52% chance of having an unsuccessful stay (95% CI, 42‐61%; Figure 3). In addition, the accessibility of a diagnostic test was inversely proportional to the chance of having a long stay; patients who received an open access test had a 12% chance of a long stay (95% CI, 8‐16%) whereas those who received a limited access test had a 29% chance of a long stay (95% CI, 20‐39%). Receiving acute care treatments was also a significant, though less important, predictor of an unsuccessful stay (Table 4).

Figure 3

Design and Setting

In this prospective cohort study, the cohort was all patients admitted to the general medicine service at Cook County Hospital, a 500‐bed public teaching hospital in Chicago, Illinois, in January and February of 2006. The general medicine inpatient service is divided into 3 firms (A, B, and C), each with 4 separate teams of physicians and students. Admissions from the emergency department or other services in the hospital, such as intensive care units (which are closed and therefore staffed by separate teams of physicians), are distributed in sequence to on‐call teams from each firm. During the study period, the availability of a bedside procedure service varied by firm. Throughout the first 4 weeks, the service was available to only 1 of 3 firms (firm A). Then, during weeks 5 through 8, the service crossed over to the other 2 firms (firms B and C) and was unavailable to the original firm. Firm assignments for residents assigned to the inpatient service for all 8 weeks did not change. Of the 16 residents assigned to firm A during the first 4 weeks, when the procedure service was available, 3 remained on the wards during the second 4 weeks, when the procedure service was not available.

We chose to collect data on 4 bedside procedures: paracentesis, thoracentesis, LP, and CVC. Similar to those at other teaching hospitals, our residents informally acquire the skills to perform these procedures while assisting and being assisted by more experienced senior residents in a see one, do one, teach one apprenticeship model of learning.4 To improve the training and performance of these bedside procedures, the Department of Medicine piloted a bedside procedure service to teach procedural skills and assist residents during these procedures. Use of the service, though voluntary, was actively encouraged at residents' monthly orientation meetings and regular conferences.

One attending inpatient physician (J.A.) staffed the bedside procedure service, which was available during normal work hours on weekdays. Requests for procedures were made by general medicine residents through an online database and, after approval by the procedure service attending physician, were performed under his direct supervision. A hand‐carried ultrasound (MicroMaxx, Sonosite, Inc., Bothell, WA) that generates a 2‐dimensional gray‐scale image was used to both confirm the presence and location of fluid prior to paracentesis and thoracentesis and provide real‐time guidance during central venous catheterization. When the bedside procedure service was unavailable, residents performed bedside procedures in the usual fashion, typically without direct attending physician supervision. But if requested, an on‐call chief medical resident with access to a hand‐carried ultrasound device used by the intensive care unit was available for assistance at any time.

Subjects

The study subjects were all patients admitted to the general medical service during the 8‐week pilot period. Patients were excluded if they had been discharged before arrival on the medical wards or if they were under the care of the general medicine service for less than 6 hours before discharge or transfer to another service. We chose 6 hours because we reasoned that such brief admissions were not potential candidates for invasive bedside procedures.

Data Collection

Each morning an investigator contacted the senior residents who had admitted patients during the previous 24‐hour shift and confirmed that newly admitted patients were under the care of the general medicine service for more than 6 hours. To examine how the number of attempts may have been affected by procedures done in the emergency room or intensive care units before admission to the general medicine service, investigators also asked admitting residents whether a bedside procedure had been attempted in the 72 hours before admission. Every general medicine service resident was asked to fill out a brief data collection form after an attempt to perform any procedure on the general medical wards. In addition, chief residents asked each member of the general medicine service at mandatory sign‐out rounds at the end of each weekday whether any procedures had been attempted, and on weekend days investigators contacted senior residents from each general medicine service team.

We report on this quality assurance study, which was conducted during a pilot phase. This report has been reviewed and judged exempt by our institutional review board.

Primary OutcomeNumber of Procedure Attempts

For all bedside procedures attempted by residents on the general medical wards, investigators determined whether the residents were members of firms that were offered the bedside procedure service and, if so, whether the procedure service attending directly supervised the procedure attempt. Multiple procedure attempts of the same type were counted for an individual patient if (1) the procedure attempts did not occur during the same admissions and (2) neither the physicians attempting the procedure nor the primary indications for it were the same. Therefore, neither attempts performed after initially unsuccessful ones nor repeated procedures, such as large‐volume therapeutic paracentesis and thoracentesis, were counted twice. We reasoned that when these criteria were met, procedure attempts could be considered independently.

Secondary Outcomes

Investigators asked residents who attempted procedures to indicate whether (1) the indication for the procedure was solely diagnostic or was, at least in part, therapeutic; (2) the procedure was successful; and (3) there were any immediate major periprocedural complications. A procedure was considered to have been successfully performed if it fulfilled 2 criteria: it had to be completed during a single continuous attempt, even if multiple sites or procedure kits were used; and it had to fulfill the indication for it being done. For example, if the indication for thoracentesis was therapeutic, this procedure would be considered successful if it yielded a large enough volume of fluid to alleviate the patient's symptoms, but if the indication was diagnostic, then thoracentesis would be considered successful if it yielded enough fluid for laboratory processing. Residents were asked to report any periprocedural complications that they considered major; 2 illustrative examples were provided: a pneumothorax and severe bleeding.

Data Analyses

On the basis of earlier pilot data, we estimated that 8%10% of all admissions to the general medicine service underwent at least 1 procedure (paracentesis, thoracentesis, lumbar puncture, or central vein catheterization). We planned for a sample size of 1900 admissions, which would have 80% power to detect a clinically meaningful 50% relative increase in the mean number of bedside procedures with a double‐sided alpha error of 0.05. We used permutation tests to compare the mean number of procedures attempted between firms and bootstrap simulation to construct 95% confidence intervals for those means and the differences between and ratios of them. Fisher's exact test was used to compare proportions of successfully performed procedures and preadmission procedure attempts. All analyses were conducted with Stata Statistical Software, Release 9 (StataCorp, LP, College Station, TX).

Subjects

During this 8‐week pilot study, there were 2157 admissions to the general medicine service. Among these admissions, 216 were excluded from our study because the patients did not arrive on the medical wards or were not under the care of the general medicine service for at least 6 hours before discharge or before being transferred to another service. Of the remaining 1941 admissions, 935 were to firms with the bedside procedure service available, and 1006 were to firms without the service available (Fig. 1)

Figure 1

Primary OutcomeNumber of Procedure Attempts

Overall, 122 patients underwent 145 procedure attempts that met our criteria for independence. The mean number of procedure attempts in firms offered the bedside procedure service was 48% higher (90 versus 61 per 1000 admissions; RR 1.48, 95% CI 1.062.10; P = .030; Fig. 1). When procedures attempted on weekends and holidays were excluded, the relative increase in procedure attempts in firms offered the bedside procedure service was even higher (70 versus 43 per 1000 admissions; RR 1.63, 95% CI 1.092.49; P = .023; Fig. 1). When grouped according to whether procedure attempts occurred before or after crossover of the procedure service, the mean number of procedure attempts in firms was higher when the service was offered: firm A dropped from 84 to 70 per 1000 admissions (P = .58) after losing the service, whereas firms B and C increased from 57 to 94 per 1000 admissions (P = .025) on gaining the service. There were 40 procedure attempts performed on patients within 72 hours before admission, but there was no difference between firms in the proportions of these preadmission procedures (P = .43).

Secondary Outcomes

Table 1 shows how of each type of procedure contributed to the overall difference. Attempts of CVC and therapeutic paracentesis and thoracentesis accounted for 86% of the overall increase in procedure attempts for admissions to firms offered the bedside procedure service, whereas only 14% of this increase was a result of diagnostic procedures. There were no differences in the proportions of successfully performed procedures, whether grouped by firm (P = 1.0) or by direct supervision from the procedure service attending (P = .64; Table 2). There were 3 self‐reported major periprocedural complications; all were related to excessive bleeding from CVC attempts. Two occurred without direct supervision from the bedside procedure service attending, one hemomediastinum from an internal jugular CVC attempt and one groin hematoma from a femoral CVC attempt. The third, a groin hematoma from a femoral CVC attempt, occurred during direct supervision from the bedside procedure service attending.

Design and Setting

In this prospective cohort study, the cohort was all patients admitted to the general medicine service at Cook County Hospital, a 500‐bed public teaching hospital in Chicago, Illinois, in January and February of 2006. The general medicine inpatient service is divided into 3 firms (A, B, and C), each with 4 separate teams of physicians and students. Admissions from the emergency department or other services in the hospital, such as intensive care units (which are closed and therefore staffed by separate teams of physicians), are distributed in sequence to on‐call teams from each firm. During the study period, the availability of a bedside procedure service varied by firm. Throughout the first 4 weeks, the service was available to only 1 of 3 firms (firm A). Then, during weeks 5 through 8, the service crossed over to the other 2 firms (firms B and C) and was unavailable to the original firm. Firm assignments for residents assigned to the inpatient service for all 8 weeks did not change. Of the 16 residents assigned to firm A during the first 4 weeks, when the procedure service was available, 3 remained on the wards during the second 4 weeks, when the procedure service was not available.

We chose to collect data on 4 bedside procedures: paracentesis, thoracentesis, LP, and CVC. Similar to those at other teaching hospitals, our residents informally acquire the skills to perform these procedures while assisting and being assisted by more experienced senior residents in a see one, do one, teach one apprenticeship model of learning.4 To improve the training and performance of these bedside procedures, the Department of Medicine piloted a bedside procedure service to teach procedural skills and assist residents during these procedures. Use of the service, though voluntary, was actively encouraged at residents' monthly orientation meetings and regular conferences.

One attending inpatient physician (J.A.) staffed the bedside procedure service, which was available during normal work hours on weekdays. Requests for procedures were made by general medicine residents through an online database and, after approval by the procedure service attending physician, were performed under his direct supervision. A hand‐carried ultrasound (MicroMaxx, Sonosite, Inc., Bothell, WA) that generates a 2‐dimensional gray‐scale image was used to both confirm the presence and location of fluid prior to paracentesis and thoracentesis and provide real‐time guidance during central venous catheterization. When the bedside procedure service was unavailable, residents performed bedside procedures in the usual fashion, typically without direct attending physician supervision. But if requested, an on‐call chief medical resident with access to a hand‐carried ultrasound device used by the intensive care unit was available for assistance at any time.

Subjects

The study subjects were all patients admitted to the general medical service during the 8‐week pilot period. Patients were excluded if they had been discharged before arrival on the medical wards or if they were under the care of the general medicine service for less than 6 hours before discharge or transfer to another service. We chose 6 hours because we reasoned that such brief admissions were not potential candidates for invasive bedside procedures.

Data Collection

Each morning an investigator contacted the senior residents who had admitted patients during the previous 24‐hour shift and confirmed that newly admitted patients were under the care of the general medicine service for more than 6 hours. To examine how the number of attempts may have been affected by procedures done in the emergency room or intensive care units before admission to the general medicine service, investigators also asked admitting residents whether a bedside procedure had been attempted in the 72 hours before admission. Every general medicine service resident was asked to fill out a brief data collection form after an attempt to perform any procedure on the general medical wards. In addition, chief residents asked each member of the general medicine service at mandatory sign‐out rounds at the end of each weekday whether any procedures had been attempted, and on weekend days investigators contacted senior residents from each general medicine service team.

We report on this quality assurance study, which was conducted during a pilot phase. This report has been reviewed and judged exempt by our institutional review board.

Primary OutcomeNumber of Procedure Attempts

For all bedside procedures attempted by residents on the general medical wards, investigators determined whether the residents were members of firms that were offered the bedside procedure service and, if so, whether the procedure service attending directly supervised the procedure attempt. Multiple procedure attempts of the same type were counted for an individual patient if (1) the procedure attempts did not occur during the same admissions and (2) neither the physicians attempting the procedure nor the primary indications for it were the same. Therefore, neither attempts performed after initially unsuccessful ones nor repeated procedures, such as large‐volume therapeutic paracentesis and thoracentesis, were counted twice. We reasoned that when these criteria were met, procedure attempts could be considered independently.

Secondary Outcomes

Investigators asked residents who attempted procedures to indicate whether (1) the indication for the procedure was solely diagnostic or was, at least in part, therapeutic; (2) the procedure was successful; and (3) there were any immediate major periprocedural complications. A procedure was considered to have been successfully performed if it fulfilled 2 criteria: it had to be completed during a single continuous attempt, even if multiple sites or procedure kits were used; and it had to fulfill the indication for it being done. For example, if the indication for thoracentesis was therapeutic, this procedure would be considered successful if it yielded a large enough volume of fluid to alleviate the patient's symptoms, but if the indication was diagnostic, then thoracentesis would be considered successful if it yielded enough fluid for laboratory processing. Residents were asked to report any periprocedural complications that they considered major; 2 illustrative examples were provided: a pneumothorax and severe bleeding.

Data Analyses

On the basis of earlier pilot data, we estimated that 8%10% of all admissions to the general medicine service underwent at least 1 procedure (paracentesis, thoracentesis, lumbar puncture, or central vein catheterization). We planned for a sample size of 1900 admissions, which would have 80% power to detect a clinically meaningful 50% relative increase in the mean number of bedside procedures with a double‐sided alpha error of 0.05. We used permutation tests to compare the mean number of procedures attempted between firms and bootstrap simulation to construct 95% confidence intervals for those means and the differences between and ratios of them. Fisher's exact test was used to compare proportions of successfully performed procedures and preadmission procedure attempts. All analyses were conducted with Stata Statistical Software, Release 9 (StataCorp, LP, College Station, TX).

Subjects

During this 8‐week pilot study, there were 2157 admissions to the general medicine service. Among these admissions, 216 were excluded from our study because the patients did not arrive on the medical wards or were not under the care of the general medicine service for at least 6 hours before discharge or before being transferred to another service. Of the remaining 1941 admissions, 935 were to firms with the bedside procedure service available, and 1006 were to firms without the service available (Fig. 1)

Figure 1

Primary OutcomeNumber of Procedure Attempts

Overall, 122 patients underwent 145 procedure attempts that met our criteria for independence. The mean number of procedure attempts in firms offered the bedside procedure service was 48% higher (90 versus 61 per 1000 admissions; RR 1.48, 95% CI 1.062.10; P = .030; Fig. 1). When procedures attempted on weekends and holidays were excluded, the relative increase in procedure attempts in firms offered the bedside procedure service was even higher (70 versus 43 per 1000 admissions; RR 1.63, 95% CI 1.092.49; P = .023; Fig. 1). When grouped according to whether procedure attempts occurred before or after crossover of the procedure service, the mean number of procedure attempts in firms was higher when the service was offered: firm A dropped from 84 to 70 per 1000 admissions (P = .58) after losing the service, whereas firms B and C increased from 57 to 94 per 1000 admissions (P = .025) on gaining the service. There were 40 procedure attempts performed on patients within 72 hours before admission, but there was no difference between firms in the proportions of these preadmission procedures (P = .43).

Secondary Outcomes

Table 1 shows how of each type of procedure contributed to the overall difference. Attempts of CVC and therapeutic paracentesis and thoracentesis accounted for 86% of the overall increase in procedure attempts for admissions to firms offered the bedside procedure service, whereas only 14% of this increase was a result of diagnostic procedures. There were no differences in the proportions of successfully performed procedures, whether grouped by firm (P = 1.0) or by direct supervision from the procedure service attending (P = .64; Table 2). There were 3 self‐reported major periprocedural complications; all were related to excessive bleeding from CVC attempts. Two occurred without direct supervision from the bedside procedure service attending, one hemomediastinum from an internal jugular CVC attempt and one groin hematoma from a femoral CVC attempt. The third, a groin hematoma from a femoral CVC attempt, occurred during direct supervision from the bedside procedure service attending.

Design and Setting

In this prospective cohort study, the cohort was all patients admitted to the general medicine service at Cook County Hospital, a 500‐bed public teaching hospital in Chicago, Illinois, in January and February of 2006. The general medicine inpatient service is divided into 3 firms (A, B, and C), each with 4 separate teams of physicians and students. Admissions from the emergency department or other services in the hospital, such as intensive care units (which are closed and therefore staffed by separate teams of physicians), are distributed in sequence to on‐call teams from each firm. During the study period, the availability of a bedside procedure service varied by firm. Throughout the first 4 weeks, the service was available to only 1 of 3 firms (firm A). Then, during weeks 5 through 8, the service crossed over to the other 2 firms (firms B and C) and was unavailable to the original firm. Firm assignments for residents assigned to the inpatient service for all 8 weeks did not change. Of the 16 residents assigned to firm A during the first 4 weeks, when the procedure service was available, 3 remained on the wards during the second 4 weeks, when the procedure service was not available.

We chose to collect data on 4 bedside procedures: paracentesis, thoracentesis, LP, and CVC. Similar to those at other teaching hospitals, our residents informally acquire the skills to perform these procedures while assisting and being assisted by more experienced senior residents in a see one, do one, teach one apprenticeship model of learning.4 To improve the training and performance of these bedside procedures, the Department of Medicine piloted a bedside procedure service to teach procedural skills and assist residents during these procedures. Use of the service, though voluntary, was actively encouraged at residents' monthly orientation meetings and regular conferences.

One attending inpatient physician (J.A.) staffed the bedside procedure service, which was available during normal work hours on weekdays. Requests for procedures were made by general medicine residents through an online database and, after approval by the procedure service attending physician, were performed under his direct supervision. A hand‐carried ultrasound (MicroMaxx, Sonosite, Inc., Bothell, WA) that generates a 2‐dimensional gray‐scale image was used to both confirm the presence and location of fluid prior to paracentesis and thoracentesis and provide real‐time guidance during central venous catheterization. When the bedside procedure service was unavailable, residents performed bedside procedures in the usual fashion, typically without direct attending physician supervision. But if requested, an on‐call chief medical resident with access to a hand‐carried ultrasound device used by the intensive care unit was available for assistance at any time.

Subjects

The study subjects were all patients admitted to the general medical service during the 8‐week pilot period. Patients were excluded if they had been discharged before arrival on the medical wards or if they were under the care of the general medicine service for less than 6 hours before discharge or transfer to another service. We chose 6 hours because we reasoned that such brief admissions were not potential candidates for invasive bedside procedures.

Data Collection

Each morning an investigator contacted the senior residents who had admitted patients during the previous 24‐hour shift and confirmed that newly admitted patients were under the care of the general medicine service for more than 6 hours. To examine how the number of attempts may have been affected by procedures done in the emergency room or intensive care units before admission to the general medicine service, investigators also asked admitting residents whether a bedside procedure had been attempted in the 72 hours before admission. Every general medicine service resident was asked to fill out a brief data collection form after an attempt to perform any procedure on the general medical wards. In addition, chief residents asked each member of the general medicine service at mandatory sign‐out rounds at the end of each weekday whether any procedures had been attempted, and on weekend days investigators contacted senior residents from each general medicine service team.

We report on this quality assurance study, which was conducted during a pilot phase. This report has been reviewed and judged exempt by our institutional review board.

Primary OutcomeNumber of Procedure Attempts

For all bedside procedures attempted by residents on the general medical wards, investigators determined whether the residents were members of firms that were offered the bedside procedure service and, if so, whether the procedure service attending directly supervised the procedure attempt. Multiple procedure attempts of the same type were counted for an individual patient if (1) the procedure attempts did not occur during the same admissions and (2) neither the physicians attempting the procedure nor the primary indications for it were the same. Therefore, neither attempts performed after initially unsuccessful ones nor repeated procedures, such as large‐volume therapeutic paracentesis and thoracentesis, were counted twice. We reasoned that when these criteria were met, procedure attempts could be considered independently.

Secondary Outcomes

Investigators asked residents who attempted procedures to indicate whether (1) the indication for the procedure was solely diagnostic or was, at least in part, therapeutic; (2) the procedure was successful; and (3) there were any immediate major periprocedural complications. A procedure was considered to have been successfully performed if it fulfilled 2 criteria: it had to be completed during a single continuous attempt, even if multiple sites or procedure kits were used; and it had to fulfill the indication for it being done. For example, if the indication for thoracentesis was therapeutic, this procedure would be considered successful if it yielded a large enough volume of fluid to alleviate the patient's symptoms, but if the indication was diagnostic, then thoracentesis would be considered successful if it yielded enough fluid for laboratory processing. Residents were asked to report any periprocedural complications that they considered major; 2 illustrative examples were provided: a pneumothorax and severe bleeding.

Data Analyses

On the basis of earlier pilot data, we estimated that 8%10% of all admissions to the general medicine service underwent at least 1 procedure (paracentesis, thoracentesis, lumbar puncture, or central vein catheterization). We planned for a sample size of 1900 admissions, which would have 80% power to detect a clinically meaningful 50% relative increase in the mean number of bedside procedures with a double‐sided alpha error of 0.05. We used permutation tests to compare the mean number of procedures attempted between firms and bootstrap simulation to construct 95% confidence intervals for those means and the differences between and ratios of them. Fisher's exact test was used to compare proportions of successfully performed procedures and preadmission procedure attempts. All analyses were conducted with Stata Statistical Software, Release 9 (StataCorp, LP, College Station, TX).

Subjects

During this 8‐week pilot study, there were 2157 admissions to the general medicine service. Among these admissions, 216 were excluded from our study because the patients did not arrive on the medical wards or were not under the care of the general medicine service for at least 6 hours before discharge or before being transferred to another service. Of the remaining 1941 admissions, 935 were to firms with the bedside procedure service available, and 1006 were to firms without the service available (Fig. 1)

Figure 1

Primary OutcomeNumber of Procedure Attempts

Overall, 122 patients underwent 145 procedure attempts that met our criteria for independence. The mean number of procedure attempts in firms offered the bedside procedure service was 48% higher (90 versus 61 per 1000 admissions; RR 1.48, 95% CI 1.062.10; P = .030; Fig. 1). When procedures attempted on weekends and holidays were excluded, the relative increase in procedure attempts in firms offered the bedside procedure service was even higher (70 versus 43 per 1000 admissions; RR 1.63, 95% CI 1.092.49; P = .023; Fig. 1). When grouped according to whether procedure attempts occurred before or after crossover of the procedure service, the mean number of procedure attempts in firms was higher when the service was offered: firm A dropped from 84 to 70 per 1000 admissions (P = .58) after losing the service, whereas firms B and C increased from 57 to 94 per 1000 admissions (P = .025) on gaining the service. There were 40 procedure attempts performed on patients within 72 hours before admission, but there was no difference between firms in the proportions of these preadmission procedures (P = .43).

Secondary Outcomes

Table 1 shows how of each type of procedure contributed to the overall difference. Attempts of CVC and therapeutic paracentesis and thoracentesis accounted for 86% of the overall increase in procedure attempts for admissions to firms offered the bedside procedure service, whereas only 14% of this increase was a result of diagnostic procedures. There were no differences in the proportions of successfully performed procedures, whether grouped by firm (P = 1.0) or by direct supervision from the procedure service attending (P = .64; Table 2). There were 3 self‐reported major periprocedural complications; all were related to excessive bleeding from CVC attempts. Two occurred without direct supervision from the bedside procedure service attending, one hemomediastinum from an internal jugular CVC attempt and one groin hematoma from a femoral CVC attempt. The third, a groin hematoma from a femoral CVC attempt, occurred during direct supervision from the bedside procedure service attending.