Academic medical centers are increasingly employing hospitalists to staff teaching wards. Although studies have demonstrated reduced lengths of stay (LOSs) associated with hospitalist care, it is unclear which patients are most likely to benefit. We sought to determine whether patients with specific diagnoses or discharge needs account for the association between hospitalist care and reduced LOS.
Hospital admissions were divided into the following 2 groups based on type of attending physician: teaching hospitalist (full-time faculty hospitalist with no outpatient responsibilities) vs nonhospitalist (full-time or voluntary faculty contributing 1 or 2 months of teaching service per year). We included all patients discharged from an academic teaching service for a 2-year period. Data were extracted from the Montefiore Medical Center's clinical information system and the Social Security Death Registry.
Mean LOS was lower for teaching hospitalists than for nonhospitalists (5.01 vs 5.87 days [P < .02]). The reduction in LOS was greatest for patients requiring close clinical monitoring (patients with congestive heart failure, stroke, asthma, or pneumonia) and for those requiring complex discharge planning. There were no significant differences between the groups in readmission, in-hospital mortality, or 30-day mortality.
Teaching hospitalist care was associated with shorter LOS in patients requiring close clinical monitoring and complex discharge planning, without adversely affecting readmission or mortality rates.
Pressure to control costs of inpatient care has led many academic medical centers to employ hospitalists to provide care for medical inpatients and to staff medical teaching rounds.1 Hospitalist care has been associated with shorter lengths of stay (LOSs)2- 16 and lower costs3,4,7,9- 14,16,17 in a variety of settings, including after adjustment for physician experience.18 This reduction in LOS has been demonstrated among patients admitted with pneumonia9,17 and congestive heart failure.19 However, whether there are other specific patient diagnoses or discharge dispositions that seem to derive particular benefit from hospitalist care has not been firmly established.
The available data on clinical outcomes associated with hospitalist care are less consistent than the data on LOS. One study11 noted a 33% decrease in in-hospital mortality among patients on the hospitalist service, and this decrease remained significant at 60 days. However, these findings were confounded by nonrandom assignment to the hospitalist service and major differences between the treatment groups.11 Another study,14 in which assignment to the hospitalist service was without preference, found a statistically significant reduction in 30-day mortality in the hospitalist-treated group in the second year of the program, but included only 2 hospitalists. Despite these promising results, randomized studies with long-term follow-up of hospitalist vs nonhospitalist care are lacking.
Our primary objective was to determine which patient groups appear more likely to benefit from hospitalist care by comparing LOS and clinical outcomes (in-hospital mortality, 30-day mortality, and readmission) between hospitalists and nonhospitalists. We compared the following 2 models of care: teaching hospitalist (full-time faculty with no outpatient responsibilities) and nonhospitalist (full- or part-time faculty with no more than 2 months per year on the inpatient service). We hypothesized that hospitalist care would be associated with shorter LOS among patients with specific diagnoses, levels of acuity (overall severity of illness), and discharge needs, without adversely affecting clinical outcomes.
Montefiore Medical Center's Weiler Hospital is a 381-bed teaching hospital affiliated with the Albert Einstein College of Medicine. We extracted data on all discharges from the medical teaching service at Weiler Hospital for 2 years, from July 1, 2002, through June 30, 2004. Patients were assigned to a teaching hospitalist or nonhospitalist team in rotation at the time of admission by a senior admitting resident. Hospitalist and nonhospitalist teams are identical in every way other than type of attending physician conducting the rounds. Patients assigned to the nonteaching service were not included in this analysis. Although team assignment appears random, concealed allocation did not occur. Rather, we used a quasirandomized study design.2 Data on patient age, race or ethnicity, sex, insurance status, history of admissions, laboratory values, International Classification of Diseases, Ninth Revision (ICD-9) and diagnosis related group (DRG) codes, LOS, readmission, and in-hospital mortality were extracted from a replicate of the Montefiore Medical Center's clinical information system (Emerging Health Information Technology, Yonkers, New York). The data replicate is merged monthly with the Social Security Death Registry, which allowed extraction of 30-day mortality rates after discharge. The institutional review board of the Montefiore Medical Center approved the study.
The main outcome variables were LOS, readmission, and in-hospital and 30-day mortality. Length of stay was analyzed as a continuous variable and, although not normally distributed, was reported as a mean (as has been done previously5,14,15). Readmission was defined as an admission for any reason, within 30 days of discharge, to the Weiler Hospital or to the other Montefiore hospital, the Moses division hospital, and was analyzed as a dichotomous variable. In-hospital and 30-day mortality were analyzed as dichotomous variables.
The major independent variable was the team assignment of each admission (teaching hospitalist vs nonhospitalist). In addition, we examined patient- and physician-level covariates for each admission. Patient-level covariates included demographic characteristics (age, sex, and race or ethnicity), insurance status, number of prior admissions within the previous 90 days, albumin value at the time of admission, diagnosis of diabetes mellitus (defined as an ICD-9 code of 250 at discharge), DRG weight, primary discharge diagnosis (ICD-9 code), and discharge disposition (home with self-care, home with services, or nursing home). Because the source of data was a clinical information system, independent variables used for adjusted analyses could only be those available on most of the patients. We used DRG weight as a surrogate for the overall severity of illness, as has been done previously,14 because it was readily available on all patients. For length-of-stay analyses, we categorized DRG weight into quartiles, which represent levels of patient acuity. Albumin level at the time of admission was also used as a marker for overall disease severity, as has been done in earlier studies,20- 22 because it is available on most patients and is thought to be useful in the quantification of risk in research settings.23 The ICD-9 codes were grouped into diagnosis clusters using the Healthcare Cost and Utilization Project of the Agency for Healthcare Research and Quality system.24 Some of these diagnoses were further collapsed into clinical categories to make data reporting manageable (eg, neoplasm contains several Agency for Healthcare Research and Quality diagnosis classifications). For LOS analyses, only those diagnoses that were among the top 30 causes for hospitalization in the United States25 were included, to ensure adequate numbers of patients in each category. The major physician-level covariate of interest was the annual number of months on the service. There was no crossover of attending physicians between groups (ie, nonhospitalists never became hospitalists or vice versa).
First, the 2 groups (teaching hospitalist vs nonhospitalist) were compared with respect to demographic and clinical characteristics using unpaired, 2-tailed t tests; Mann-Whitney tests; or χ2 tests. Because DRG weight is not normally distributed, it is reported as a mean (using a t test with unequal variances for comparison between the groups) and as a median (using the Mann-Whitney test for comparison between the groups). Categorical variables are compared using χ2 tests. Second, we calculated the Somers D statistic and used it as a rank-based test26 to compare LOS between the groups, with standard errors robust to the clustering of patients within each physician. The analysis was repeated with and without truncation of LOS to the mean plus 3 SDs to assess the influence of outliers. Then, multivariate linear regression was used to compare LOS between the groups after adjusting for differences in demographic and clinical characteristics. For multivariate analyses, the variables for insurance and race or ethnicity were recoded into multiple dummy variables.
For the dichotomous outcomes of readmission and in-hospital and 30-day mortality, we initially compared rates using χ2 tests and then used multivariate logistic regression models to assess the effect of hospitalist care after adjusting for differences between the groups. Variables chosen for multivariate analyses included those that measured a significant difference between the groups (albumin level, DRG weight, and insurance); other variables were added, one by one, keeping those with a Wald statistic of P < .20. We calculated 95% confidence intervals using standard errors, which are robust to the clustering of patients by physician.
We used STATA statistical software, version 9.0 (StataCorp, College Station, Texas) for all statistical analyses and data manipulation.
From July 1, 2002, through June 30, 2004, there were 9047 discharges from the teaching service at the Weiler Hospital. Significant data were missing for 10 hospital stays; thus, the study sample included 9037 discharges. This included 2913 discharges from teaching hospitalist teams (32.2%) and 6124 discharges from nonhospitalist teams (67.8%). The patient populations were similar with respect to age, sex, race or ethnicity, presence of prior admission, and mix of acute diagnoses. The teaching hospitalist group included more patients with Medicaid, a lower mean albumin level, and a lower mean DRG weight (Table 1).
The mean ± SD LOS was 5.01 ± 6.12 days in the teaching hospitalist group vs 5.87 ± 7.51 days in the nonhospitalist group (P = .02; Table 2). The reduction in LOS was still large and significant after adjustment for potential confounders (unadjusted reduction, 0.86 day [P = .004]; reduction adjusted for race or ethnicity, insurance, DRG weight, albumin level, and prior admissions, 0.78 day [P < .001]). There was a statistically significant negative linear association between LOS (in days) and the number of months the attending physician was on the service per year, with each extra month of service associated with a 0.19-day reduction in LOS (β = −0.19 day per month of service [P < .002]).
To assess the influence of outliers on LOS, the analysis was repeated using the variable LOS truncated at the mean plus 3 SDs. In the truncated analysis, teaching hospitalist care was associated with reduced LOS when compared with nonhospitalist care (4.82 vs 5.54 days [P = .02]).
To assess the influence of patients returning with multiple admissions, the analysis was performed using only the first admission for each patient during the time period (n = 6883). This first-admission analysis also yielded significant reductions in LOS for teaching hospitalists vs nonhospitalists (4.96 vs 5.69 days [P = .04]).
To determine whether reduction in LOS is more pronounced among patients with specific diagnoses, level of acuity, or discharge needs, LOS data were analyzed after stratifying by clinical conditions, DRG weight, and discharge disposition (Table 3). Reductions in LOS associated with hospitalist care were most pronounced in patients with diagnoses of cerebrovascular accident (stroke), sepsis, asthma/chronic obstructive pulmonary disease, or urinary tract infection, without appreciable differences in LOS among patients with diagnoses of acute myocardial infarction, coronary disease without myocardial infarction, arrhythmia, or gastrointestinal tract bleeding. Lengths of stay were marginally shorter in patients with pneumonia (P = .13) and congestive heart failure (P = .06). The mean LOS was low for patients admitted with chest pain in both groups, but was slightly lower in the teaching hospitalist group (P = .04). Differences between hospitalists and nonhospitalists were significant in patients with lower overall acuity (DRG weight quartiles 1-3; mean difference, 0.49 day [P = .02]), but widened and remained significant in patients with the highest overall acuity (DRG weight quartile 4; mean difference, 1.95 days [P < .001]). Significant differences between groups were smaller in patients who were discharged home with self-care, widened when home services were required, and were largest when transfer to a nursing facility at the time of discharge was required.
After adjustment for demographic and clinical characteristics, there were no differences between the groups for readmission, in-hospital mortality, or 30-day mortality (Table 4). In first-admission-only analysis, after adjustment, there were no differences between the 2 groups for readmission (odds ratio, 0.92; 95% confidence interval, 0.80-1.06), in-hospital mortality (odds ratio, 0.95; 95% confidence interval, 0.72-1.27), or 30-day mortality (odds ratio, 0.99; 95% confidence interval, 0.82-1.18).
Our experience with a teaching hospitalist service confirmed several now-familiar findings of hospitalist care, including shorter LOS for patients treated by teaching hospitalist teams. We further showed that teaching hospitalist care seems to have a stronger association with LOS in patients requiring specialized services on discharge, such as home care services or transfer to a nursing facility. We were able to measure discharge planning needs with the use of a separate discharge disposition code, which was assigned independent of billing at the time of discharge. We believe that the greater reduction in LOS associated with complex discharge planning reflects hospitalist skills in working with ancillary staff, such as social workers or discharge planners.
In addition, we found that hospitalist care had the strongest association with LOS in patients with the highest acuity, as measured by the DRG weight. This important finding might help guide the use of hospitalist services, given that small numbers of patients at the highest acuity levels account for disproportionate numbers of total hospital days.27
Hospitalist care had the strongest association with LOS in patients with specific diagnoses, including cerebrovascular accidents (strokes), congestive heart failure, pneumonia, sepsis, urinary tract infections, and asthma/chronic obstructive pulmonary disease. These findings may reflect hospitalists' ability to closely monitor patients with dynamic conditions such as congestive heart failure, pneumonia, sepsis, urinary tract infections, and asthma/chronic obstructive pulmonary disease. The close monitoring and continuous presence offered by hospitalists may allow for earlier discharge because hospitalists are more likely to detect clinical improvement in real time and to make appropriate adjustments in treatment regimens. For patients with stroke, in particular, shorter LOS associated with hospitalist care may also reflect more effective attention to complex discharge needs such as rehabilitation services.
When we analyzed LOS by the number of months the attending physician spent on the service per year, we showed a statistically significant negative linear trend toward shorter LOS with increasing numbers of inpatient months. This suggests that extra inpatient experience is associated with even larger reductions in LOS and supports earlier findings that reductions in LOS with hospitalist care were greater14 or became significant11 in the second year of hospitalist programs.
In contrast to other analyses, the patients in our 2 groups were largely similar, but some differences persisted. Slightly sicker patients may have been assigned to teaching hospitalist teams because, at any given time, the existing census was likely to be lower for a teaching hospitalist team owing to the shorter LOS of their patients.
Despite the strengths of our study, it has some limitations. Similar to other studies of hospitalist care, the groups of attending physicians were small (5 teaching hospitalists and 54 nonhospitalists); thus, an individual outlier may have had a significant effect on the outcomes. Given the large number of diagnoses examined, the results of LOS for specific diagnoses need to be interpreted with caution. In addition, we had limited power to detect small differences in mortality rates. While it is reassuring that we did not find a statistically significant difference in readmission or mortality rates associated with hospitalist care, it would be erroneous to infer that there is no difference. Given our sample size and the event rates in the nonhospitalist group, and requiring a power of 80%, we found minimal detectable odds ratios of 1.18 for readmission, 1.38 for in-hospital events, and 1.24 for 30-day mortality. The strongest claim that can be inferred from this analysis is that if there is a difference in these outcomes with hospitalist care, it is likely less than these. Smaller differences in mortality rates would be detectable only with a larger sample. Finally, although our data were collected in 2002 to 2004, they reflect a practice in common use today, namely, hospitalists being used for inpatient teaching service in academic medical centers.
In conclusion, teaching hospitalist care was associated with reduced LOS on an academic teaching service, but not with increased readmission or mortality rates. We identified several patient groups in whom this association was particularly strong: patients who require close clinical monitoring, those in whom real-time adjustment of therapy is critical, those with high overall acuity, and those in whom complex discharge planning is necessary.
Correspondence: William N. Southern, MD, MS, Department of Medicine, Weiler Hospital of Albert Einstein College of Medicine, 1825 Eastchester Rd, Bronx, NY 10461 (email@example.com).
Accepted for Publication: May 17, 2007.
Author Contributions:Study concept and design: Southern, Berger, Bellin, and Arnsten. Acquisition of data: Southern, Berger, and Bellin. Analysis and interpretation of data: Southern, Bellin, Hailpern, and Arnsten. Drafting of the manuscript: Southern, Hailpern, and Arnsten. Critical revision of the manuscript for important intellectual content: Southern, Bellin, and Arnsten. Statistical analysis: Southern and Hailpern. Obtained funding: Arnsten. Administrative, technical, and material support: Berger and Arnsten. Study supervision: Bellin and Arnsten.
Financial Disclosure: None reported.
Funding/Support: This study was supported by the Institute for Medical Effectiveness Research, a joint project of the Albert Einstein College of Medicine and the North Shore–Long Island Jewish Health System, and the Clinical Investigation Core of the Center for AIDS Research at the Albert Einstein College of Medicine and Montefiore Medical Center (National Institutes of Health grant P30 AI51519).
Previous Presentation: Portions of this paper were presented at the 29th Annual Meeting of the Society for General Internal Medicine; April 28, 2006; Los Angeles, California.
Additional Contributions: Lori Ashton, BS, provided programming support.
Country-Specific Mortality and Growth Failure in Infancy and Yound Children and Association With Material Stature
Use interactive graphics and maps to view and sort country-specific infant and early dhildhood mortality and growth failure data and their association with maternal
Thank you for submitting a comment on this article. It will be reviewed by JAMA Internal Medicine editors. You will be notified when your comment has been published. Comments should not exceed 500 words of text and 10 references.
Do not submit personal medical questions or information that could identify a specific patient, questions about a particular case, or general inquiries to an author. Only content that has not been published, posted, or submitted elsewhere should be submitted. By submitting this Comment, you and any coauthors transfer copyright to the journal if your Comment is posted.
* = Required Field
Disclosure of Any Conflicts of Interest*
Indicate all relevant conflicts of interest of each author below, including all relevant financial interests, activities, and relationships within the past 3 years including, but not limited to, employment, affiliation, grants or funding, consultancies, honoraria or payment, speakers’ bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued. If all authors have none, check "No potential conflicts or relevant financial interests" in the box below. Please also indicate any funding received in support of this work. The information will be posted with your response.
Register and get free email Table of Contents alerts, saved searches, PowerPoint downloads, CME quizzes, and more
Subscribe for full-text access to content from 1998 forward and a host of useful features
Activate your current subscription (AMA members and current subscribers)
Purchase Online Access to this article for 24 hours
Some tools below are only available to our subscribers or users with an online account.
Download citation file:
Web of Science® Times Cited: 35
Customize your page view by dragging & repositioning the boxes below.
and access these and other features:
Enter your username and email address. We'll send you a link to reset your password.
Enter your username and email address. We'll send instructions on how to reset your password to the email address we have on record.
Athens and Shibboleth are access management services that provide single sign-on to protected resources. They replace the multiple user names and passwords necessary to access subscription-based content with a single user name and password that can be entered once per session. It operates independently of a user's location or IP address. If your institution uses Athens or Shibboleth authentication, please contact your site administrator to receive your user name and password.