PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptNIH Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Ann Intern Med. Author manuscript; available in PMC Aug 2, 2012.
Published in final edited form as:
PMCID: PMC3196599
NIHMSID: NIHMS327394

Association of Hospitalist Care With Medical Utilization After Discharge: Evidence of Cost Shift From a Cohort Study

Abstract

Background

Hospitalist care has grown rapidly, in part because it is associated with decreased length of stay and hospital costs. No national studies examining the effect of hospitalist care on hospital costs or on medical utilization and costs after discharge have been done.

Objective

To assess the relationship of hospitalist care with hospital length of stay, hospital charges, and medical utilization and Medicare costs after discharge.

Design

Population-based national cohort study.

Setting

Hospital care of Medicare patients.

Patients

A 5% national sample of enrollees in Medicare parts A and B with a primary care physician who were cared for by their primary care physician or a hospitalist during medical hospitalizations from 2001 to 2006.

Measurements

Length of stay, hospital charges, discharge location and physician visits, emergency department visits, rehospitalization, and Medicare spending within 30 days after discharge.

Results

In propensity score analysis, hospital length of stay was 0.64 day less among patients receiving hospitalist care. Hospital charges were $282 lower, whereas Medicare costs in the 30 days after discharge were $332 higher (P < 0.001 for both). Patients cared for by hospitalists were less likely to be discharged to home (odds ratio, 0.82 [95% CI, 0.78 to 0.86]) and were more likely to have emergency department visits (odds ratio, 1.18 [CI, 1.12 to 1.24]) and readmissions (odds ratio, 1.08 [CI, 1.02 to 1.14]) after discharge. They also had fewer visits with their primary care physician and more nursing facility visits after discharge.

Limitation

Observational studies are subject to selection bias.

Conclusion

Decreased length of stay and hospital costs associated with hospitalist care are offset by higher medical utilization and costs after discharge.

Primary Funding Source

National Institute on Aging and National Cancer Institute.

Care of hospitalized patients by full-time, hospital-based physicians— called hospitalists—has grown rapidly over the past decade (1, 2). This growth was fueled in part by an increasing emphasis on efficiency, coupled with evidence from prospective trials that hospitalist care reduced average length of hospital stay (36). Major concerns about hospitalists relate to threats to continuity of care (7, 8). Discontinuity across the transition from outpatient to inpatient settings and back again might result in poor communication of important information (7, 912). Miscommunication at the time of hospital discharge can result in serious medication errors, which may in turn precipitate an emergency department visit (1315). In addition, a hospitalist’s lack of familiarity with the patient’s preferences and values might lead to less-optimal treatment choices (16, 17). Pressure to reduce length of stay may promote early discharge to other health care facilities rather than discharge to home.

We used a 5% national sample of Medicare patients to examine the growth of hospitalist care over time (1, 18) and then used data from this sample to estimate hospital and Medicare costs in the 30 days after discharge that were associated with hospitalist care for all medical conditions. Our underlying hypothesis was that hospitalist care would be associated with cost shifting from the hospital setting to the posthospital setting—specifically, that hospitalist care would be associated with decreased discharges to home, and that discontinuity of care associated with hospitalist care would lead to more visits to the emergency department and readmissions to the hospital, resulting in increased Medicare costs.

Methods

Study Sample

We selected hospital admissions with a medical diagnosis-related group (DRG) from January 2001 to November 2006 in a 5% representative national sample of Medicare beneficiaries. We then constructed data files to include beneficiaries’ demographic and enrollment information (denominator file) and claims for hospital stays (Medicare Provider Analysis and Review [MEDPAR] files), claims from outpatient facility files (outpatient statistical analysis files [OUTSAF]), and claims from physician services (carrier files).

To make patients cared for by hospitalists and those cared for by their primary care physicians (PCPs) more comparable, we included only admissions for patients with an identified PCP before admission (8). We defined patients as having a PCP if they had 2 or more visits to the same generalist physician (general internist, family physician, general practitioner, or geriatrician) in the year before hospitalization (19). We identified the treating physicians for each hospitalization by linking inpatient evaluation and management (E&M) codes in the carrier files to the admission record in the MEDPAR files. We excluded admissions if neither the patient’s PCP nor a hospitalist provided care or if patients were cared for by both their PCP and a hospitalist. For patients with more than 1 admission in a given year, we randomly selected 1 admission per patient per year. We excluded admissions with intensive care unit services or inpatient deaths and those outside the 50 states, resulting in 205 190 admissions at 4657 hospitals in the entire study cohort. For our main analysis, we selected only hospitals with at least 20 admissions cared for by hospitalists and at least 20 admissions cared for by PCPs during the study period. This final study cohort included 58 125 admissions at 454 hospitals.

Measurements

Medicare enrollment files provided data on patient age, sex, and race. We used Medicaid eligibility in the enrollment file as a proxy for low income. We obtained rates of weekend versus weekday admissions and DRGs (categorized as neurology, pulmonary, cardiac, gastrointestinal, and other) from the MEDPAR files. We determined residence in a nursing facility before admission from the MEDPAR files and by searching for any E&M codes associated with nursing facilities in the 3 months before admission (20). We generated Elixhauser comorbidity measures (21), total hospitalizations, and outpatient visits by using both inpatient and physician claims from MEDPAR, OUTSAF, and carrier files in the year before admission.

The provider-of-service file provided information on the hospitals, including census region, size (<200, 200 to 349, 350 to 499, or ≥500 beds), medical school affiliation (major, minor, or nonteaching), and type of hospital (public, for-profit, or nonprofit). Hospitals with a major medical school affiliation host clinical clerkship programs, whereas those with a minor affiliation have only residency programs or occasional student rotations (22). We generated metropolitan size from 2000 U.S. Census data. We defined “hospitalists” as generalist physicians who had at least 5 Medicare E&M claims and generated 90% or more of these claims from care provided to hospitalized patients at each year in the study period. When validated at 7 hospitals (1), this definition had a sensitivity of 84.2% and specificity of 96.5%. A further validation, which used generalists (n = 8928) with 90% or more E&M claims from hospitalized patients in the 100% Medicare data from Texas in 2006 as the gold standard, found a sensitivity of 84.7%, specificity of 98.7%, and positive predictive value of 90.9% for the algorithm with the 5% Medicare sample. We used The Dartmouth Atlas of Health Care definition of hospital referral regions as naturally occurring tertiary care markets within which Medicare beneficiaries receive inpatient services (23).

Context

Most studies of hospitalist care have been limited to inpatient stay.

Contribution

This study compared the patients of primary care physicians with those of hospitalists and found that hospitalist patients had shorter and less expensive admissions. After discharge, however, they had more visits to the emergency department, more readmissions to the hospital, and higher total expenses.

Caution

Observational studies (such as this one) may be biased in ways that are difficult to detect.

Implication

The decreased inpatient expenses associated with hospitalist care are offset by higher expenses after discharge.

The Editors

Study Outcomes

Data on hospital length of stay were from the MEDPAR files. We estimated the cost of hospitalization by the amount of total charges in the MEDPAR files. We estimated Medicare costs in the 30 days after discharge by the Medicare paid amounts in the MEDPAR, OUTSAF, and carrier files, following the payment-calculation worksheets provided by the Research Data Assistance Center (24, 25). On the basis of E&M codes in the carrier files, we classified provider visits as outpatient, emergency department, or nursing facility services. We also calculated total Medicare spending in the period from 12 months to 1 month before admission for each admission and adjusted for that in the cost-estimation model. We presented the adjusted cost differences in 2006 U.S. dollars by using the gross domestic product deflator.

For the analysis of 30-day rehospitalization and emergency department visits, we excluded admissions with transfers to other acute care hospitals and deaths without an event (rehospitalization or emergency department visit) occurring within 30 days after discharge. In the analysis of discharge location, we excluded admissions with transfers to other acute care hospitals and those in which the patients had resided in a nursing facility any time in the 3 months before admission.

Statistical Analysis

To assess differences between patients cared for by hospitalists and those cared for by their PCP within a hospital, we used a nonpooling approach of propensity analysis (26, 27) and treated hospital as a covariate. We generated the propensity that an admission would be cared for by a hospitalist from a logistic regression model that incorporated the patient characteristics listed in Table 1 for each hospital. This approach implicitly balances hospital-level characteristics that affect choice of hospitalist care. We examined the assumptions for use of propensity score analysis, including overlap, balance, and ability to pull across deciles (Appendix 1, Appendix Figure, and Appendix Table 1, available at www.annals.org). We excluded patients with propensity scores in the nonoverlap regions within each hospital (27).

Appendix Figure
Distribution of propensity scores for admissions cared for by hospitalists and primary care physicians.
Table 1
Characteristics of Hospitalized Patients Cared for by Hospitalists or Their PCPs
Appendix Table 1
Association Between Hospitalist Care and Rates of Readmission, ED Visits, and Discharge to Home, by Decile of Propensity Score

We used conditional logistic regression models to avoid inconsistent estimates of the within-hospital effect for the binary outcomes. For nonbinary outcomes, we used generalized linear models with log-link, normal distribution for length of stay; Poisson regression for number of provider visits after discharge; and log-gamma models for hospital costs (28). Some admissions had no Medicare spending in the 30 days after discharge. Therefore, in estimating differences in postdischarge Medicare costs, we used a 2-part model: a logit model that estimated the percentage of patients with any Medicare costs and a log-gamma model that estimated the average costs in patients with any costs (29, 30). We computed the 95% CIs of the cost differences from the 2-part model by using the cluster bootstrap method with 1000 bootstrap samples of the hospitals. All of these models included the hospital as a covariate and controlled for the propensity score decile within hospitals. To allow for nonlinear relationships, we included the propensity score decile as a categorical covariate in the analysis. For the cost models, we also adjusted for Medicare spending in the year before admission.

We also conducted a sensitivity analysis for our final study cohort (31). We postulated prevalence differences and effect sizes for an unmeasured confounder on the basis of our findings with the known confounders (Appendix 2, available at www.annals.org). To determine whether the effect of hospitalist care varied by hospital or patient characteristics, we tested interactions between hospitalist care and these characteristics in multivariate conditional logistic regression models for the final study cohort (Appendix 2).

Only 28.3% of admissions and 9.7% of hospitals in our entire study cohort met the criteria for the main analysis. We also analyzed the entire study cohort (205 190 admissions at 4657 hospitals) by using marginal models to study the average differences between patients cared for by hospitalists and those cared for by their PCPs. Appendix 2 details the methods used. We performed all analyses by using SAS, version 9.1 (SAS Institute, Cary, North Carolina), and STATA, version 9.2 (StataCorp, College Station, Texas).

Role of the Funding Source

The National Institute on Aging and National Cancer Institute funded this study. The funding sources had no role in the design and conduct of the study; the collection, management, analysis, and interpretation of the data; or the preparation, review, or approval of the manuscript.

Results

Table 2 shows the characteristics of the 454 hospitals included in the final study cohort. Compared with the 4657 hospitals in the entire study cohort (Appendix Table 2, available at www.annals.org), the 454 hospitals were larger, were more likely to be teaching and nonprofit hospitals, and were located in larger metropolitan areas. Table 1 presents the characteristics of the final cohort of hospitalized patients, stratified by whether their in-hospital care was provided by their PCP or a hospitalist. Because of the large sample size, even small differences will be statistically significant. The patients cared for by hospitalists were slightly younger, were more likely to be white, had slightly more comorbid conditions, and were more likely to be admitted on the weekends and from nursing homes. Geographic region, hospital size, and hospital type also differed between the 2 groups. All differences except race were removed after we controlled for the propensity score decile.

Table 2
Characteristics of the 454 Hospitals*
Appendix Table 2
Characteristics of the 4657 Hospitals in the Entire Cohort*

Table 3 compares hospitalized patients cared for by hospitalists with those cared for by their PCPs on medical utilization and costs during and after hospitalization. We estimated length of stay, overall hospital costs, and Medicare spending in the 30 days after discharge. We also stratified the postdischarge spending by type of service (such as professional service or rehospitalization), controlling for the propensity of receiving hospitalist care within each hospital. The adjusted length of stay was 0.64 day shorter and the hospital charges were $282 lower for patients cared for by hospitalists, whereas total Medicare spending in the 30 days after discharge was $332 higher (P < 0.001 for all). When the postdischarge costs were stratified by type of service, 59% was from rehospitalization, 19% from skilled-nursing facilities, and 22% from professional and other services.

Table 3
Propensity Analyses Estimating Mean Length of Stay, Mean Hospital Charges, and Medicare Costs in the 30 Days After Discharge for Hospitalized Patients Cared for by Hospitalists or Their PCPs*

We also examined the number of provider visits in the 30 days after discharge categorized by type of provider (Table 4). Patients cared for by their PCPs had more visits to their PCP and fewer visits to other outpatient providers. Patients cared for by their PCPs also had fewer provider visits in emergency departments and nursing facilities (P < 0.001 for all comparisons). The results were similar in the unadjusted and the propensity score–adjusted analyses (Table 4).

Table 4
Provider Visits in the 30 Days After Hospitalization for Patients Cared for by Hospitalists or Their PCPs, by Type of Provider Visit

Table 5 shows the association of hospitalist care with the proportion of patients discharged to home, readmissions, and emergency department visits within 30 days of discharge. The unadjusted rates and propensity score–adjusted odds ratios are presented for each measure. Patients cared for by hospitalists were less likely to be discharged directly to home (odds ratio, 0.82 [95% CI, 0.78 to 0.86]) and were more likely to have emergency department visits (odds ratio, 1.18 [CI, 1.12 to 1.24]) and readmissions (odds ratio, 1.08 [CI, 1.02 to 1.14]) in the 30 days after discharge.

Table 5
Rate of Readmission, ED Visits, and Discharge to Home in the 30 Days After Discharge for Patients Cared for by Hospitalists or Their PCPs

The decreased odds of discharge to home and increased odds of emergency department visits associated with hospitalist care (Table 5) were robust in the sensitivity analysis, postulating an unmeasured confounder with a range of prevalence differences and effect sizes. However, the higher 30-day readmission rate associated with hospitalist care became insignificant with relatively small postulated differences in prevalence and magnitude of effect size (Appendix Table 3, available at www.annals.org).

Appendix Table 3
Sensitivity Analyses of the Association Between Hospitalist Care and Rates of Readmission, ED Visits, and Discharge to Home, Controlling for an Unmeasured Binary Confounder

Interactions between hospitalist care and other covariates were tested in multivariate conditional logistic regression models (Appendix Table 4, available at www.annals.org). We found no statistically significant interactions between hospitalist care and any hospital characteristic on odds of discharge to home, readmission, and emergency department visits and no interactions with patient age, DRG group, comorbid condition, or year of hospitalization. A subgroup analysis removing patients with a nursing home stay before hospitalization produced similar results.

Appendix Table 4
Multivariate Analyses of Odds of Readmission, ED Visit, and Discharge to Home

Finally, we conducted confirmatory analyses by using the entire study cohort (205 190 admissions at 4657 hospitals) (Appendix Tables 2 and and5,5, available at www.annals.org), similar to those presented in Tables 3 and and5.5. The results, presented in Appendix Tables 6 and and77 (available at www.annals.org), are similar to those found with the smaller cohort.

Appendix Table 5
Characteristics of Hospitalized Patients Cared for by Hospitalists or Their PCPs in the Entire Cohort*
Appendix Table 6
Mean Length of Stay, Mean Hospital Charges, and Medicare Costs in the 30 Days After Discharge for Hospitalized Patients Cared for by Hospitalists or Their PCPs for the Entire Cohort*
Appendix Table 7
Association of Hospitalist Care With Rates of Readmission, ED Visit, and Discharge to Home for the Entire Cohort*

Discussion

The reduction in hospital charges associated with hospitalist care in this national study of Medicare patients is smaller than the cost savings reported in the prospective trials (3, 6) but is similar to that in an observational study of hospitalist care in 45 hospitals (32). It could be argued that the randomized trials, unblinded by necessity, were conducted by early adopters of the hospitalist model, and their results might not translate nationally. Our finding of $332 more Medicare spending associated with hospitalist care in the 30 days after discharge means that all of the reduction in hospital costs shifted to costs after discharge. The cost shift might be considered modest. However, if applied to the approximate 25% of Medicare admissions cared for by hospitalists (1, 2), this represents more than $1.1 billion in additional Medicare costs annually (33).

The prospective trials and several observational studies found no significant increase in postdischarge visits to emergency departments or readmissions associated with hospitalist care (36, 32). Most prior studies were from single institutions and lacked the statistical power to detect the differences we found. As the hospitalist model has rapidly disseminated nationally, the outcomes may also have changed. A recent prospective analysis of hospitalist care for patients with upper gastrointestinal hemorrhage at 6 academic hospitals also found higher readmission rates in the patients assigned to hospitalists (34). We also found a higher rate of readmissions after hospitalist care in patients with stroke (35).

In the unadjusted analysis, 5.5% more patients who were cared for by their PCPs were discharged to home compared with patients cared for by hospitalists. The differences remained significant in multivariate, propensity, and sensitivity analyses. The adjusted difference translates to a reduction of 120 000 patients discharged to home per year, suggesting that the decrease in length of stay by hospitalists may be obtained at a cost of increased discharges to other health care facilities, such as skilled-nursing facilities.

Our study has limitations. We limited the patient cohort to those who had an identified PCP, and we compared patients cared for by their PCP in the hospital with those cared for by hospitalists. Thus, our results may not be applicable to patients without an identified PCP. In addition, the study included only patients admitted with medical diagnoses. Hospitalists have a smaller effect on length of stay with surgical patients (18), so the results may differ. Also, we did not include patients cared for by medical subspecialists who were hospitalists.

Another limitation is that we studied patients with fee-for-service Medicare coverage. The results may not be applicable to younger patients and patients in HMOs. Our study period is 2001 to 2006; results might differ in later years. However, within our study period, we did not find a significant interaction between hospitalist care and year of hospitalization. We did not directly assess costs of hospitalization or costs after discharge. We used total charges as an indicator of hospital costs during hospitalization and calculated total Medicare spending in the 30 days after discharge. Charges reflect price setting rather than resource consumption and, as a result, overestimate costs. Medicare reimbursement does not include other payers’ payments, out-of-pocket expenses, and copayments; therefore, it underestimates costs. Thus, the estimated cost shift in our study is conservative.

The hospitals in the study cohort were limited to those with at least 20 patients in each group (hospitalist and PCP) to allow for propensity adjustment at the hospital level. However, analyses without propensity adjustment for the entire cohort produced similar results (Appendix Tables 6 and and77).

The estimates from the unadjusted, multivariate-adjusted, and propensity score–adjusted analyses were remarkably similar, which reflects the fact that the patients cared for by hospitalists versus those cared for by their PCPs differed only slightly in key characteristics (Table 1). In observational trials, it is always possible that an unmeasured confounder is responsible for the results. The fact that the measured potential confounders differed very little in prevalence even before statistical adjustment allays that concern somewhat.

In conclusion, in this national study, patients cared for by hospitalists versus their PCPs had a shorter length of stay but were less likely to be discharged to home; were less likely to see their PCP after discharge; and had more readmissions, emergency department visits, and nursing home visits after discharge. Hospital cost savings associated with hospitalists were offset by increased medical utilization and costs after discharge. Many incentives and opportunities for cost shifting are in the present Medicare system (36, 37). In the current example, the cost shift is from a fixed prospective payment (DRG) to post-discharge services in a fee-for-service system, thus lowering hospital costs while increasing overall Medicare cost. Hospitalists, who typically are employed or subsidized by hospitals (2), may be more susceptible to behaviors that promote cost shifting. Current efforts to increase bundling of payments based on episodes of care should reduce these incentives and clarify the effect of different models of hospital care on overall medical costs.

Acknowledgments

Grant Support: By the National Institute on Aging (grants 1R01-AG033134 and P30AG024832) and the National Cancer Institute (grant K05-CA134923).

Appendix 1: Propensity Score Analysis

We restricted the analysis to the 454 hospitals with at least 20 admissions in each category: PCPs and hospitalists. We estimated the propensity score by using a separate logistic regression model for each hospital. In each model, the dependent variable was 1 for hospitalized patients cared for by a hospitalist and 0 for hospitalized patients cared for by their PCP. We included the patient characteristics listed in Table 1: age, sex, race, low income, weekend admission, admission from nursing facilities, DRG weight, DRG group, number of comorbid conditions, number of hospitalizations, and number of outpatient physician visits in the prior year. We examined the assumptions of the propensity score model within hospitals and overall. The 16.8% of admissions in the nonoverlapping regions within a hospital were excluded (27). The Appendix Figure shows the distribution of propensity scores and decile markers for the admissions in the overlapping regions across the 454 hospitals. For hospitalized patients cared for by hospitalists and their PCPs, the ranges of the propensity scores were 0.003 to 0.981 and 0.003 to 0.982, respectively.

All patient characteristics except for race were balanced between the 2 groups in the overall study cohort. Within each hospital, patient characteristics were balanced, except for race and low income for 5 hospitals (1.1% of studied hospitals).

Appendix Table 1 shows the association between hospitalist care and the odds of readmission, emergency department visits, and discharge to home, stratified by decile of propensity score. We found little evidence of an interaction of hospitalist care and propensity score decile on odds of readmission (P = 0.766), emergency department visits (P = 0.748), and discharge to home (P = 0.555).

Appendix 2: Confirmatory Analyses

We conducted 3 sets of confirmatory analyses to examine the robustness of study results: sensitivity analysis, multivariate analysis for the final study cohort (n = 58 125), and population average models for the entire study cohort (n = 205 190).

Sensitivity Analysis

We performed a sensitivity analysis to estimate the potential effect of an unmeasured confounder or confounders on the odds of readmission, emergency department visits, and discharge to home. Appendix Table 3 shows the results of controlling for a postulated unmeasured confounder with different prevalences and odds ratios for the unknown factor. On the basis of our known confounding variables listed in Table 1, the largest difference in prevalence of confounders between the hospitalist and PCP groups was 4.6%. Therefore, we chose a 5– and 10– percentage point difference in prevalence for the 2 groups, with a prevalence level of 5% or 20% in the PCP group. We chose low prevalence levels for the postulated unknown confounder in the PCP group because such situations produce the largest effect in estimating the association between hospitalist care and outcomes.

We used a similar approach to construct possible effect sizes for the postulated unknown confounder. Except for the effect of DRG group on discharge to home, the known confounding variables had effect sizes (odds ratios) below 1.5 (Appendix Table 4). Therefore, we chose effect sizes for the unmeasured confounders with odds ratios of 1.5, 2.5, and 3.5.

Appendix Table 3 shows the prevalence of the postulated confounder for each group (hospitalist and PCP) and the postulated effect size (odds ratio). We then applied this unmeasured confounder to the propensity analysis, calculating the association between hospitalist care and outcomes. For example, if the difference in prevalence of an unmeasured confounder between hospitalized patients cared for by hospitalists and those cared for by their PCPs is 5 percentage points (25% vs. 20% in the 2 groups) and the association of the confounding variable with readmission is 1.5, then the odds of readmission associated with hospitalist care becomes 1.06 (CI, 1.00 to 1.11). The association of hospitalist care with increased odds of readmission became insignificant, with relatively small postulated effect sizes and small differences in prevalence for an unknown confounder. However, the association of hospitalist care with odds of emergency department visits and discharge to home was very robust.

Multivariate Analysis

We examined the association between hospitalist care and costs in the 30 days after discharge for the final study cohort by using a multivariate 2-part model. We also assessed the associations of hospitalist care and readmission, emergency department visits, and discharge to home for the final study cohort by using conditional logistic regression models. This allowed us to show the magnitude of the associations of the covariates with the outcomes and also allowed for testing of interactions. Appendix Tables 4 and and88 show the complete models, including covariates.

Appendix Table 8
Multivariate Analyses of 30-Day Postdischarge Costs*

Analysis Involving the Entire Cohort

In Appendix Tables 2, ,5,5, ,6,6, and and7,7, we present analyses with the entire cohort of 205 190 admissions at 4657 hospitals. Appendix Table 2 presents the characteristics of the 4657 hospitals. Appendix Table 5 presents patient characteristics stratified by whether the patients received hospitalist or PCP care. Appendix Table 6 presents the results of the analysis of length of stay, hospital charges, and Medicare costs in the 30 days after discharge for patients cared for by hospitalists or their PCPs. This analysis is similar to that presented in Table 3. Appendix Table 7 presents the results of analysis of readmission, emergency department visits, and discharge to home within 30 days after discharge for patients who received care from hospitalists or their PCPs. This analysis is similar to that presented in Table 5.

We analyzed the average differences between patients cared for by hospitalists and PCPs by using marginal models—generalized estimating equations with an exchangeable working correlation matrix (28). We reported the differences for each outcome associated with hospitalist care with their 95% CIs on the basis of a modified sandwich estimate of variance. In the cost-estimation models, we also adjusted the geographic adjustment factor available from the Centers for Medicare & Medicaid Services Web site (38). To further adjust for clustering within hospital referral regions, we treated hospital referral region as a covariate in the generalized estimating equation analysis. We also conducted a propensity score analysis by using a complete pooling approach.

Comparing results shown in Appendix Tables 2 and and55 with those in Tables 1 and and2,2, the differences in characteristics of hospitalized patients cared for by hospitalists versus their PCPs were larger in the entire cohort than in the final cohort, especially for hospital characteristics. The differences in average hospital length of stay, average hospital costs, and average costs in the 30 days after discharge between hospitalized patients cared for by hospitalists and those cared for by their PCPs estimated from the marginal model with the entire cohort were similar to those with the final cohort (Appendix Table 6 vs. Table 3). Average adjusted length of stay was 0.63 day shorter and average adjusted hospital charges were $296 lower for hospitalized patients cared for by hospitalists. Average adjusted total Medicare spending in the 30 days after discharge was $228 higher for admissions cared for by hospitalists (Appendix Table 6). The estimates of the association between hospitalist care and odds of readmission, emergency department visits, and discharge to home from the marginal models in the entire cohort were also close to those from the conditional logistic regression models used in the final cohort (Appendix Table 7 vs. Table 5). Results of the propensity analysis were similar to those of multivariate analysis. Further adjustments for clustering of hospitals within hospital referral regions were almost identical to results without adjustments.

Footnotes

Potential Conflicts of Interest: Disclosures can be viewed at www.acponline.org/authors/icmje/ConflictOfInterestForms.do?msNum=M11-0086.

Reproducible Research Statement: Study protocol and data set: Not available. Statistical code: Available from Dr. Kuo (yokuo/at/utmb.edu).

Author Contributions: Conception and design: Y.F. Kuo, J.S. Goodwin.

Analysis and interpretation of the data: Y.F. Kuo, J.S. Goodwin.

Drafting of the article: Y.F. Kuo, J.S. Goodwin.

Critical revision of the article for important intellectual content: Y.F. Kuo, J.S. Goodwin.

Final approval of the article: Y.F. Kuo, J.S. Goodwin.

Provision of study materials or patients: Y.F. Kuo, J.S. Goodwin.

Statistical expertise: Y.F. Kuo.

Administrative, technical, or logistic support: Y.F. Kuo, J.S. Goodwin.

Collection and assembly of data: Y.F. Kuo, J.S. Goodwin.

References

1. Kuo YF, Sharma G, Freeman JL, Goodwin JS. Growth in the care of older patients by hospitalists in the United States. N Engl J Med. 2009;360:1102–1112. [PMID: 19279342] [PMC free article] [PubMed]
2. Society of Hospital Medicine releases results of the 2007–2008 survey on the state of hospital medicine movement [News release] Philadelphia: Society of Hospital Medicine; 2008. Apr 4, Accessed at www.hospitalmedicine.org/AM/Template.cfm?Section_Press_Releases&Template_/CM/HTMLDisplay.cfm&ContentID_17477 on 26 May 2011.
3. Wachter RM, Katz P, Showstack J, Bindman AB, Goldman L. Reorganizing an academic medical service: impact on cost, quality, patient satisfaction, and education. JAMA. 1998;279:1560–1565. [PMID: 9605901] [PubMed]
4. Palmer HC, Armistead NS, Elnicki DM, Halperin AK, Ogershok PR, Manivannan S, et al. The effect of a hospitalist service with nurse discharge planner on patient care in an academic teaching hospital. Am J Med. 2001;111:627–632. [PMID: 11755506] [PubMed]
5. Kearns PJ, Wang CC, Morris WJ, Low DG, Deacon AS, Chan SY, et al. Hospital care by hospital-based and clinic-based faculty: a prospective, controlled trial. Arch Intern Med. 2001;161:235–241. [PMID: 11176737] [PubMed]
6. Meltzer D, Manning WG, Morrison J, Shah MN, Jin L, Guth T, et al. Effects of physician experience on costs and outcomes on an academic general medicine service: results of a trial of hospitalists. Ann Intern Med. 2002;137:866–874. [PMID: 12458986] [PubMed]
7. Pham HH, Grossman JM, Cohen G, Bodenheimer T. Hospitalists and care transitions: the divorce of inpatient and outpatient care. Health Aff (Millwood) 2008;27:1315–1327. [PMID: 18780917] [PubMed]
8. Sharma G, Fletcher KE, Zhang D, Kuo YF, Freeman JL, Goodwin JS. Continuity of outpatient and inpatient care by primary care physicians for hospitalized older adults. JAMA. 2009;301:1671–1680. [PMID: 19383958] [PMC free article] [PubMed]
9. Roy CL, Poon EG, Karson AS, Ladak-Merchant Z, Johnson RE, Maviglia SM, et al. Patient safety concerns arising from test results that return after hospital discharge. Ann Intern Med. 2005;143:121–128. [PMID: 16027454] [PubMed]
10. Hinami K, Farnan JM, Meltzer DO, Arora VM. Understanding communication during hospitalist service changes: a mixed methods study. J Hosp Med. 2009;4:535–540. [PMID: 20013853] [PubMed]
11. Kripalani S, LeFevre F, Phillips CO, Williams MV, Basaviah P, Baker DW. Deficits in communication and information transfer between hospital-based and primary care physicians: implications for patient safety and continuity of care. JAMA. 2007;297:831–841. [PMID: 17327525] [PubMed]
12. Sharma G, Kuo YF, Freeman JL, Zhang DD, Goodwin JS. Outpatient follow-up visit and 30-day emergency department visit and readmission in patients hospitalized for chronic obstructive pulmonary disease. Arch Intern Med. 2010;170:1664–1670. [PMID: 20937926] [PMC free article] [PubMed]
13. Moore C, Wisnivesky J, Williams S, McGinn T. Medical errors related to discontinuity of care from an inpatient to an outpatient setting. J Gen Intern Med. 2003;18:646–651. [PMID: 12911647] [PMC free article] [PubMed]
14. Coleman EA, Min SJ, Chomiak A, Kramer AM. Posthospital care transitions: patterns, complications, and risk identification. Health Serv Res. 2004;39:1449–1465. [PMID: 15333117] [PMC free article] [PubMed]
15. Forster AJ, Murff HJ, Peterson JF, Gandhi TK, Bates DW. Adverse drug events occurring following hospital discharge. J Gen Intern Med. 2005;20:317–323. [PMID: 15857487] [PMC free article] [PubMed]
16. Sharma G, Freeman J, Zhang D, Goodwin JS. Continuity of care and intensive care unit use at the end of life. Arch Intern Med. 2009;169:81–86. [PMID: 19139328] [PMC free article] [PubMed]
17. Fried TR, Bradley EH, Towle VR, Allore H. Understanding the treatment preferences of seriously ill patients. N Engl J Med. 2002;346:1061–1066. [PMID: 11932474] [PubMed]
18. Kuo YF, Goodwin JS. Effect of hospitalists on length of stay in the medicare population: variation according to hospital and patient characteristics. J Am Geriatr Soc. 2010;58:1649–1657. [PMID: 20863324] [PMC free article] [PubMed]
19. Shah BR, Hux JE, Laupacis A, Zinman B, Cauch-Dudek K, Booth GL. Administrative data algorithms can describe ambulatory physician utilization. Health Serv Res. 2007;42:1783–1796. [PMID: 17610448] [PMC free article] [PubMed]
20. Koroukian SM, Xu F, Murray P. Ability of Medicare claims data to identify nursing home patients: a validation study. Med Care. 2008;46:1184–1187. [PMID: 18953230] [PMC free article] [PubMed]
21. Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data. Med Care. 1998;36:8–27. [PMID: 9431328] [PubMed]
22. Donini-Lenhoff F, editor. Guidebook for GME Program Directors. Chicago: American Med Assoc; 2005. p. 59.
23. ZIP code crosswalk files: HRR boundaries. The Dartmouth Atlas of Health Care. Accessed at www.dartmouthatlas.org/Downloads.aspx on 5 May 2010.
24. McWilliams JM, Meara E, Zaslavsky AM, Ayanian JZ. Medicare spending for previously uninsured adults. Ann Intern Med. 2009;151:757–766. [PMID: 19949141] [PubMed]
25. Yabroff KR, Lamont EB, Mariotto A, Warren JL, Topor M, Meekins A, et al. Cost of care for elderly cancer patients in the United States. J Natl Cancer Inst. 2008;100:630–641. [PMID: 18445825] [PubMed]
26. D’Agostino RB., Jr Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med. 1998;17:2265–2281. [PMID: 9802183] [PubMed]
27. Griswold ME, Localio AR, Mulrow C. Propensity score adjustment with multilevel data: setting your sites on decreasing selection bias [Editorial] Ann Intern Med. 2010;152:393–395. [PMID: 20231571] [PubMed]
28. Rabe-Hesketh S, Skrondal A. Multilevel and Longitudinal Modeling Using STATA. College Station, TX: StataCorp; 2005.
29. Finkelstein EA, Trogdon JG, Cohen JW, Dietz W. Annual medical spending attributable to obesity: payer-and service-specific estimates. Health Aff (Millwood) 2009;28:w822–w831. [PMID: 19635784] [PubMed]
30. Mullahy J. Much ado about two: reconsidering retransformation and the two-part model in health econometrics. J Health Econ. 1998;17:247–281. [PMID: 10180918] [PubMed]
31. Lin DY, Psaty BM, Kronmal RA. Assessing the sensitivity of regression results to unmeasured confounders in observational studies. Biometrics. 1998;54:948–963. [PMID: 9750244] [PubMed]
32. Lindenauer PK, Rothberg MB, Pekow PS, Kenwood C, Benjamin EM, Auerbach AD. Outcomes of care by hospitalists, general internists, and family physicians. N Engl J Med. 2007;357:2589–2600. [PMID: 18094379] [PubMed]
33. Centers for Medicare & Medicaid Services. 100% MEDPAR inpatient hospital fiscal year 2008. Short stay inpatient by state. Assessed at www.cms.gov/MedicareFeeforSvcPartsAB/Downloads/DRGState08.pdf on 12 May 2010.
34. Go JT, Vaughan-Sarrazin M, Auerbach A, Schnipper J, Wetterneck TB, Gonzalez D, et al. Do hospitalists affect clinical outcomes and efficiency for patients with acute upper gastrointestinal hemorrhage (UGIH)? J Hosp Med. 2010;5:133–139. [PMID: 20235292] [PMC free article] [PubMed]
35. Howrey B, Kuo YF, Goodwin JS. Association of care by hospitalists on discharge destination and 30 day outcomes following acute ischemic stroke. Med Care. [Forthcoming] [PMC free article] [PubMed]
36. Anderson GF, Weller WE. Methods of reducing the financial risk of physicians under capitation. Arch Fam Med. 1999;8:149–155. [PMID: 10101986] [PubMed]
37. Frakt AB, Pizer SD, Schmitz RJ, Mattke S. Voluntary partial capitation: the Community Nursing Organization Medicare demonstration. Health Care Financ Rev. 2005;26:21–37. [PMID: 17288066] [PubMed]
38. Report to Congressional Committees GAO-05-119. Washington, DC: U.S. Government Accountability Office; 2005. Medicare Physician Fees: Geographic Adjustment Indices Are Valid in Design, But Data and Methods Need Refinement. Accessed at www.gao.gov/new.items/d05119.pdf on 16 August 2010.