Duke 2008 | ASHE | Equity and Efficiency in Health and Healthcare

Introduction Health outcomes research often involves comparing data across hospitals. Comparisons are frequently confounded by a number of unobserved factors including patient preferences, provider skills/preferences and the hospital's technology. The effect of these factors is well-known and not new. Over three decades ago, Wennberg and colleagues showed how health service use varied by geographic area. Since then, new fields of research, including the health disparities literature, have emerged, but the results are still frequently confounded by these factors.

Methods The main innovation in this paper comes from a comparison of all inpatient discharges at a large, urban hospital where patients were randomly assigned to physician teams affiliated with one of two academic institutions: one that is among the top residency programs in the U.S. and another that ranked outside of the top 50. All patients admitted to the hospital were randomized based on the last digit of the social security number (odd or even). Patients were exempt from randomization in some tertiary care programs (e.g., stroke) where there was only one physician team. The universe of over 36,000 inpatient admissions from 1993-2006 is considered, along with subsamples of patients with a diagnosis of chronic heart failure (CHF), a gastro-intestinal bleed, and a myocardial infarction. We also ran models for stroke patients and at geographically proximal hospitals where randomization did not happen. We used a range of models including semi-log, general linear models (log link and gamma distributions) and logit models, to analyze costs, length of stay (LOS), 28-day readmission, 1-year readmission, and all cause mortality. Our key independent variable was an even-ending social security number, which meant that the patient was assigned to the higher-ranked residency program. In our models, we controlled for age, gender, disease severity (Charlson Index), marital status, and time, day, month, and year of admission. We also linked the patient's zip code to census data and controlled for the zip code's average income, education, race/ethnicity, density and age. All costs were inflated to 2006 using the Bureau of Labor Statistics general Consumer Price Index.

Results The results indicate that the randomization was consistently applied to all patients. The two groups were not statistically different by age, gender, disease severity, date of admission, month of admission, year of admission, marital status or zip code characteristics; out of 58 comparisons, only two were significant at an alpha of <0.05.

Semi-log and GLM models indicated that patients assigned to the higher ranked residency program had approximately 10-20% lower costs and shorter lengths of stay. Cost differences existed even after controlling for length of stay. No differences were found for stroke patients or at the hospitals were randomization did not occur. The largest differences were found for patients with CHF; when treated by residents at the higher ranked program, CHF patients had 26% lower costs and 29% shorter lengths of stay. Few differences in mortality and readmission rates were found. Ongoing research is examining why residents from the higher ranked program have fewer costs and short lengths of stay. Illumination of practice patterns that result in lower costs without compromising outcomes could have broad implications regarding health care policy implementation.