Dissertation Title:  "Improving Comparisons in Health Services Research"

Comparisons of service use and outcomes data across providers, across regions and over time are made to assess performance and to guide resource allocation in health care systems. These data are, however, inherently observational and subject to many sources of variability. To be informative and useful to policymakers, comparative analyses must recognize and account for variability in underlying data. Methods that achieve these goals improve the precision of estimates, account for important covariates, and assess the impact of trends. This dissertation analyzed three prominent observational data sets, two for cardiac care and one for mental health care, to investigate methods for improving the characterization and comparison of health service utilization and outcomes.

In Chapter One, hierarchical models were extended and applied to examine outcomes for cardiac surgeons and mental health networks over time. Provider-specific outcomes were compared based on level, rate of change over time, and a combination of both. Longitudinal information supplemented cross-sectional analysis by revealing whether and how much more rapidly the outcomes of an individual provider were increasing (or decreasing) relative to other providers. A multivariate approach helped to describe and combine multiple outcomes.

In Chapter Two, risk-adjusted, in-hospital patient mortality rates for cardiac surgeons were compared using three different analytical models to determine differences in the detection of outliers. Compared with pooling data over three years, a model that accounted for variation within and between surgeons identified fewer surgeons as high or low outliers. Similar results were obtained for a model that incorporated additional longitudinal variation.

In Chapter Three, the extent to which factors beyond characteristics of the patient, such as discharging hospital attributes and state factors, contributed to variations in post-acute service use were examined in a cohort of elderly Medicare patients after acute myocardial infarction. Using detailed data sets and a hierarchical model, patient severity of illness at hospital discharge, for-profit ownership of the hospital, and hospital provision of home health services were identified as important predictors of post-acute service use. After adjusting for many patient and hospital characteristics, however, variation in post-acute service use remained across states.