Aspects of hierarchical regression modeling in health services and outcomes research

Abstract

From a statistical perspective, the goals of the analyses of health care data require the estimation of: covariate effects; cluster-specific measures of utilization, costs, outcomes; and systematic and random components of variation. These estimates need to account for within cluster correlations and to accommodate substantial variations in cluster size. The growing literature on hierarchical regression modeling (HRM) and its applications to health services and outcomes research includes work that is relevant to a broad set of subject-matter and methodologic questions. We focus on two illustrative examples: the HRM approach to profiling of medical care providers; and the use of HRM in the estimation and proper interpretation of the effect of covariates.