Hierarchical Modeling Using HCUP (Healthcare Cost and Utilization Project) Data. HCUP Methods Series

摘要

This document continues the HCUP Methods Series of reports, which features information on a broad array of methods as they relate to the HCUP databases and software tools. This series is designed to help HCUP data users work efficiently and effectively with HCUP data. This report is consistent with that goal. Hierarchical linear modeling (HLM) is a regression technique designed to deal with clustered or grouped data in which analytic units are naturally nested or grouped within other units of interest. For example, a physician's patients form a group nested within that physician. In analyzing outcomes from a sample of patients treated by a number of physicians, interest centers on the effects of both patient and physician characteristics. Since each group of patients is treated by a single physician, it is expected that those patients outcomes will be correlated, which violates one of the assumptions of standard regression methods. If this correlation is ignored, wrong inferences can result with respect to the effects of both patient and physician factors. HLM accounts for this within-group correlation to produce better inferences when the proper model is specified.