Regression methods in the empiric analysis of health care data.

Abstract

The aim of this paper is to provide health care decision makers with a conceptual foundation for regression analysis by describing the principles of correlation, regression, and residual assessment. Researchers are often faced with the need to describe quantitatively the relationships between outcomes and predictors , with the objective of explaining trends, testing hypotheses, or developing models for forecasting. Regression models are able to incorporate complex mathematical functions and operands (the variables that are manipulated) to best describe the associations between sets of variables. Unlike many other statistical techniques, regression allows for the inclusion of variables that may control for confounding phenomena or risk factors. For robust analyses to be conducted, however, the assumptions of regression must be understood and researchers must be aware of diagnostic tests and the appropriate procedures that may be used to correct for violations in model assumptions. Despite the complexities and intricacies that can exist in regression , this statistical technique may be applied to a wide range of studies in managed care settings. Given the increased availability of data in administrative databases, the application of these procedures to pharmacoeconomics and outcomes assessments may result in more varied and useful scientific investigations and provide a more solid foundation for health care decision making.