Approaches to predictive modeling.

Abstract

A four-component clinical model for process improvement is presented: (1) patient-related risk factors, (2) clinical processes ordered by the attending physician, (3) the hospital's execution of the physician's plan, and (4) the patient's outcome, or outcomes, resulting from the first three factors. The goal of risk adjustment in the analysis of quality of care is to account for the contribution of patient-related risk factors, so that the patient's outcome can be used as an indicator of the care ordered by the physician and executed by the hospital. Risk adjustment is usually accomplished by comparing the patient's predicted outcome, based on the patient's risk factors, to the observed outcome. Historically, three approaches to the development of prediction models have been used: (1) selection and weighting of risk factors by expert opinion, (2) univariate analyses, and (3) multivariate analyses. Future prediction models will be based on neural network techniques or cluster analysis. As these prediction models have evolved, there has been a steady increase in their predictive power.