The problem this dissertation proposal addresses is the insufficient understanding among payers and providers alike of what predicts variation in resource use among ambulatory medical care visits. The general hypothesis of this proposal is that the predictive accuracy of the new ambulatory medical classification system, the Ambulatory Patient Groups (APGs), will be improved by incorporating additional patient characteristics into the system. The project will develop a risk adjusted utilization model for ambulatory medical care. This product can be thought of as an outpatient corollary to the inpatient RDRG system. Currently, the APGS, which are the most likely candidate for a Congressionally mandated outpatient Medicare prospective payment system, use only the principal diagnosis and occasionally, a pediatric/adult split to classify medical visits. Although this parsimony in classification is administratively desirable, it does not produce the most accurate prediction of resources used. Previous research has shown that secondary diagnosis, age, and visit status (new patient, new diagnosis) can be powerful additions to a case mix model. The technique of building and applying a risk-of-resource-use prediction model will proceed in three phases. First, the risk index will be built using clinically predictive variables. The risk of increased resource use will be estimated for each APG based upon attributes of the patient before receiving care. Second, the risk index will be subject to statistical and clinical validation, and the ability to create an across-APG unified risk model investigated. In the third step, the stability and utility of the model will be tested by examining the circumstances under which any disparities between observed and expected resource use are attenuated or exaggerated. For example, controlling for risk, differences in resource use by payor source, provider type, and setting will be examined. The proposed research will draw upon the risk index modeling techniques developed by Blumberg and applied extensively to inpatient data to create risk-adjusted resource use-of-care models. Data for this project will include a national sample of private physician offices and clinic data from a number of sites.
