NEW OBSERVATIONAL DATA ANALYSIS METHODS FOR COMPARATIVE EFFECTIVENESS RESEARCH

Washington University, Saint Louis MO

Investigators

Abstract

DESCRIPTION (provided by applicant): Comparative effectiveness research (CER) is designed to identify healthcare interventions having the best patient outcomes to direct patients to receive the best treatment and to direct our healthcare dollars to where they will be most productive. When comparing observational data to determine the best intervention, CER requires that we apply risk or case-mix adjustment methods before examining outcomes of care. For example, to compare survival in treatment or hospital for inpatient acute myocardial infarction (AMI) patients using the proportion surviving may be misleading if the severity of disease is significantly different across interventions or hospital. To make comparisons valid, risk adjustment must balance patient factors, such as disease severity and co-morbidities, which result in different likelihood of death. A standard approach to risk adjustment is to use measures of "observed-to-expected" rates, where expected outcome for patients are estimated by an existing, often unknown and proprietary, regression model previously fit to a standard or reference population of patient data said to be representative of all patients. The observed outcome is obtained from the patient's discharge data. The goal of the risk adjustment is to determine if an intervention (or provider) on average shows better, worse, or the same observed outcomes compared to expected outcomes. We propose to develop and release an open-source HealthCare Rankings (HCR) case-mix adjustment software package combining methods from observational data analysis, operations research, statistics, and mathematics that have not been applied in combination previously in CER and health services research. The HCR algorithm ranks two or more interventions or providers simultaneously based on direct comparison of patient-level data. This algorithm avoids the need to have a reference database for observed-to-expected comparisons. This proposal is a joint effort of investigators in the Washington University School of Medicine (WUSM) Dept. of Medicine's Biostatistical Consulting Center and the BJC HealthCare Center for Clinical Excellence (CCE). There are 11 hospitals in the BJC network with a comprehensive informatics system of patient level clinical and administrative data available for developing and validating the HCR algorithm. PUBLIC HEALTH RELEVANCE: The goal of this project is to develop and validate novel mathematical methods from operations research and voting theory to perform more accurate comparisons of outcomes and performance among health care interventions and providers. The importance of this project is that if successful there will be new data analysis tools for directing patients to the best treatment and providers for their care based on their level of disease severity and other patient characteristics, and for directing health care dollars to the most cost effective options.