2057. Comparing the Clinical Validity of Diagnosis-based Risk Adjusters
Supicha Sookanan, MPH, Houston Center for Quality of Care and Utilization Studies, Houston VAMC, K Pietz, Houston Center for Quality of Care and Utilization Studies, Houston VAMC and Baylor College of Medicine, LD Woodard, Houston Center for Quality of Care and Utilization Studies, Houston VAMC, H Nelson, Houston Center for Quality of Care and Utilization Studies, Houston VAMC, M Byrne, Center for Bioethics and Health Law and Department of Medicine, University of Pittsburg, LA Petersen, Houston Center for Quality of Care and Utilization Studies, Houston VAMC and Baylor College of Medicine

Objectives: Many possible methods of risk adjustment exist, but there is a dearth of comparative data on their performance. To compare the face validity of two widely used methods (Diagnostic Cost Groups [DCGs] and Adjusted Clinical Groups [ACGs]), we assessed predictive ability for three different clinical outcomes using VA data.

Methods: We studied patients who used VA medical services in fiscal year (FY) 2001 (n=3,070,957) and assigned both a DCG and an ACG to each. We used logistic regression analyses to compare predictive ability for death, rehabilitation stays, and long-term care (LTC) hospitalization for age-gender models, DCG models, and ACG models.

Results: Patients who died were more likely to be in the highest DCG categories and the most severe ACGs, indicating higher severity of illness. The lowest c-statistics were obtained when using age-gender models to predict LTC or rehabilitation stays (c-statistic 0.59 for each). In contrast, the c-statistics when DCGs were used to predict LTC and rehabilitation stays rose to 0.88 and 0.89, respectively, indicating good prediction. The c-statistics when ACGs were used to predict LTC and rehabilitation stays was significantly lower, at 0.78. Surprisingly, the age-gender model predicted death slightly more accurately than the ACG model (c-statistic of 0.71 vs. 0.69, respectively).

Conclusions: In this comparative analysis using VA data, DCG models were superior to ACG models in predicting clinical outcomes.

Impact: The face validity of risk adjusters developed in non-VA databases for VA data is acceptable. DCGs appear to be preferable to ACGs for the VA population.