2006 HSR&D National Meeting Abstract
3063 — Comparative Performance of Function-Based Risk Adjustment Systems for Stroke Outcomes in the VHA
Vogel WB (VA Rehabilitation Outcomes Research Center)
Berlowitz DR (Center for Health Outcomes, Quality, and Economics Research)
Tsilimingras D (Center for Health Outcomes, Quality, and Economics Research)
Hoenig HM (Duke University)
Young LJ (VA Rehabilitation Outcomes Research Center)
Cowper DC (VA Rehabilitation Outcomes Research Center)
Duncan PW (VA Rehabilitation Outcomes Research Center)
Wing KL (VA Rehabilitation Outcomes Research Center)
To evaluate the comparative performance of function-based risk adjustment systems in predicting three- and six-month rehospitalization and mortality in a sample of VHA stroke patients.
We used data from the VHA’s Integrated Stroke Outcomes Database (ISOD) to compare the performance of the Function Related Groups-Length of Stay (FRG-LOS), Function Related Groups-Discharge Motor Function (FRG-DMF), and Case Mix Groups (CMG) risk adjustment systems in predicting three- and six-month rehospitalization and mortality. Our analytic sample consisted of 1,742 VHA stroke patients from the FY2001 ISOD. We estimated logistic regression models of mortality and rehospitalization controlling for sociodemographics, service-connected status, admission source, and discharge location, along with variables from each risk adjustment system separately.
For mortality, the best sensitivity and specificity tradeoff was achieved by the FRG-DMF system (c=.83-.86). For rehospitalization, the best sensitivity and specificity tradeoff was achieved by the CMG system for three-month rehospitalization (c=.56) and by the FRG-LOS system for six-month rehospitalization (c=.55). Our logistic regressions also pass the Hosmer-Lemeshow goodness-of-fit test. Among the additional covariates, patient age, time spent in rehabilitation, and type of rehabilitation unit exhibited statistical significance across the various models.
No single risk-adjustment system performed best in predicting all outcomes. While the FRG-DMF system performed best for mortality, the other two systems outperformed the FRG-DMF system for predicting rehospitalization. In comparing predictive power across outcomes, all three systems yielded markedly better predictive power for mortality than for rehospitalization. None of these function-based risk adjustment systems have much predictive power when examining post-stroke rehospitalization, and more work is needed to improve risk adjustment for this outcome.
Using the FRG-DMF system for quality assurance and provider profiling of post-stroke mortality will improve the accuracy of these activities, thereby improving both the quality of VHA stroke care and the overall health of VHA stroke patients.