Go backSearch Session number: 1028

Abstract title: Predicting Outpatient Pharmacy Cost using Diagnosis- and Pharmacy-Based Case-Mix Instruments

Author(s):
KL Sloan - VA Puget Sound Health Care System & Department of Psychiatry and Behavioral Sciences, University of Washington
CF Liu - VA Puget Sound Health Care System & the Department of Health Services, University of Washington
AE Sales - VA Puget Sound Health Care System & the Department of Health Services, University of Washington
P Fishman - Group Health Cooperative of Puget Sound
J Todd-Stenberg - VA Puget Sound Health Care System
ND Sharp - VA Puget Sound Health Care System
AK Rosen - Bedford Center for Health Quality, Outcomes and Economic Research
S Loveland - Bedford Center for Health Quality, Outcomes and Economic Research

Objectives: Case-mix instruments are commonly used to predict total cost. Little work, however, has been done to explore how well these measures predict pharmacy cost, a rapidly rising cost component that is potentially one of the more controllable components of total cost. We compare the ability of the pharmacy-based VA-adapted RxRisk (RxRisk-V) with two leading diagnosis-based case-mix measures, the Diagnostic Cost Groups/Hierarchical Co-existing Conditions (DCG/HCC) model and Ambulatory Diagnostic Groups (ADGs), to explain concurrent and prospective outpatient pharmacy cost.

Methods: We examined all Fiscal Year (FY) 1998 veteran users in the VHA Northwest Network (n=121,074). Outpatient pharmacy information was obtained from the VHA Northwest Network data warehouse, diagnostic data were obtained from the Austin Automation Center, and outpatient pharmacy cost was obtained from DSS. Concurrent and prospective models were fitted using FY 1998 case-mix classifications to predict FY 1998 and 1999 outpatient pharmacy cost, respectively. Age and gender were included all models. We obtained estimation and validation R-squareds, adjusted R-squareds, mean squared errors, mean absolute errors, predictive ratios and estimates of cost within quintiles of actual cost.

Results: The RxRisk-V model had better explanatory power than DCG/HCCs or ADGs in both the estimation and validation samples. Validation sample R-squareds for concurrent models were 0.185, 0.137 and 0.104, respectively. Prospective model validation R-squareds were 0.155 for the RxRisk-V, 0.113 for the DCG/HCC model, and 0.088 for ACGs. Compared to the other measures, predicted estimates were closer to actual estimates of cost using RxRisk-V in all quintiles except one.

Conclusions: RxRisk-V performed better than other models in explaining concurrent and prospective outpatient pharmacy cost as reflected in both goodness of fit as well as estimates within cost quintiles.

Impact statement: These results suggest that the RxRisk-V should be strongly considered for applications involving case-mix adjustment of outpatient pharmacy cost in either concurrent (e.g., provider profiling) or prospective (e.g., predicting next-year cost) models.