90. Assessing the Performance of Adjusted Clinical Groups (ACGs) in Explaining Variation in Utilization among Age and Gender Subgroups in the VA
C Rakovski, Center for Health Quality, Outcomes, and Economic Research; AK Rosen, Center for Health Quality, Outcomes, and Economic Research; S Loveland, Center for Health Quality, Outcomes, and Economic Research; JJ Anderson, Center for Health Quality, Outcomes, and Economic Research; D Berlowitz, Center for Health Quality, Outcomes, and Economic Research
Objectives: Previous research indicates that ACGs explain variation in utilization among veterans. Little work, however, has been done to examine how well ACGs characterize subgroups of veterans. Because ACGs were developed on an HMO population that is younger and more female than the VA population, we expected that ACGs would capture the case-mix of these groups more accurately than for other groups. We examined how well this diagnosis-based, case-mix measure performed in explaining utilization among age and gender subgroups of veterans.
Methods: We measured utilization as the total number of annualized service days (inpatient and outpatient days). We selected a 40% random sample of all veterans using healthcare services in FY 1997, excluding those who had only telephone or dental encounters. This resulted in a sample of 1,046,803 people who received acute, long-term, or outpatient care during FY 1997. For each age/gender group, we compared the mean absolute prediction error (MAPE), defined as the average absolute difference between the observed (O) and expected (E) utilization, under a basic age/gender model and an ACG model. MAPE is used to evaluate model performance; the model with the smaller average difference between O and E has the better performance. The ACG model was estimated by a least squares model that consisted of age, gender, and 32 ADGs (clinical groupings from the ACG model). This model was validated on a 20% random sample of veteran users (N=524,461). We compared the performance of the two models within 12 age/gender categories: 18-34, 35-44, 45-54, 55-64, 65-74, and 75+.
Results: The ACG model resulted in an approximately 20% smaller MAPE than the age/gender model in 11 of the 12 categories. The MAPE from the ACG model was proportional to utilization (i.e., smaller for young age groups who had little utilization, and increasingly higher in older age groups). The ACG model performed consistently well in both men and women and in each of the 5 oldest age categories. As age increased, performance of the ACG model also improved with respect to the age/gender model. There were no differences in MAPEs with respect to gender, except among elderly veterans (aged 65+), where the ACG model fit better for men than women. However, older women used substantially more service days than older men.
Conclusions: As age increases, ACGs become increasingly more important for explaining utilization. ACGs did not perform better in subgroups of young adults and women as had been expected; rather their performance was consistent in all age/gender categories. Women and men used similar amounts of services, except those 65+, where women used substantially more services.
Impact: ACGs may be used to characterize the veteran population as well as age/gender subgroups with respect to utilization. Since older veterans use a higher than average number of services, capturing the case-mix of this group is important. VA researchers and policymakers should be aware that, although women 65 and older are a relatively small subgroup, they are consuming a large amount of services at a higher rate than other subgroups.