Utilization of Custodial Nursing Homes by Elderplan Social HMO Members: Predicting Expected Rates with the Medicare Health Outcomes Survey Data

Rigorous efforts to see if home and community-based care programs substantially reduce nursing home utilization among the elderly have had conflicting results. This research tested the hypothesis that the community long-term care benefit, which was available to Social HMO enrollees that are nursing home certifiable, has led to lower custodial nursing home residence.

We compare predicted vs. actual nursing home admission rates for Social HMO members based on the HOS analytic sample provided to our agency with the published final report. To predict risk, we apply an HOS-based, Cox proportional-hazard model developed by Yael Harris and includes age, income, ethnicity, marital status, education, Medicaid status, baseline physical health status, baseline functional status, and baseline mental health status to predict expected nursing home rates over a study period of 3.5 years. However, the analytic sample of the individual level data file excludes 96 individuals, and the excluded subset of individuals includes disenrollers and decedents, in both cases a likely source of bias in our results. Effectively, disenrollers are excluded almost totally from individual HOS outcomes and after disenrollment from all-plan outcomes. To ensure the validity of the initial results of analysis, we requested the use of the cumulative dataset, including non-respondents. This allows us to reduce selection bias in our HOS-based analysis, and the possibility of confounding factors, such as plan disenrollment.