OBJECTIVE: To test the substitution hypothesis, that community-based care reduces the probability of institutional placement for at-risk elderly. RESEARCH DESIGN: The closure of the Social Health Maintenance Organization (Social HMO) at HealthPartners (HP) in Minnesota in 1994 and the continuation of the Social HMO at Kaiser Permanente Northwest (KPNW) in Oregon/Washington comprised a "natural experiment." Using multinomial logistic regression analyses, we followed cohorts of Social HMO enrollees for up to 5 years, 1995 to 1999. To adjust for site effects and secular trends, we also followed age- and gender-matched Medicare-Tax Equity and Fiscal Responsibility Act (TEFRA) cohorts, enrolled in the same HMOs but not in the Social HMOs. SUBJECTS: All enrollees in the Social HMO for at least 4 months in 1993 and an age-gender matched sample of Medicare-TEFRA enrollees. To be included, individuals had to be alive and have a period out of an institution after January 1, 1995 (total n = 18,143). MEASURES: The primary data sources were the electronic databases at HP and KPNW. The main outcomes were long-term nursing home placement (90+ days) or mortality. Covariates were age, gender, a comorbidity index, and geographic site effect. RESULTS: Adjusting for variations in the 2 sites, we found no difference in probability of mortality between the 2 cohorts, but approximately a 40% increase in long-term institutional placement associated with the termination of the Social HMO at HealthPartners (odds ratio, 1.43; 95% confidence interval, 1.15-1.79). CONCLUSIONS: The Social HMO appears to help at-risk elderly postpone long-term nursing home placement.