Biased selection and regression toward the mean in three Medicare HMO demonstrations: a survival analysis of enrollees and disenrollees

Mortality of aged Medicare enrollees in three demonstration health maintenance organizations (HMOs) was compared with that of three cohorts of local fee-for-service (FFS) beneficiaries as a measure of group differences in health status. Mortality of the HMO and FFS cohorts was tracked via life-table analysis for 6 years after enrollment to measure the persistence of health status differences existing at enrollment. After adjustment for age, sex, Medicaid-eligibility, and institutional status, HMO enrollee mortality was lower than FFS at all three plans in the first year after enrollment. Enrollee mortality at two plans increased to near FFS levels in year 2 and remained relatively stable thereafter. Enrollee mortality at the third plan increased toward FFS levels more gradually and was significantly below FFS levels for the first 5 years of follow-up. Mortality of disenrollees in the 2 years after disenrollment was significantly higher than that of the continuously enrolled in one plan and marginally significantly higher in a second. These findings suggest a pattern of favorable selection at enrollment, followed by different rates of decay in the favorable health status of enrollees. Other health status measures may exhibit different patterns, however. Selection effects may cause payments based on Medicare's AAPCC to be too high or low, in some cases for several years after enrollment.