Aging
Who Do These Estimates Really Represent? An Application of Pre-Enrollment Inverse Probability Selection Weighting to a Late-Life Aging Cohort Sirena Gutierrez* Sirena Gutierrez Gutierrez Gutierrez Gutierrez Gutierrez Gutierrez Gutierrez Gutierrez Gutierrez Gutierrez Gutierrez University of California San Francisco
Studies of older adults are shaped by multiple socially patterned selection processes prior to enrollment, including survival, retention within health systems, and self-selection into study participation. These processes challenge the estimation and interpretation of causal effects of earlier life determinants of healthy aging. Inverse probability weighting (IPW) is commonly used to address post-enrollment attrition, but studies rarely account for pre-enrollment selection due to limited pre-enrollment data. We developed a composite IPW for the Kaiser Healthy Aging and Different Life Experiences (KHANDLE) study, a multi-ethnic cohort of long-term Kaiser Permanente Northern California (KPNC) members, to account for selection into the cohort along 3 selection processes: dementia-free survival, health system membership, and enrollment. KHANDLE participants were current KPNC members on 1/1/2017, born before 1952, who participated in at least one Kaiser Permanente Multiphasic Health Checkup (MHC) exam between 1964 and 1985. The target population was all MHC participants born before 1952, regardless of survival or membership status as of 1/1/2017. Using midlife health covariates measured in the MHCs and sociodemographic covariates measured from electronic health records, we estimated: 1) dementia-free survival to 1/1/2017; 2) KPNC membership on 1/1/2017 conditional on survival; and 3) KHANDLE participation conditional on survival and membership. We will compare estimated effects of midlife hypertension on late-life dementia risk in KHANDLE participants and after extending estimates to the MHC target population. Our framework is applicable to other studies of lifecourse determinants of health among older adults by enabling correction for survival bias, selective enrollment, and unmeasured structural processes that may otherwise bias late-life estimates.