Adjustment for Selective Attrition

To account for potential sex differences in outcome-related dropouts (i.e., study dropouts due to impaired cognition) we used inverse probability weighting (IPW) in our models (Rouanet et al., 2021). We calculated probabilities for study participation at the two follow-ups with logistic regression models. These logistic regression models comprise age, sex, sociodemographic variables, distance to study center and variables describing the health condition in prior surveys, including the most recent cognitive global score. The inverse values of these probabilities were used as weights in the mixed models described above. Further information on IPW can be found here (Seaman and White, 2013; Hernán and Robins, 2020). As sensitivity analyses, we re-analyzed all models using the same cohort without IPW.