To model the relationship between CDR‐SB and each functional outcome, we selected participants without impairment in the outcome at the baseline assessment (i.e., different subsamples were used for each outcome). IADLs and BADLs were coded as dichotomous variables, and the four FAQ items (personal finances, driving, remembering meds/appointments, and meal preparation) were coded as ordinal variables. For each functional outcome, we fitted generalized linear mixed‐effects models to CDR‐SB with participant‐level random effects. The glmer function of the lme4 R package 28 was used to fit dichotomous outcomes. We ran ordinal logistic regression for the ordinal outcomes using the mixed_model function of the GLMMadaptive R package. 29 For independent IADLs and BADLs, we used the regression models to identify the first CDR‐SB where 50% of participants were predicted to be dependent (referred to as cutoffs for IADLs and BADLs).
