Trajectories of cognitive performance in late middle‐aged adults: Bayesian random change point mixed model analysis in WRAP

Abstract

AbstractBackgroundPrevious studies indicate that rates of cognitive decline in Alzheimer’s disease (AD) and other dementias may accelerate significantly as brain pathology increases. The objectives of this study included to: 1) characterize changepoints and related slope parameters across four cognitive composites in a longitudinal cohort study; and 2) examine how these parameters varied across predictors of interest.MethodsDementia‐free participants at baseline from the Wisconsin Registry for Alzheimer’s Prevention (WRAP) with > = 3 cognitive assessments were eligible (n≈1071; mean(sd) baseline age = 58.4(6.4)). We used newly‐developed Bayesian random change point mixed models (BRCPMM) to estimate fixed (group‐level) and random (person‐level) change points (CPs), slopes pre‐ and post‐CP, and intercepts at CP for four cognitive composites: three‐test preclinical Alzheimer’s cognitive composite (PACC3; primary outcome), immediate learning, delayed learning, and executive function. We used non‐parametric statistics to test whether PACC3 estimates differed by: last cognitive status observed (cognitively unimpaired‐stable or ‐declining (CU‐S, CU‐D), MCI, and dementia); sex/gender, and APOE‐e4 carrier status.ResultsPosterior median fixed and random effect estimates (corresponding 95% credible intervals from 20000 BRCPMM runs) are summarized in Table 1 by outcome. The PACC3 group‐level CP was earliest at ∼75.56 [72.98,78.44] years old with 95% of individuals hitting CP between 59.69 and 91.86. Timing of CP was significantly associated with last observed cognitive status (Table 2); those 37 (3.46%) progressing to MCI/dementia had estimated CP’s before or during study participation while those remaining unimpaired had estimated CP’s after their last visit. All random effects (person‐level) PACC3 estimates differed by worst cognitive status (Figure_1) and sex/gender (Figure_2). For example, while the CU‐D and MCI groups had similar slopes before CP, CP occurred earlier, estimated PACC3 at CP was lower, and post‐CP slope was worse in the MCI group compared to CU‐D (Figure_1). APOE‐e4 random‐effects estimates differed by CP and post‐CP slopes (Figure 3).ConclusionsThe timing and rates of change in different cognitive composites differ by cognitive status, gender, and APOE‐e4 status. Predicting individual cognition trajectories and change points may provide an opportunity for early intervention by diagnosing high‐risk subjects at the right time for treatment and/or enrollment in a clinical trial.