Examining the influence of changes in amyloid burden on both contemporaneous and subsequent cognitive decline: using a latent change score approach

Abstract

AbstractBackgroundA critical question for understanding the natural history of preclinical Alzheimer’s disease (AD) is the extent to which changes in pathology affect contemporaneous and/or subsequent (lagged) cognitive manifestations of the disease. The most common approach in the literature involves the extraction and correlation of time‐demarcated slopes from different variables. Our objective was to employ a powerful and flexible structural equation modelling approach, latent change score models (LCSM), to directly address this question while also adjusting for demographics, a medial temporal lobe (MTL) tau burden composite, and an AD‐relevant glucose metabolism composite.Method131 Harvard Aging Brain Study participants (Agemean:73(6), Female:62%) had complete data for eight time‐points of neuropsychological assessment (Preclinical Alzheimer’s Cognitive Composite; PACC) and three timepoints of global PiB‐PET. In a contemporaneous bivariate LCSM (Fig1), we tested the influence of both cross‐sectional PiB‐PET(α) and change in PiB‐PET(β) on changes in PACC in the following year. Within a bivariate lagged LCSM (Fig2), we examined the effect of both cross‐sectional PiB‐PET(α) and change in PiB‐PET(β) on changes in PACC two years later. Regression coefficients were specified to be invariant over time. Both models included assumptions of cross‐sectional PiB(λ) and PiB change(μ) to predict subsequent PiB changes, and the same for PACC. Baseline age, sex, education, APOEe4, an MTL tau (Flortaucipir)‐PET composite (amygdala/entorhinal/parahippocampal), and an FDG‐PET composite (hippocampal/inferior parietal/posterior cingulate) were included as covariates. All data were z‐scored (PACCmean:0.21(0.6), PiBmean:1.16(0.19)).ResultHigher cross‐sectional PiB‐PET was not significantly associated with contemporaneous cognitive decline (α;Table1), but PiB‐PET accumulation was significantly associated with faster cognitive decline in the following year (β;Table1). Within the lagged model, PiB‐PET accumulation, not cross‐sectional, was significantly associated with faster PACC decline two years later (β;Table1). The lagged model was the better fitting model and demonstrated much larger change‐on‐change effect sizes (Table1). The strongest covariate relationships were between age/education and PACCintercept, and APOEe4/tau‐PET burden with the global PiB‐PETintercept and PiB‐PETslope (Table1).ConclusionOur findings suggest that change in PiB‐PET is more strongly associated with subsequent changes in cognition rather than contemporaneous cognitive change. Further, this relationship exists above and beyond the influence of demographics, cross‐sectional MTL tau burden, and glucose metabolism.