Measuring longitudinal cognition: Individual tests versus composites

Longitudinal cohort studies of cognitive aging must confront several sources of within-person variability in scores. In this article, we compare several neuropsychological measures in terms of longitudinal error variance and relationships with biomarker-assessed brain amyloidosis (Aβ). Analyses used data from the Wisconsin Registry for Alzheimer's Prevention. We quantified within-person longitudinal variability and age-related trajectories for several global and domain-specific composites and their constituent scores. For a subset with cerebrospinal fluid or amyloid positron emission tomography measures, we examined how Aβ modified cognitive trajectories. Global and theoretically derived composites exhibited lower intraindividual variability and stronger age × Aβ interactions than did empirically derived composites or raw scores from single tests. For example, the theoretical executive function outperformed other executive function scores on both metrics. These results reinforce the need for careful selection of cognitive outcomes in study design, and support the emerging consensus favoring composites over single-test measures. •Identifying early cognitive change requires tests with low error variance.•In a middle-aged sample, composites were less noisy than single tests.•Global and theory-driven composites outperformed data-driven composites.