Brain MRI markers and dropout in a longitudinal study of cognitive aging: The Three-City Dijon Study

Objective: Longitudinal studies of dementia rely on the assumption that individuals who drop out are comparable to those who remain in the study, adjusting for measured covariates. Existing methods to handle dropouts account for differences based on past health and cognitive measures. We assess whether such adjustments fully account for differences in future dementia risk. Methods: Among Three-City Study participants in Dijon, France, with 1 (n = 1,633) or 2 (n = 1,168) brain MRI scans, we tested whether white matter lesion volume (WMLV), hippocampal volume, or brain CSF volume predicted dropout (“unable to contact” or “refused interview”) in repeated-measures logistic regression with up to 4 follow-ups (average 3.5 waves). Using linear regression, we also estimated differences in MRI volumes and MRI changes by subsequent dropout status and estimated plausible ranges for selective attrition bias based on these associations. Models were adjusted for demographic, health, and cognitive score covariates. Results: Baseline greater WMLV predicted increased odds of dropping out (adjusted odds ratio = 1.71; 95% confidence interval [CI] 1.20−2.43). Among participants with 2 MRI scans, individuals who subsequently dropped out had significantly worse declines in hippocampal volume (−0.30 SD difference; 95% CI −0.43 to −0.17) between the first and second MRI scans. Conclusions: Higher future dementia risk, indicated by worse past brain MRI findings, predicted future study dropout. Adjustment for selective attrition, based on MRI markers when available, may help reduce bias in estimates of dementia incidence and improve research on dementia risk factors. MRI findings may also help prospectively identify cohort members at elevated risk of attrition.