Brain age is not a significant predictor of relapse risk in late-life depression.

Introduction: Late-life depression (LLD) has been associated cross-sectionally with lower brain structural volumes and accelerated brain aging compared to healthy controls (HC). There are few longitudinal studies on the neurobiological predictors of recurrence in LLD. We tested a machine learning (ML) brain age model and its prospective association with LLD recurrence risk.

Methods: We recruited individuals with LLD (n=102) and HC (n=43) into a multi-site 2-yr longitudinal study. Individuals with LLD were enrolled within 4 months of remission. Remitted LLD participants underwent baseline neuroimaging and longitudinal clinical follow-up. Over 2 years, 43 LLD participants relapsed (REL) and 59 stayed in remission (REM). We used a previously developed ML brain age algorithm to compute brain age at baseline and we evaluated brain age group differences (HC vs. LLD and HC vs. REM vs. REL). We conducted a Cox proportional hazards model to evaluate whether baseline brain age predicted time to relapse.

Results: We found that brain age did not significantly differ between HC and LLD as well as HC, REM, and REL groups. Brain age did not significantly predict time to relapse.

Discussion: In contrast to our hypothesis, we found that brain age did not differ between non-depressed controls and individuals with remitted LLD, and brain age was not associated with subsequent recurrence. This is in contrast to existing literature which has identified baseline brain age differences in late life but in line with work that shows no differences between those who do and do not relapse on gross structural measures.