Preprint · Nov 13, 2025
Brain Aging–Based Subtypes of Major Depressive Disorder (MDD)
By Lim J, Dunlop K
This secondary analysis tests whether subtyping through structural MRI-based brain‑age models yield stable, clinically meaningful dimensions linking brain aging to major depressive disorder (MDD) symptoms and treatment response. We will analyze 3,240 adults (MDD + HC) aggregated across 11 multi‑site datasets in the COORDINATE‑MDD consortium. Seven datasets (n = 1,376) are antidepressant trials (escitalopram, duloxetine, sertraline). T1‑weighted structural MRI features from HC data (extracted via the MUSE Pipeline and harmonized) were used to train an XGBoost brain‑age model, and a previous analysis computed brain‑predicted age difference (brain‑PAD) and SHAP values indexing anatomical drivers of atypical aging in the MDD and HC samples. We will leverage these SHAP values in the proposed analyses. Z‑scored SHAP and clinical symptom measures (HDRS, QIDS‑16, or MADRS‑10) will be used to train and optimze a regularized CCA within a nested cross‑validation framework to derive stable brain–symptom dimensions. We will then apply k‑means clustering (evaluated using the Calinski–Harabasz and adjusted mutual information) to define reproducible subtypes and test differences in brain‑PAD, symptom profiles, and antidepressant outcomes. Inclusion requires HDRS ≥ 14 (or scale‑equivalent); participants with missing imaging are excluded. Inference will use 1,000‑permutation testing corrected for multiple comparisons where appropriate using the BH‑FDR. Sensitivity analyses will include leave‑one‑site‑out and covariate checks for age/sex and site effects. Data collection is complete and the authors report no prior analyses of these data.