Mood Dynamics and Response to Group Music Therapy: Findings from MUSED

Background: Group music therapy is increasingly used as a tool to supplement care for depression, yet we know less about who benefits most. A previously published study named MUSED randomised trial enrolled women with depression to receive music therapy (MT) plus treatment-as-usual (TAU) or TAU alone and embedded ecological momentary assessment (EMA) of daily-life mood. Methods: In this study, a secondary analysis of the processed Randomised Controlled Trial dataset MUSED was conducted. Outcomes were a post-treatment Back Depression Inventory (BDI-II) and a binary responder classification (≥50% BDI reduction) within the MT arm. Analyses included (i) ANCOVA adjusting for baseline BDI, baseline EMA mean, study cohort, and group; (ii) 5×5 cross-validated models (elastic-net, random forest, Gradient Boosting (XGBoost)) for classification (MT only) and regression (across all participants); and (iii) model interpretability using SHAP rankings, partial-dependence plots, and simplified logistic regression.Results: Post-treatment BDI was analysed for 87 participants; in the MT arm (n=52), 38.5% were responders. After adjustment, MT was associated with a 3-point lower BDI versus TAU (with adjusted difference −2.87). Higher baseline BDI and worse baseline EMA mood predicted higher (worse) post-treatment BDI; the ANCOVA explained ~63% of variance. In classification (MT only), random forest performed best (ROC AUC 74±3%; PR AUC 85±2%) with high sensitivity (83%) and modest specificity (at around 40%). For regression (all participants), random forest achieved RMSE=8.9, MAE=7.5, and R²=0.43. Across model families, SHAP analyses consistently ranked baseline EMA mean (and within-day EMA slope) as top predictors; elastic-net coefficients aligned in direction (People who began treatment with worse mood tended to have worse outcomes, while receiving music therapy was linked to lower (better) predicted BDI scores).Conclusions: MT+TAU reduced depressive symptoms beyond TAU, and everyday mood at intake, captured through EMA, was the most informative signal for both who responds and how well patients do. These baseline-only models can support care planning by prioritising likely beneficiaries, pinpointing those needing extra support, and contextualising MT’s added value alongside usual care.