Activity-function transitions and interpretable machine learning for predicting incident depressive symptoms among...

Ad--verse childhood experiences (ACEs) are associated with increased risk of depressive symptoms in later life. This study aimed to develop and externally validate an interpretable machine-learning model for predicting incident depressive symptoms among ACE-exposed middle-aged and older adults.

Data were obtained from the English Longitudinal Study of Aging (ELSA) and the Health and Retirement Study (HRS). Participants with baseline depressive symptoms were excluded. ELSA was used for model development and HRS for external validation. LASSO regression was applied for predictor selection, followed by comparison of multiple machine-learning algorithms. Pre-outcome activity-function transition profiles were further examined in relation to observed incident depressive symptoms.

LASSO identified stable predictors mainly involving physical function, physical activity, chronic conditions, and sociodemographic characteristics. Among the evaluated models, random forest demonstrated the best performance, with an AUC of 0.701 in ELSA and 0.693 in HRS external validation. SHAP analysis identified walking time, grip strength, falls, arthritis, and physical activity as important contributors to prediction. Participants with unfavorable activity-function status at both pre-outcome assessments showed higher odds of incident depressive symptoms in both cohorts.

An interpretable machine-learning framework integrating physical performance, physical activity, and health-related factors achieved moderate and externally validated prediction of incident depressive symptoms among ACE-exposed adults. Pre-outcome activity-function transition profiles provided additional information on subsequent depressive-symptom risk. Given the moderate discrimination and limited case-detection ability, the framework should be regarded as exploratory and requires further validation and refinement before any clinical application.