Predictive model for mild cognitive impairment risk in community-dwelling older adults with depressive symptoms based on explainable machine learning: a cross-sectional study

Abstract Background: Mild cognitive impairment (MCI) represents a high-risk prodromal stage of Alzheimer’s disease, and its prevalence and conversion risk are higher among community-dwelling older adults presenting with depressive symptoms. Currently, there is a lack of interpretable and easily implementable risk stratification tools specifically tailored for this “high-risk within high-risk” population. Methods: This cross-sectional study enrolled 297 community-dwelling older adults with depressive symptoms (GDS-15 ≥ 5). Guided by the biopsychosocial model, 18 candidate predictors were collected. Core predictors were selected using LASSO regression and stepwise logistic regression, and four predictive models were developed: logistic regression, support vector machine, random forest, and XGBoost. Model performance was evaluated using five-fold cross-validation and an independent test set, and model interpretability was addressed using SHAP for both global and local explanations. Results: Five core predictive factors were ultimately identified: sleep quality, educational level, activities of daily living (ADL) ability, social support, and anxiety status. The XGBoost model exhibited the best performance on the test set , and both calibration and decision curve analyses indicated favorable clinical net benefit. SHAP analysis further revealed that poor sleep, ADL dependency, and anxiety substantially increased the risk of MCI, whereas higher education and strong social support acted as protective factors. Conclusion： We developed an interpretable MCI risk stratification model based on XGBoost and SHAP, capable of effectively identifying individuals at high risk of MCI among community-dwelling older adults with depression. The model demonstrates strong discriminative performance and clinical interpretability, and has the potential to provide community healthcare workers with an intuitive and reliable tool for early screening and intervention decision support.