Development and validation of an explainable machine learning model for predicting depressive symptoms in older adults with edentulism: a study based on the CHARLS cohort

Depression is highly prevalent among older adults with edentulism, yet no dedicated prediction tool exists for this population. This study aimed to develop and temporally validate an interpretable machine learning model for estimating the probability of concurrent clinically significant depressive symptoms in older adults with edentulism. We analyzed data from the 2015 ( N = 2,078) and 2018 ( N = 2,311) waves of the China Health and Retirement Longitudinal Study (CHARLS). Three complementary feature selection methods (univariate logistic regression, least absolute shrinkage and selection operator regression, and random forest importance ranking) were applied to identify robust predictors. Five machine learning algorithms were trained and compared using area under the receiver operating characteristic curve (AUC). Shapley additive explanation (SHAP) was performed to interpret the model, and a nomogram was constructed to facilitate clinical application. Decision curve analysis (DCA) was used to validate the clinical utility of the nomogram. A scatter plot of predicted risk and actual depression status was used to validate its ability to stratify high- and low-risk individuals. Twelve key predictors were identified through the intersection of three feature selection methods. Logistic regression achieved the highest predictive performance (training AUC = 0.765, 95% CI: 0.703–0.815; validation AUC = 0.768, 95% CI: 0.749–0.787). SHAP analysis revealed life satisfaction, limitations in activities of daily living (ADL), self-rated health, and sleep time as the four most influential predictors. A paper-based nomogram integrating all 12 predictors was constructed to enable individualized probability estimation. The calibration plot and decision curve analysis demonstrated good calibration and clinical utility of the nomogram. The scatter plot confirmed good discrimination between participants above and below the threshold. The present study provides a tool for estimating the probability of concurrent clinically significant depressive symptoms in older adults with edentulism. It may serve as a triage aid that identifies patients who should receive direct symptom screening or referral.