Mental health literacy predicts depression in older adults in China: an interpretable machine learning model

ObjectivesThis study was to construct an interpretable model for predicting the risk of depression among older individuals. The XGBoost model was developed to investigate the protective and risk factors associated with depression in older people.MethodsThis study accomplished model development and performance comparisons regarding depression in older individuals. The predictive ability of the XGBoost model was compared with that of three other machine learning techniques using the area under the curve (AUC) metric. To interpret the XGBoost model’s outcomes, we utilized the SHapley Additive exPlanation (SHAP) method.ResultsCompared with the other three models, the XGBoost model exhibited superior predictive efficacy, with an AUC value of .806. Decision curve analysis revealed that the XGBoost model provided greater net benefits than other machine learning algorithms when the threshold probability was less than .4. The SHAP method revealed all the predictors of depression in older individuals ranked by importance, and considering older adults as a burden on society was identified as the most significant risk predictor variable. Resilience was the most important protective factor for depression in older adults. Overall, mental health literacy was the most crucial risk and protective factor for depression in older people.DiscussionThe interpretable predictive model facilitates a more precise estimation of the risk of depression in older individuals by health care professionals, providing better prevention plans and treatment targets. Additionally, this transparent framework enhances the clarity of the model, making it easier to understand the reliability of the predictions for mental health concerns.