Depression and chewing difficulty are prevalent health concerns among middle-aged and older adults. This study aimed to develop group-specific machine learning models to identify depression in middle-aged and older adults with preserved versus impaired chewing ability, enabling tailored prevention and management strategies. The research data was derived from the 2018 China Health and Retirement Longitudinal Study (CHARLS). Five machine learning algorithms were employed to develop group-specific models stratified by chewing ability, which were compared against a unified model for validation. The evaluation metrics for the models included accuracy, precision, recall, F1 score, and the area under the receiver operating curve (AUC). The SHapley Additive exPlanations (SHAP) method was utilized to analyze feature importance and to interpret the individualized identification process within the optimal model. The analyzed sample consisted of 15,459 middle-aged and older adults, with 82.84% of them aged between 45 and 69. Of the participants, 32.06% were identified as having depressive symptoms, while 38.63% exhibited impaired chewing ability. Surpassing the unified model in performance, the Deep Neural Network (DNN) and Multilayer Perceptron (MLP) models emerged as the optimal algorithms for identifying depression in middle-aged and older adults with preserved versus impaired chewing ability, respectively. Feature importance analysis revealed that self-rated health was the most significant feature in the DNN model, while life satisfaction was the most important factor in the MLP model. This study developed group-specific DNN and MLP models to identify depression in middle-aged and older adults with preserved versus impaired chewing ability. These findings provide valuable references for developing targeted screening tools to support the prevention and early intervention of depression.
Song et al. (Mon,) studied this question.