From Health Checkups to Mental Health Referrals: Developing an AI-Based Depression Screening Model Using Routine Data in Older Adults Living Alone
H. W. Jung, J. J. Lee, O. KimIntroduction
Depression in older adults often remains underdiagnosed due to somatic symptoms, stigma, and low self-recognition. This challenge is particularly pronounced among those living alone, who face social isolation and limited access to mental-health care. To address this gap, an AI-based screening model using routine health checkup data—termed the Depression Prediction based on Anthropometric and Clinical Signs (DEPACS) framework—was developed to support early detection and referral in community settings (Figure 1).
Objectives
This study aimed to develop and validate an explainable AI model capable of predicting depression risk among older adults living alone using non-psychological, routinely collected health data, and to compare its predictive performance with a PHQ-9–based model.
Methods
Data were obtained from 1,087 participants in the
Intensive Managed Care for Elderly Living Alone (IMCELA)
program in Chungcheongnam-do, South Korea. Two models were trained:
a PHQ-9–based model to validate predictive performance, and a DEPACS-based screening model using seven domains (socioeconomic, vital, anthropometric, physical, behavioral, chronic disease, and sleep quality).
Five algorithms—SVM, Random Forest, Gradient Boosting, XGBoost, and KNN—were optimized via grid search and 5-fold cross-validation. AUROC and AUPRC evaluated performance, and SHapley Additive exPlanations (SHAP) visualized feature importance.
Results
The PHQ-9–based model achieved excellent discrimination (AUROC > 0.98) across algorithms (Figure 2). SHAP analysis revealed that sleep disturbance, fatigue, anhedonia, and poor concentration were the strongest predictors, followed by subjective memory complaints and chronic pain.
In contrast, the DEPACS model showed fair accuracy (AUROC = 0.74; AUPRC = 0.73). Key predictors included sleep quality, waist circumference, balance, age, and exercise (Figure 3). Poor sleep quality and elevated heart rate increased depression risk, whereas greater balance ability and higher waist circumference reduced it. These findings indicate that routine physiological and lifestyle indicators can capture meaningful depressive patterns without explicit psychological items.