Predicting depression in middle-aged adults using kidney deficiency questionnaire: an exploratory machine learning study.

In Korean medicine (KM), kidney deficiency represents age-related physiological decline theoretically linked to depression. However, its empirical predictive value in modern clinical frameworks remains underexplored. This study aimed to develop and validate a machine learning model predicting depression in middle-aged adults by integrating kidney deficiency with psychosocial variables. We analyzed baseline data from 1000 community-dwelling adults aged 50-65 from the Korean Medicine for Aging Cohort study. Depression was defined using the Geriatric Depression Scale Short Form (GDS-SF-K ≥ 5). Predictors included the kidney deficiency questionnaire (KDQ), psychosocial variables (social support), and demographics (BMI). Machine learning models were trained and selected using 5-fold cross-validation based on precision-recall area under the curve (PR-AUC). The final model was validated on an independent test set. Among participants (mean age 57.4 ± 4.2 years, 76.3% female), 164 (16.4%) met depression criteria. The optimal model was a multilayer perceptron incorporating KDQ total score, social support, and BMI. On the independent test set, this parsimonious model achieved ROC-AUC of 0.820 (95% CI: 0.752-0.879), PR-AUC of 0.521 (95% CI: 0.381-0.660), sensitivity of 0.673, and specificity of 0.757. Integrating standardized kidney deficiency assessment with key psychosocial factors provides a moderately accurate, parsimonious model for predicting depression. This study offers the first quantitative evidence for this KM construct's predictive utility, suggesting that integrative approaches considering systemic somatic decline and psychosocial health can enhance depression risk stratification in middle-aged adults.