A machine-learning-derived online screening tool for depressive symptoms in chronic digestive system diseases patients: A cross- sectional study with temporal validation from CHARLS

Chronic digestive system diseases (CDSD) are common in older adults, and depressive symptoms substantially worsen prognosis and quality of life. We developed and temporally validated a machine-learning model to identify depressive symptoms in individuals with CDSD using data from the China Health and Retirement Longitudinal Study (CHARLS). This study included 3,762 participants with CDSD from the 2011 survey and examined 46 behavioral, health, psychological, and sociodemographic variables. Feature selection was performed using logistic regression and LASSO regression, and seven machine-learning algorithms were compared. Temporal validation was conducted in newly diagnosed CDSD participants from the 2015 CHARLS wave. Among 3,762 participants, 1,900 had depressive symptoms. Thirteen variables were retained, including education, residence, life assessment, health assessment, fall history, disability, kidney disease, arthritis, heart disease, eyesight, instrumental activities of daily living, sleep duration, and grip strength. XGBoost showed the best performance in the testing set, with an area under the curve of 0.793 and an F1-score of 0.724, together with good calibration and clinical utility. These findings suggest that machine-learning approaches may support early identification of depressive symptoms in people with CDSD.