Interpretable ensemble learning model with shapley additive explanations for predicting anxiety symptoms risk in Chinese older adults with body shape index abnormality

Abstract

This study aimed to construct and validate an interpretable risk prediction model for anxiety symptoms in Chinese older adults with abnormal body shape, explore the association between A Body Shape Index (ABSI) and anxiety symptoms, and identify key predictive factors via explainable methods. Data were from the Chinese Longitudinal Healthy Longevity Survey (CLHLS) 2008–2014 (n = 1,844/2,663/3,058 for 2008/2011/2014). The 2008 data (80% training, 20% internal validation) and 2011/2014 data (external validation) were used. Feature selection, data balancing, ensemble learning (Boosting/Stacking/Voting), and Shapley Additive exPlanations (SHAP) were applied. ABSI was positively associated with anxiety symptoms (P = 0.038), with stronger effects in males (trend slope = 0.03) than females (0.02); female anxiety prevalence (39.37%) was higher than males (20.79%). The Boosting-ADASYN model performed best (internal AUC = 0.814, external AUC = 0.766–0.772). SHAP identified marital status, age, self-reported health, education, and happiness as top predictors. ABSI outperformed BMI in capturing abnormal body fat distribution. This study provides an interpretable tool for early anxiety identification in this population, supporting precise interventions combining ABSI and psychosocial strategies.