Machine learning-based cross-sectional exploration of depression-chronic pain associations in chronic respiratory disease patients

BACKGROUND: Chronic respiratory diseases (CRD) are a leading global cause of death, with high comorbidity rates of depression and chronic pain forming a bidirectional relationship that is associated with worse prognosis. Existing research lacks systematic analysis of their association in CRD patients, and traditional statistical methods are limited in identifying complex relationships, highlighting the need for machine learning-based exploratory analysis. METHODS: This cross-sectional study used data from the 2015 wave of the China Health and Retirement Longitudinal Study (CHARLS), including 1,702 eligible CRD patients. Depression was assessed by the CESD-10 scale, and chronic pain was defined as self-reported long-term pain. Six machine learning algorithms (XGBoost, SVM, MLP, LightGBM, RF, LR) were employed, with model performance evaluated by AUROC, accuracy, and other metrics. LASSO regression selected key features, and subgroup and restricted cubic spline analyses explored heterogeneity and dose-response associations. RESULTS: Among participants, 43.2% had chronic pain. CESD-10 score showed the strongest association with chronic pain (AUC = 0.762), with a significant dose-response relationship: the highest CESD-10 tertile had a 7.18-fold higher risk of chronic pain (95% CI: 2.45-4.78, P < 0.001) after full adjustment. The XGBoost model achieved the best predictive performance among the six tested models (test set AUROC = 0.830). Subgroup analysis showed a stronger depression-pain association in urban than rural residents (P for interaction = 0.016). Key predictive factors include self-rated health status, activities of daily living (ADL), and gender, among others. CONCLUSION: Depression severity is robustly associated with chronic pain in CRD patients in a dose-dependent manner. The XGBoost model provides reliable early identification tools. These findings support routine depression assessment and geographically tailored bio-psycho-social interventions to disrupt the depression-pain cycle and improve patient outcomes.