Development and interpretation prediction model for depressive symptoms in patients with chronic kidney disease: a cross-sectional study based on NHANES

Depressive symptoms are prevalent and impactful among patients with chronic kidney disease (CKD). However, traditional screening methods often yield high false positive and false negative rates in this population. This study aimed to develop a machine learning-based model to improve the early identification of depressive symptoms risk among CKD patients.Data were extracted from NHANES 2011-March 2020 (pre-pandemic) as the model development cohort and split into a training cohort and an internal validation cohort. A temporally independent NHANES cycle (August 2021-August 2023) was used as an external validation cohort. Predictors were selected using least absolute shrinkage and selection operator regression. Eight machine learning algorithms were trained. Given outcome imbalance, discrimination was primarily compared using recall, precision, and F1 score, with AUCs reported as secondary measures. Calibration was assessed using calibration curves and the Brier score, and clinical utility was evaluated using decision curve analysis. SHAP was used to interpret feature contributions in the selected model.In the internal validation cohort, the RF model achieved recall 0.826, precision 0.569, and F1 score 0.673, with an AUC of 0.941. In the external validation cohort, the RF model achieved recall 0.826, precision 0.136, and F1 score 0.234, with an AUC of 0.711. SHAP identified poverty-income ratio, smoking status, weight, and urine albumin-to-creatinine ratio as the top predictors, followed by marital status, education level, gender, BMI, albumin, and hypertension.This study provides a clinically actionable tool to stratify the risk of clinically significant depressive symptoms in patients with CKD, which may support targeted screening, timely mental health assessment or referral, and more efficient allocation of clinical resources. SHAP-based explanations help interpret individual risk estimates to inform patient-centered management.