Machine-learning prediction of depression in chronic kidney disease: development and internal validation using NHANES 2007-2018

Background: Depression is a frequent and underrecognized comorbidity in patients with chronic kidney disease (CKD), associated with poor adherence, reduced quality of life, and increased mortality. Despite its clinical relevance, depression is often inadequately screened in routine CKD care. Early identification of individuals at high risk may facilitate timely psychological support and improve clinical outcomes. This study aimed to develop and validate a machine learning–based model to predict depression risk in patients with CKD using nationally representative survey data. Methods: We analyzed NHANES data from 2009 to 2018. CKD and depression were defined based on eGFR values and PHQ-9 scores, respectively. After data preprocessing and exclusions, the dataset was randomly split into training and test sets (80:20). Feature selection was conducted using least absolute shrinkage and selection operator (LASSO) regression. Nine algorithms were tuned; performance was assessed with AUC, Brier score, calibration, and decision‑curve analysis. SHAP explained feature contributions. Results: Depressed participants were younger (p < 0.001), had higher BMI (p < 0.001) and lower PIR (p < 0.001). They consumed less protein, fat, and vitamins B1, B2, B6, B12 (p < 0.01). Albumin was lower and triglycerides higher (p < 0.05); calcium, BUN, ALT, AST and total cholesterol showed no difference (p > 0.05). Logistic regression achieved the highest test performance (AUC = 0.771; Brier = 0.083) and the greatest net benefit. SHAP confirmed Insomnia, low PIR, Smoking, younger Age, higher BMI and low Vitamin B6 Intake as dominant risk factors. Conclusion: We developed a concise and interpretable machine learning model for predicting depression risk in CKD patients. The model demonstrates strong performance and potential value in clinical decision support for early psychological screening.