Identification of cardiovascular disease in patients with kidney stone disease using explainable machine learning

Background Kidney stone disease is an independent risk factor for cardiovascular disease (CVD), but specific tools for identifying CVD in patients with kidney stone disease are lacking. This study aimed to validate the association between kidney stones and CVD and to develop an interpretable machine learning model for the identification of prevalent CVD in individuals with kidney stones. Methods Using data from 34,770 participants in the NHANES 2007–2018 cycle, weighted multivariable logistic regression and subgroup analysis were employed to examine the association between kidney stones and CVD. A total of 1,491 NHANES participants from 2007 to 2016 were used for model development and internal validation, while 296 participants from the 2017–2018 cycle were used as an independent temporal validation cohort. The Shapley Additive exPlanation (SHAP) method was used for global and local interpretation. Results Model 3 revealed a 47% increased risk of CVD in participants with kidney stones compared to those without (OR = 1.47, 95% CI: 1.20–1.80). In the internal test set, the logistic regression (LR) model performed best, with an area under the receiver operating characteristic curve of 0.801, sensitivity of 0.721, specificity of 0.771, accuracy of 0.759, recall of 0.721, and Brier score of 0.169. LR also demonstrated the best performance in the temporal validation cohort. SHAP analysis identified the importance of 15 predictors. Conclusions This study highlights an association between kidney stones and prevalent CVD, though causality cannot be inferred due to the cross-sectional design. The LR model demonstrated strong performance in identifying prevalent CVD in patients with kidney stone disease.