Explainable machine learning framework using visceral adiposity index to predict cardiorenal syndrome: a survey-weighted NHANES study with SHAP interpretation

Cardiorenal syndrome (CRS) involves bidirectional pathophysiology between cardiovascular and renal dysfunction. The visceral adiposity index (VAI), a sex-specific composite metric, serves as an indicator of visceral adipose accumulation and associated cardiometabolic risk. This cross-sectional study aims to investigate the previously underexplored association between CRS and VAI. We analyzed National Health and Nutrition Examination Survey (NHANES) data from 33,605 adults. Logistic regression models and restricted cubic splines (RCS) were utilized to examine the association. Machine learning (ML) models (XGBoost, SVM, and GLM) were developed and evaluated using receiver operating characteristic curves, Youden's J, and F1 score. The model interpretability was evaluated by Shapley Additive exPlanations (SHAP). The logistic regression analysis, adjusted for confounders, demonstrated a positive association between VAI and CRS. Higher VAI was independently associated with increased risks of CRS (OR = 1.29, 95% CI = 1.13-1.49). Quartile analysis demonstrated a 53% elevated risk in the highest versus lowest VAI quartile (Q4 vs Q1: OR = 1.53, 95% CI = 1.15-2.03). The RCS did not indicate significant nonlinearity (P for non-linear = 0.98), suggesting a linear association between VAI and CRS. Subgroup analyses revealed that hypertension status exhibited a significant interaction. The XGBoost model demonstrated superior predictive performance. The SHAP plot of XGBoost revealed that age, VAI, and hypertension were the three most important features for predicting CRS. Elevated VAI is independently associated with an increased risk of CRS. We introduce the first explainable ML-driven CRS prediction benchmark using VAI within a nationally representative population.