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

Higher VAI was independently associated with increased risk of cardiorenal syndrome after adjustment for confounders.

• OR = 1.29, 95% CI = 1.13–1.49 for the continuous association between VAI and CRS
• Quartile analysis showed a 53% elevated risk in the highest versus lowest VAI quartile (Q4 vs Q1: OR = 1.53, 95% CI = 1.15–2.03)
• Analysis was based on NHANES data from 33,605 adults using survey-weighted logistic regression
• VAI is described as a sex-specific composite metric serving as an indicator of visceral adipose accumulation and associated cardiometabolic risk

The association between VAI and cardiorenal syndrome was linear rather than nonlinear.

• Restricted cubic splines (RCS) analysis did not indicate significant nonlinearity (P for non-linear = 0.98)
• This suggests a linear dose-response relationship between VAI and CRS risk
• RCS was used alongside logistic regression to examine the shape of the association

Hypertension status exhibited a significant interaction with the VAI–cardiorenal syndrome association in subgroup analyses.

• Subgroup analyses were conducted to evaluate effect modification across different population characteristics
• Hypertension was identified as a significant effect modifier in the relationship between VAI and CRS
• This interaction was detected among multiple subgroup analyses examining potential moderating variables

The XGBoost model demonstrated superior predictive performance compared to SVM and GLM models for predicting cardiorenal syndrome.

• Three machine learning models were developed and compared: XGBoost, SVM, and GLM
• Model performance was evaluated using receiver operating characteristic curves, Youden's J statistic, and F1 score
• XGBoost outperformed the other models across the evaluated metrics
• This represents the first explainable ML-driven CRS prediction benchmark using VAI within a nationally representative population

SHAP analysis of the XGBoost model identified age, VAI, and hypertension as the three most important features for predicting cardiorenal syndrome.

• Shapley Additive exPlanations (SHAP) were used to evaluate model interpretability
• Age was ranked as the most important predictive feature
• VAI was ranked as the second most important predictive feature
• Hypertension was ranked as the third most important predictive feature

The study used a large, nationally representative cross-sectional sample from NHANES to investigate the association between VAI and CRS.

• Data were drawn from the National Health and Nutrition Examination Survey (NHANES)
• The analytic sample included 33,605 adults
• Survey-weighted analyses were conducted to account for the complex sampling design
• The authors describe this as investigating a 'previously underexplored association between CRS and VAI'