Interpretable machine learning analysis of immunoinflammatory biomarkers for predicting CHD among NAFLD patients.

Abstract

Background: Coronary Heart Disease (CHD) and Non-Alcoholic Fatty Liver Disease (NAFLD) share overlapping pathogenic mechanisms including adipose tissue dysfunction, insulin resistance, and systemic inflammation mediated by adipokines. However, the specific impact of inflammation and immune responses on CHD risk in NAFLD patients remains poorly understood. This study evaluated the predictive value of ten immunoinflammatory indexes for CHD risk in NAFLD patients using an interpretable machine learning framework.

Methods: We retrospectively analyzed 407 NAFLD patients undergoing coronary angiography, and stratifying them into NAFLD + CHD (n = 250) and NAFLD (n = 157) groups. Ten immunoinflammatory indexes were derived from the blood laboratory results. Lasso regression analysis and propensity score matching (PSM) were employed to mitigate confounding effects. Subsequently, univariate and multivariate logistic regression analyses were used to identify independent risk factors for CHD occurrence among NAFLD patients. While restricted cubic splines (RCS) and Receiver operating characteristic (ROC) curve evaluated the relationship between each immunoinflammatory indexes and CHD risk. Linear correlation methods were employed to evaluate the relationship between Gensini score and immunoinflammatory indexes. Finally, three machine learning algorithms (RF, SVM and GLM) were used to identify significant risk factors. To interpret the diagnostic model built by Random Forest, the SHapley Additive exPlanations (SHAP) method was employed, and features were ranked according to their SHAP values. Based on these rankings, a diagnostic nomogram was further constructed and the accuracy of the diagnostic model was evaluated using ROC curves.

Result: After PSM, among the 282 included patients with NAFLD, 141 cases (50%) were complicated with CHD. Multivariate logistic regression analysis revealed that after adjusting for age, sex, hypertension, and smoking history, the NHR index was identified as the most significant risk factor for CHD in NAFLD patients (OR, 1.375; 95% CI, 1.021-1.852; P < 0.001). Additionally, NLR, SII, SIRI and NMR were also identified as risk factors. PNR was a protective factor for CHD events in patients with NAFLD. RCS analysis demonstrated linear relationships between the NHR, NLR, and PNR index with CHD occurrence, whereas the SII index exhibited a non-linear J-shaped relationship with CHD risk (non-linear P = 0.025). Correlation analysis with Gensini score showed that the NHR index had the highest correlation with the severity of CHD (R = 0.256, P < 0.001). ROC curves indicated that the NHR index had good predictive and diagnostic performance (AUC = 0.703,95% CI, 0.652-0.754). Finally, the diagnostic nomogram constructed based on SHAP values demonstrated good accuracy and predictive performance (AUC = 0.834,95% CI, 0.795-0.873; P < 0.001).

Conclusion: Six immunoinflammatory markers demonstrated significant associations with CHD risk in NAFLD populations, among which the NHR index exhibited particularly promising predictive potential.