Life's Crucial 9 and NAFLD from association to SHAP-interpreted machine learning predictions.
Journal:
Scientific reports
PMID:
40102489
Abstract
Non-alcoholic fatty liver disease (NAFLD) is the most prevalent chronic liver disease worldwide. Cardiovascular disease (CVD) and NAFLD share multiple common risk factors. Life's Crucial 9 (LC9), a novel indicator for comprehensive assessment of cardiovascular health (CVH), has not yet been studied in terms of its association with or predictive value for NAFLD. This study analyzed data from 10,197 participants in the National Health and Nutrition Examination Survey (NHANES) from 2007 to 2018. The association between LC9 and NAFLD was assessed using weighted logistic regression, while weighted Cox proportional hazards models were applied to evaluate the relationship between LC9 and all-cause mortality among NAFLD patients. Restricted cubic spline (RCS) analysis was conducted to explore dose-response relationships, and Kaplan-Meier survival curves were utilized to examine differences in survival outcomes. Machine learning (ML) approaches were employed to construct predictive models, with the optimal model further interpreted using SHapley Additive exPlanations (SHAP). An increase of 10 points in LC9 was negatively associated with the risk of NAFLD (model 3: OR = 0.39, 95% CI = 0.36 - 0.42, P < 0.001) and all-cause mortality in NAFLD patients (model 3: HR = 0.78, 95% CI = 0.67 - 0.91, P < 0.001). A non-linear relationship was observed between LC9 and NAFLD (P < 0.0001 for nonlinearity). Among the eight ML models, the Support Vector Machine (SVM) demonstrated the best predictive performance (AUC = 0.873). SHAP analysis indicated that LC9 was the most significant predictor in the model. LC9 demonstrated a nonlinear negative association with NAFLD and a linear negative association with all-cause mortality in NAFLD patients. Maintaining a higher LC9 score may reduce the risk of NAFLD and all-cause mortality among NAFLD patients. The predictive model developed using Support Vector Machine (SVM) exhibited strong clinical predictive value, with LC9 being the most critical factor in the model, facilitating self-risk assessment and targeted intervention.