Machine Learning–Based Comparative Analysis of Blood Cell–Derived Inflammatory Indices for Predicting MAFLD and Liver Fibrosis: Evidence From NHANES 2017–2020

BACKGROUND: Metabolic dysfunction associated fatty liver disease (MAFLD) is linked to chronic low-grade inflammation, but the diagnostic value of blood cell-derived inflammatory indices remains unclear. This study assessed their associations and predictive ability for MAFLD and liver fibrosis in a nationally representative population. METHODS: This cross-sectional study included 3501 participants from NHANES 2017-2020. Multivariable logistic regression was used to evaluate associations between six inflammatory indices (AISI, NLR, LMR, PLR, SII and SIRI) and MAFLD with or without liver fibrosis. Restricted cubic spline analysis assessed dose-response relationships, while receiver operating characteristic (ROC) curves and random forest analysis evaluated diagnostic performance and variable importance. RESULTS: Multivariable logistic regression analysis showed that higher values of ln (AISI) and ln (SIRI) were significantly associated with increased odds of MAFLD, whereas ln (SII), ln (LMR) and ln (NLR) were not significantly associated with MAFLD (p > 0.05). In contrast, ln (PLR) was inversely associated with MAFLD. Among participants with liver fibrosis (LF), higher values of ln (AISI), ln (NLR) and ln (SIRI) were positively associated with MAFLD, whereas ln (PLR) remained inversely associated. No significant associations were observed for ln (LMR) or ln (SII) with MAFLD in the presence of LF. Among the evaluated inflammatory indices, AISI demonstrated the highest discriminatory performance for MAFLD (AUC = 0.752), whereas PLR showed the best performance for predicting liver fibrosis. These findings were further supported by machine learning analysis. CONCLUSION: Inflammatory indices showed distinct associations with MAFLD and liver fibrosis. AISI was most strongly associated with MAFLD, whereas PLR showed the best performance for fibrosis assessment.