Association between phenolic compound exposure and liver function: Using machine learning models to identify primary risk factors

Objective: This study aims to evaluate the impacts of exposure to phenolic compounds on liver function, while also using machine learning models to identify primary risk pollutants.

Methods: Based on the Henan rural cohort, this cross-sectional study enrolled 876 normal glucose tolerance (NGT) individuals and 860 impaired fasting glucose (IFG) individuals. Ultra-performance liquid chromatography tandem mass spectrometry was utilized to measure the levels of endocrine-disrupting chemicals, including bisphenol A (BPA), bisphenol E (BPE), bisphenol P (BPP), bisphenol S (BPS), bisphenol Z (BPZ), methyl paraben (MeP), ethyl paraben (EtP), propyl paraben (PrP), butyl paraben (BuP), and benzyl paraben (BzP). Linear models, the quantile-based g-computation (QGC) model, the Bayesian Kernel Machine Regression (BKMR) model, and a machine learning model were employed to assess the associations between single- and mixed-exposure and liver injury indicators.

Results: In the IFG group, linear regression, QGC models, and BKMR models all showed positive associations of exposure with Aspartate aminotransferase and Fibrosis-4 index, in which EtP, BuP, BPS, and BPZ were found in single-exposure models and EtP, as well as BPZ, contributing mostly among mixed-exposure models. In addition, QGC and BKMR models presented an increased Aspartate aminotransferase-to-platelet ratio index with mixed exposure. However, no significant results were found in the NGT group. By integrating the results of the ML model and the mixed-exposure analysis, for the IFG population, EtP, BuP, BPP, BPZ, and BPE were risk factors for AST, Fib-4, and APRI. Subgroup analysis showed more evident results in females, with IFG having a significant modifying effect.

Conclusion: Phenolic compound exposure was associated with impaired liver function among the IFG population. EtP, BuP, BPP, BPZ, and BPE were identified as primary risk factors for liver fibrosis indicators. IFG females demonstrated a stronger association between mixed exposure to phenolic EDCs and liver fibrosis indicators, and hyperglycemia modified the association.