Longitudinal association between metal mixture exposure and non-alcoholic fatty liver disease: A prospective cohort study based on generative adversarial networks.

Environmental metal exposure represents a potential risk factor for non-alcoholic fatty liver disease (NAFLD); nevertheless, existing observational studies often lack reliable methodologies for exploring potential associations within a counterfactual framework, particularly when addressing complex co-exposures. Addressing these constraints, we conducted a prospective cohort study involving 4095 participants from the Yinchuan community-dwelling elderly cohort with a 3-year follow-up. Twelve baseline urinary metal concentrations were quantified utilizing inductively coupled plasma mass spectrometry (ICP-MS). We employed a multi-tiered analytical strategy encompassing cross-sectional models and longitudinal prospective analyses, alongside Bayesian Kernel Machine Regression (BKMR) for assessing joint mixture effects and interactions. Furthermore, the Generative Adversarial Networks (GAN) framework estimated individualized exposure effects within a counterfactual paradigm. Our initial cross-sectional analysis linked zinc (Zn), cobalt (Co), iron (Fe), and tellurium (Te) with NAFLD prevalence. Conversely, longitudinal assessments demonstrated that baseline Zn and Co were not prospective risk factors, whilst Te was significantly associated with elevated longitudinal repeated measures of the hepatic steatosis index (HSI). Notably, longitudinal analyses revealed that elevated baseline selenium (Se) and copper (Cu) were inversely associated with the longitudinal repeated measures of liver fibrosis indices (Fibrosis-4 index and aspartate aminotransferase-to-platelet ratio index). Additionally, interaction models identified pronounced synergistic relationships of Fe with both Te and vanadium (V) regarding hepatic steatosis. Within the counterfactual GAN-based models, Se emerged as a primary potential contributor to NAFLD risk (average estimated exposure effect = 0.072), followed by lead (Pb). Model estimates linked 18.8% and 5.1% of NAFLD cases to Se and Pb exposures, respectively. Ultimately, this multi-tiered study suggests potential associations between long-term environmental metal exposure and the progression of NAFLD within a causal framework, highlighting Se and Pb as primary factors potentially associated with the disease burden, while Te exhibits a potential contribution to elevating longitudinal hepatic steatosis burden through synergistic interactions.