A plasma metabolome-derived model predicts severe liver outcomes of nonalcoholic fatty liver disease in the UK Biobank.

Aims: Severe liver disease (SLD) in nonalcoholic fatty liver disease (NAFLD) is often diagnosed late due to the long asymptomatic period of progressive fibrosis. We aimed to identify metabolomic profiles associated with SLD and develop a predictive model to improve risk stratification.

Materials and methods: We enrolled 59 579 UK Biobank participants with a positive fatty liver index (≥60) and plasma metabolomic profiles, evaluating the incidence of cirrhosis, decompensated liver disease, hepatocellular carcinoma and/or liver transplantation. Cox regression models were applied to evaluate the associations between individual metabolites and SLD risk. Using an interpretable machine-learning framework, a metabolomics-integrated nomogram prediction model was developed and compared with conventional scoring systems.

Results: After Bonferroni correction, 110 of 249 metabolites were significantly associated with the risk of incident SLD in the Cox regression model. Among them, 11 metabolites were ultimately prioritised as predictors to construct the metabolomic score based on the optimal machine learning algorithm. The nomogram integrating metabolomic score, gamma glutamyltransferase, platelet count, waist/hip ratio, diabetes and sex showed better predictive capacity of 10-year SLD risk (area under the receiver operating characteristic 0.841 [95% CI: 0.800-0.881]) than the fibrosis-4 index (0.712, 0.662-0.763), NAFLD fibrosis score (0.659, 0.609-0.709) and aspartate aminotransferase-to-platelet ratio index (0.705, 0.652-0.759) in the validation cohort. Categorisation of participants according to selected cutoffs revealed a distinct cumulative risk of SLD, with a hazard ratio of 25.71 (95% CI: 17.10-38.66) for the high-risk group compared with the low-risk group.

Conclusions: Integrating plasma metabolomics with routine indicators enhanced the predictive capacity for severe liver outcomes of NAFLD, which shows the potential benefits in disease risk stratification and precise interventions.