10-Protein Blood Test Accurately Subtypes Liver Disease and Predicts Mortality Risk
Researchers used proteomics to split a hybrid liver disease into two distinct subtypes with starkly different survival and cirrhosis outcomes.
Summary
A new study analyzing over 443,000 UK Biobank participants used plasma proteomics to classify MetALD — a liver disease sitting between alcohol-related and metabolic liver disease — into two biologically distinct subtypes. A 10-protein model achieved 93% accuracy in distinguishing alcohol-driven from cardiometabolic-driven disease. Patients classified as alcohol-predominant faced significantly higher risks of cirrhosis, liver cancer, and death over 15 years compared to those with cardiometabolic-predominant MetALD. The proteome outperformed metabolomics alone and was not improved by adding clinical variables like BMI or liver enzymes, suggesting proteins carry uniquely powerful diagnostic information for personalizing liver disease management.
Detailed Summary
Metabolic dysfunction- and alcohol-associated liver disease (MetALD) is a newly recognized condition affecting millions who have both significant alcohol consumption and metabolic risk factors like obesity or diabetes. Until now, clinicians lacked tools to determine whether a given MetALD patient's disease biology more closely resembles alcohol-related liver disease (ALD) or metabolic steatotic liver disease (MASLD) — a distinction that could profoundly shape treatment and prognosis.
Researchers from Oxford and Novo Nordisk analyzed data from 443,453 European participants in the UK Biobank, including over 34,000 with MetALD. Using elastic net regression on 2,941 plasma proteins and 249 plasma metabolites, they built classification models to differentiate ALD from MASLD. They then applied the best-performing model to MetALD patients to sort them into alcohol-predominant or cardiometabolic-predominant subgroups.
The proteome model achieved a remarkable AUC of 0.96 for distinguishing ALD from MASLD — far outperforming metabolomics alone (AUC 0.86). A streamlined 10-protein signature maintained an AUC of 0.93. Crucially, adding age, sex, BMI, liver enzymes, or metabolomic data did not improve the proteome model, highlighting proteins as uniquely informative biomarkers.
When applied to MetALD patients over 15 years of follow-up, alcohol-predominant MetALD patients had significantly higher risks of cirrhosis, hepatocellular carcinoma, and all-cause mortality compared to cardiometabolic-predominant patients. They also showed elevated fibrosis scores and more advanced fibrosis stages at baseline.
These findings validate proteomic subtyping as a clinically actionable approach, enabling physicians to identify the highest-risk MetALD patients and tailor interventions accordingly. Limitations include reliance on self-reported alcohol intake and a predominantly European cohort, warranting replication in diverse populations.
Key Findings
- A 10-protein plasma model distinguished ALD from MASLD with 93% accuracy (AUC 0.93).
- The full proteome model achieved AUC 0.96, outperforming metabolomics alone (AUC 0.86).
- Adding BMI, liver enzymes, or metabolites did not improve the proteome classification model.
- Alcohol-predominant MetALD patients had significantly higher 15-year risks of cirrhosis and death.
- Proteomic subtyping of MetALD enables personalized risk stratification for over 34,000 affected individuals.
Methodology
Cross-sectional and longitudinal analysis of 443,453 UK Biobank participants using elastic net regression on plasma proteomics (2,941 proteins) and metabolomics (249 metabolites). Cox regression modeled 15-year risks of major outcomes including cirrhosis, hepatocellular carcinoma, and mortality. Multiple sensitivity analyses were performed to validate model robustness.
Study Limitations
Alcohol consumption was self-reported, introducing potential misclassification bias. The cohort is predominantly European, limiting generalizability to other ethnic populations. Diagnostic criteria specificity and the observational nature of UK Biobank data require validation in independent, diverse clinical cohorts.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.