Development of a model predicting 90-day survival in severe alcohol-related hepatitis using a retrospective cohort

Objectives: Severe alcohol-related hepatitis (sAH) is associated with high mortality. Identification of at-risk patients at baseline is challenging for adapting treatment. We aimed to develop baseline models predictive of 90-day survival. Methods: We retrospectively included 72 biopsy-proven sAH patients, characterized their baseline clinicobiological features and gut microbiota (GM) using shotgun metagenomics in combination with short-chain fatty acids and bile acids profiling. Univariate then bivariate logistic regression models predictive of 90-day outcome were constructed, with and without multiple imputation of missing data. Model performance was assessed through computation of discrimination and calibration metrics. Results: 19 sAH patients were deceased 90-day following presentation (D90) and 53 were alive (A90). At baseline, D90 presented altered polymorphonuclear score (PNS), with low circulating eosinophils and basophil proportions over neutrophils. The GM composition and its related metabolites was similar for both groups except for the phylum Fusobacteria, enriched in D90. Addition of the PNS to the baseline Glasgow Alcoholic Hepatitis Score (GAHS) identified 89.5 % of D90, with a specificity of 75.5% and an area under the curve (AUC) of 0.892 (0.817-0.967), a performance significantly superior to that of GAHS alone (sensitivity: 94.7%, specificity: 45.3%, AUC: 0.746 (0.633-0.860) (p = 0.004, DeLong). Conclusions: PNS may be routinely calculated to identify sAH patients at-risk of death in the following weeks. Quantification of fecal Fusobacteria load may prove an additional clinically relevant predictor. External validation is warranted.