Five ACLF diagnostic models and prognostic scores identified markedly different proportions of patients at risk of 28-day mortality in a cohort study.

Background and Aims: Acute-on-chronic liver failure (ACLF) is associated with high short-term mortality, but substantial heterogeneity among existing diagnostic and prognostic models results in inconsistent patient identification and risk assessment. We conducted a systematic head-to-head comparison of major ACLF diagnostic and prognostic models to evaluate concordance, short-term mortality prediction and clinical utility, with the goal of informing harmonization of ACLF assessment. Methods: We analysed 3,370 patients with acute decompensation of cirrhosis in the COSSH cohort, with external validation in an independent Ambi-Spective cohort from India (n=2,055). Five ACLF diagnostic models were evaluated for identification of patients at risk of 28-day mortality. Reclassification was assessed using net reclassification improvement. Prognostic scores were compared using concordance index, integrated discrimination improvement, calibration, and decision-curve analysis. Results: Diagnostic frameworks identified markedly different proportions of ACLF. A-TANGO and COSSH-ACLF classified the largest high-risk populations while maintaining substantial short-term mortality and balanced sensitivity-specificity profiles. Compared with COSSH-ACLF, A-TANGO improved net reclassification by 7.7%, with further gains versus EASL-CLIF (11.8%), APASL-ACLF (36.4%), and NACSELD-ACLF (45.9%). In the external cohort, A-TANGO and COSSH-ACLF showed similar discrimination and identified comparable proportions of patients. Combined application of the two models delineated three clinically meaningful strata, identifying a discordant intermediate-risk group with approximately 11% 28-day mortality. Among prognostic scores, COSSH-ACLF II and A-TANGO OF scores demonstrated strong and complementary performance across cohorts. Conclusions: Outcome-anchored ACLF definitions converge in identifying patients at highest short-term risk across diverse populations. Alignment between A-TANGO and COSSH-ACLF, together with identification of an intermediate-risk phenotype, supports a data-driven framework for improving consistency and advancing global harmonization of ACLF diagnosis and risk stratification.