Acute-on-chronic liver failure (ACLF) is a distinct clinical syndrome that develops in patients with preexisting chronic liver disease, most commonly cirrhosis. It is characterized by acute hepatic decompensation, extrahepatic organ failure, systemic inflammation, and high short-term mortality [
1,
2]. Global 28-day mortality in cirrhotic patients with ACLF is estimated at 45%, with marked geographic variation reflecting differences in etiologies, populations, and healthcare systems [
3]. Despite extensive study, a universally accepted definition of ACLF remains lacking. The European Association for the Study of the Liver-Chronic Liver Failure (EASL-CLIF) consortium, the North American Consortium for the Study of End-Stage Liver Disease (NACSELD), and the Asian Pacific Association for the Study of the Liver (APASL) each emphasize different aspects, including disease stage, precipitating events, and organ failure thresholds [
4-
6]. Nonetheless, all definitions converge on ACLF as an acute, rapidly progressive, high-risk condition requiring urgent attention.
Traditional prognostic tools such as the CLIF consortium organ failures score and the APASL ACLF Research Consortium score have provided important frameworks for risk stratification [
7,
8]. Yet, their performance has been inconsistent, owing to heterogeneity in definitions, patient selection, disease severity, and management practices. Moreover, while mortality has traditionally been linked to the number of organ failures, recent evidence indicates that not all organ failures carry equal prognostic weight [
9,
10]. This underscores the need for more refined models capable of addressing the complexity of ACLF.
Against this backdrop, the recent study by Yeo et al. [
11] offers a notable advance. Using data from over 5,000 intensive care unit (ICU) admissions of cirrhotic patients, the authors defined ACLF with both EASL-CLIF and NACSELD criteria and developed cohort-specific machine learning models to predict 30-day mortality in patients with ≥2 organ failures. The models showed excellent discrimination (area under the receiver operating characteristic curve [AUC] 0.87 for NACSELD; 0.83 for EASL-CLIF), outperforming conventional scores such as model for end-stage liver disease (MELD), MELD 3.0, MELD with sodium (MELDNa), and CLIF-C ACLF (all <0.80). By incorporating multiple variables and nonlinear associations, the models captured complex interactions between organ failures and other predictors. SHapley Additive exPlanations (SHAP) further enhanced interpretability, identifying the international normalized ratio (INR) as the most influential predictor in both cohorts.
Robustness was demonstrated through internal and external validation in an independent German cohort, with sustained accuracy (AUC 0.79 for NACSELD; 0.73 for EASL-CLIF). Performance remained strong (≥0.80) across subgroups defined by precipitating events such as alcohol use and bacterial infection. To promote clinical adoption, the authors developed a web-based calculator for bedside risk estimation, providing an accessible decision-support tool.
Several limitations deserve attention. Hepatic encephalopathy was assessed using the Glasgow coma scale rather than the West Haven criteria, potentially leading to discrepancies in identifying brain failure. Prognosis was derived from baseline variables only, without accounting for dynamic changes or treatment effects during hospitalization, an important limitation in a highly dynamic condition such as ACLF [
12]. The study population was drawn from the MIMIC-IV database, which is predominantly White (>70%), raising concerns about generalizability. Finally, although SHAP improved interpretability, it does not establish causality, and caution is warranted when applying these associations to clinical decisions.
From a public health perspective, this study addresses an urgent need. ACLF is associated with frequent ICU admission, high short-term mortality, and significant healthcare resource utilization. Improved prognostic models can guide timely identification of high-risk patients, inform discussions regarding intensive care, and facilitate early transplant referral. Moreover, an accessible web-based platform increases feasibility across diverse clinical and research settings. Beyond clinical application, the findings advance pathophysiological understanding by clarifying the relative contribution of individual organ failures and systemic factors. Demonstrating that not all organ failures contribute equally to prognosis challenges traditional assumptions and highlights disease heterogeneity. These insights may refine future guidelines, patient selection for trials, and therapeutic targets.
Looking forward, several directions merit exploration. Prospective validation in ethnically diverse and geographically distinct cohorts is needed to ensure generalizability. Incorporating longitudinal data and treatment variables into prediction models may enable dynamic, real-time risk assessment more aligned with clinical trajectories. Integration with electronic health records could facilitate seamless application in routine care, supporting personalized management. Finally, linking prognostic models to therapeutic algorithms—such as standardized infection control, organ support strategies, and transplant evaluation—may help translate predictive accuracy into improved outcomes.
In conclusion, Yeo et al. [
11] provide compelling evidence that machine learning can substantially enhance prognostication in ACLF, a complex and devastating syndrome. By surpassing conventional scoring systems, offering interpretability through SHAP, and enabling bedside application via a web-based tool, their study marks a meaningful step forward in the management of critically ill patients with cirrhosis. Although further refinement and validation are necessary, this work illustrates the promise of data-driven approaches to long-standing challenges in hepatology, with potential to improve outcomes, optimize resource use, and reduce the broader societal burden of ACLF.
FOOTNOTES
-
Authors’ contribution
Jung Hee Kim confirms sole responsibility for the following: study conception, literature review, drafting of the manuscript, and final approval.
-
Conflicts of Interest
The author has no conflicts to disclose.
Abbreviations
acute-on-chronic liver failure
Asian Pacific Association for the Study of the Liver
area under the receiver operating characteristic curve
European Association for the Study of the Liver-Chronic Liver Failure
international normalized ratio
model for end-stage liver disease
North American Consortium for the Study of End-Stage Liver Disease
SHapley Additive exPlanations
REFERENCES
- 1. Br VK, Sarin SK. Acute-on-chronic liver failure: terminology, mechanisms and management. Clin Mol Hepatol 2023;29:670-689.
- 2. Luo J, Li J, Li P, Liang X, Hassan HM, Moreau R, et al. Acute-on-chronic liver failure: far to go-a review. Crit Care 2023;27:259.
- 3. Mezzano G, Juanola A, Cardenas A, Mezey E, Hamilton JP, Pose E, et al. Global burden of disease: acute-on-chronic liver failure, a systematic review and meta-analysis. Gut 2022;71:148-155.
- 4. Sarin SK, Choudhury A, Sharma MK, Maiwall R, Al Mahtab M, Rahman S, et al. Acute-on-chronic liver failure: consensus recommendations of the Asian Pacific association for the study of the liver (APASL): an update. Hepatol Int 2019;13:353-390.
- 5. O’Leary JG, Reddy KR, Garcia-Tsao G, Biggins SW, Wong F, Fallon MB, et al. NACSELD acute-on-chronic liver failure (NACSELD-ACLF) score predicts 30-day survival in hospitalized patients with cirrhosis. Hepatology 2018;67:2367-2374.
- 6. Moreau R, Jalan R, Gines P, Pavesi M, Angeli P, Cordoba J, et al. Acute-on-chronic liver failure is a distinct syndrome that develops in patients with acute decompensation of cirrhosis. Gastroenterology 2013;144:1426-1437.e9.
- 7. Rashed E, Soldera J. CLIF-SOFA and CLIF-C scores for the prognostication of acute-on-chronic liver failure and acute decompensation of cirrhosis: a systematic review. World J Hepatol 2022;14:2025-2043.
- 8. Choudhury A, Jindal A, Maiwall R, Sharma MK, Sharma BC, Pamecha V, et al. Liver failure determines the outcome in patients of acute-on-chronic liver failure (ACLF): comparison of APASL ACLF research consortium (AARC) and CLIF-SOFA models. Hepatol Int 2017;11:461-471.
- 9. Arroyo V, Moreau R, Jalan R. Acute-on-chronic liver failure. N Engl J Med 2020;382:2137-2145.
- 10. Gustot T, Stadlbauer V, Laleman W, Alessandria C, Thursz M. Transition to decompensation and acute-on-chronic liver failure: role of predisposing factors and precipitating events. J Hepatol 2021;75 Suppl 1:S36-S48.
- 11. Yeo YH, Zhang M, McCoy MS, Zu J, He Y, Liu Y, et al. Predictive machine learning model in intensive care unit patients with acute-on-chronic liver failure and two or more organ failures. Clin Mol Hepatol 2025;31:1355-1371.
- 12. Song DS, Kim HY, Jung YK, Kim TH, Yim HJ, Yoon EL, et al. Dynamic analysis of acute deterioration in chronic liver disease patients using modified quick sequential organ failure assessment. Clin Mol Hepatol 2024;30:388-405.
Citations
Citations to this article as recorded by
- Correspondence to editorial on “Predictive machine learning model in intensive care unit patients with acute-on-chronic liver failure and two or more organ failures”
Mengyi Zhang, Yee Hui Yeo, Jian Zu, Jonel Trebicka, Fanpu Ji
Clinical and Molecular Hepatology.2026; 32(3): e346. CrossRef
