Profesionales médicos

AI-based ACE score accurately predicts decompensation and liver outcomes in patients diagnosed with cirrhosis

Jan. 07, 2026

Determining accurate estimates of disease severity and prognosis is an essential but challenging part of care management in patients diagnosed with cirrhosis. Having this information can facilitate the timely implementation of preventive therapies and help optimize organ allocation among candidates for liver transplantation (LT).

Pioneered at Mayo Clinic in the 2000s, the Model for End-Stage Liver Disease (MELD) score offered a major improvement over previous methods for risk stratification for liver transplant patients. This score uses objective laboratory tests to provide a clear rating and stratification. Unfortunately, because conditions such as ascites and hepatic encephalopathy related to portal hypertension are not necessarily captured in this model, some patients end up at a severe disadvantage when evaluated by the MELD score alone.

Recognizing the potential of artificial intelligence (AI)-based tools, Mayo Clinic researchers conducted a study to evaluate whether a deep learning-based AI-Cirrhosis-ECG (ACE) score could detect hepatic decompensation and predict clinical outcomes in cirrhosis. The results of that study were published in JHEP Reports in 2025.

"Advances in artificial intelligence that have occurred over the last several years have allowed clinicians and researchers to explore novel ways of addressing a variety of knowledge gaps in patient care," explains Doug A. Simonetto, M.D., a gastroenterologist, transplant hepatologist and researcher at Mayo Clinic in Rochester, Minesota. "This study investigates a novel machine learning tool that leverages routine 12-lead electrocardiograms to predict outcomes in patients with advanced liver disease."

Methods

The researchers analyzed 2,166 ECGs from three groups of patients diagnosed with compensated cirrhosis:

  • 472 patients seen at Mayo Clinic and identified retrospectively, called the retrospective Mayo Clinic cohort.
  • 420 patients seen at Mayo Clinic and identified prospectively while they were on the waitlist for LT, called the prospective transplant cohort.
  • 341 patients at Hospital Clínic de Barcelona while undergoing evaluation for LT, called the external validation cohort.

The researchers assessed the ACE score performance using receiver-operating characteristic analysis for decompensation detection and competing risks Cox regression for outcome prediction.

Results

"After controlling for other factors, we found that our machine learning model, AI-Cirrhosis-ECG (ACE), can accurately estimate severity of disease and predict the risk of liver-related death in patients with cirrhosis," explains Dr. Simonetto. "The ACE model's ability to predict mortality was comparable to the Model for End-Stage Liver Disease (MELD). However, when combined with MELD, the ACE-MELD performance was superior to the individual models. This suggests that the ACE model is capturing disease features not accounted for in the MELD score, potentially improving prognostication for these patients."

"After controlling for other factors, we found that our machine learning model, AI-Cirrhosis-ECG (ACE), can accurately estimate severity of disease and predict the risk of liver-related death in patients with cirrhosis."

— Doug A. Simonetto, M.D.

Study data demonstrating the performance of the ACE model include the following:

  • The ACE score showed high accuracy in detecting hepatic decompensation, with an area under the curve (AUC) of 0.933; 95% confidence interval (CI); a sensitivity of 88.0%; and a specificity of 84.3% at an optimal threshold of 0.25.
  • In the multivariable analysis, each 0.1-point increase in ACE score was independently associated with an increased risk of liver-related death, hazard ratio (HR) of 1.44, 95% CI.
  • Adding the ACE score to the MELD-sodium score significantly improved the prediction of adverse outcomes across all cohorts (c-statistics: retrospective cohort, 0.903 versus 0.844; prospective cohort, 0.779 versus 0.735; external validation cohort, 0.744 versus 0.732).

Next steps

Dr. Simonetto and colleagues are currently in the process of validating these findings across a broader and more-diverse range of patient populations. "Additionally, we are examining the model's performance using point-of-care electrocardiograms — both single- and six-lead ECGs — which can be administered at patients' homes to monitor disease progression," says Dr. Simonetto.

For more information

Ahn JC, et al. AI-Cirrhosis-ECG (ACE) score for predicting decompensation and liver outcomes. JHEP Reports. 2025;7:101356.

Refer a patient to Mayo Clinic.