AI model improves risk stratification for liver transplant patients

Sept. 30, 2025

A new machine learning model based on electrocardiogram (ECG or EKG) is giving transplant surgeons a better prediction of severe liver disease and likelihood of liver-related death that has the potential to improve risk stratification for liver transplant patients. This new model, AI-Cirrhosis-ECG (ACE), picks up on other indicators of disease better than the current standard, the Model for End-Stage Liver Disease (MELD) score, alone.

Mayo Clinic transplant physicians led the development of the ACE model, publishing their findings in JHEP Reports in 2024. Here they share more about this innovative approach and what it could mean for the future of risk stratification for transplant patients.

Innovating on innovation: Going beyond the MELD score

The MELD score, pioneered at Mayo Clinic in the 2000s, offered a major improvement over previous methods for risk stratification for liver transplant patients. This score used objective laboratory tests to provide a clear rating and stratification.

Unfortunately, some patients end up at a severe disadvantage when only the MELD score is used because conditions such as ascites, hepatic encephalopathy and portal hypertension are not necessarily captured in this model.

"The MELD score is an incredible tool, but patients with low MELD scores and significant complications of portal hypertension are at a severe disadvantage," says Doug A. Simonetto, M.D., a transplant hepatologist at Mayo Clinic in Rochester, Minnesota. "The MELD score doesn't accurately reflect how sick these patients are and their risk of death. In comparison, the ACE model performed significantly better since it is likely capturing complications not included in the MELD score."

The ACE model is a logical improvement on the standard since it is also an objective measure of liver health based on ECGs. The deep learning model maintains objectivity while capturing other markers that, when combined with the MELD score, may improve transplant prioritization for those who are the sickest.

Training the AI model, collaboratively

As Mayo Clinic continues investing in technology and AI innovation, the timing of the 2024 ACE model study was perfect. Dr. Simonetto had previously collaborated with colleagues in cardiology to develop an AI model leveraging ECGs to detect cirrhosis. This study, published in the American Journal of Gastroenterology in 2022, showed that ECG findings were associated with the presence and severity of chronic liver disease.

This proof of concept led to a follow-up study, still in progress, using a machine learning model trained on ECGs to see whether it could capture early signals of cirrhosis. Using a cohort of over 75,000 patients, the team trained a model that can distinguish between patients with cirrhosis and those without.

"After controlling for comorbidities and other factors, we found strong evidence that a 12-lead EKG could tell if someone has cirrhosis," says Dr. Simonetto. "Our model output is a continuous value from 0 to 1. A score of close to zero suggests low risk of disease, while a higher output carries a stronger association with the presence of more-severe cirrhosis."

In fact, in the 2022 study, Dr. Simonetto found a progressive increase in the output of the model predicting more-severe liver disease that peaked at the time of transplant — before dramatically dropping after the transplant as patients' health improved.

As a next step, Dr. Simonetto and his team are continuing to validate the model with more-diverse patient populations and integrating it with other established predictors of liver disease before rolling it out clinically.

A future of better risk stratification

Eventually, this model could be implemented not only for risk stratification but also potentially for organ allocation as well. While Dr. Simonetto doesn't expect the ACE model to replace MELD score, he does foresee it being incorporated with the MELD score to optimize organ allocation.

The ACE model can be a helpful, low-cost tool for stratification, especially since all patients going through transplant evaluation receive cardiac workups, including ECGs.

"We are continuing to innovate and find new ways to prioritize transplants, including by capturing the severity of disease with machine learning models," says Bashar A. Aqel, M.D., a transplant hepatologist and the director of the Mayo Clinic Transplant Center in Phoenix, Arizona. "We plan to continue expanding the utility of this model to predict future complications."

Andrew P. Keaveny, M.D., a transplant hepatologist at Mayo Clinic in Jacksonville, Florida, concurs. "When a physician encounters a complicated or serious case, this is what we do each and every day at Mayo Clinic," he says. "With the ACE model, we have the potential to improve the timing of interventions, optimize transplant listing decisions and ultimately enhance patient outcomes. We're constantly thinking of how we can continue to improve with these types of innovations to support patients."

For more information

Ahn JC, et al. AI-Cirrhosis-ECG (ACE) score for predicting decompensation and liver outcomes. JHEP Reports. 2025;19:101356.

Ahn JC, et al. Development of the AI-Cirrhosis-ECG score: An electrocardiogram-based deep learning model in cirrhosis. The American Journal of Gastroenterology. 2022;173:424.

Refer a patient to Mayo Clinic.