As primary sclerosing cholangitis (PSC) progresses, patients can face the potential of a liver transplant. In order to be transplanted with a deceased donor liver, they first need to get on the liver transplant list and then receive an offer of a liver. In most places, the place on the transplant list is determined by a MELD-Na score (Model for End Stage Liver Disease). Some PSC patients find that while their health is deteriorating, their MELD scores don’t reflect that. Fortunately, Dr. Mamatha Bhat and her team at the University Health Network (UHN) in Toronto are working to address that.
In 2021, PSC Partners Canada awarded Dr. Bhat a two-year grant totalling US$60,000 to help support the development of a machine learning algorithm that optimizes outcome predictions for PSC patients. The intention is to have an equitable alternative to the MELD-Na score for liver transplantation waitlists.
Based on Dr. Bhat’s summary report dated February 2024, “In year one (2022-2023), we successfully developed this algorithm, with a good degree of accuracy, by analyzing PSC specific clinical features such as recurrent cholangitis, biliary stenting and concomitant IBD. We validated that the algorithm improves upon the MELD-Na score as a predictor of mortality in waitlisted PSC patients”.
The research considered many factors – sex, race/ethnicity, weight and BMI, as well as white blood cell (WBC), platelets and bilirubin levels. Additional considerations were PSC-related complications such as frequency of cholangitis, superimposed cirrhosis, biliary dysplasia and removal surgeries.
Using available databases, over 4,600 patients were used as a training dataset to predict waitlist outcomes and to refine the machine learning algorithm. The models developed by Dr. Bhat and her team were compared against the current MELD score systems.
The team’s research found that “MELD-Na scoring was the greatest predictor, and the addition of PSC-specific variables significantly contributed to the accuracy of the algorithm’s predictive capabilities. This leads us to conclude that our machine learning algorithm makes more accurate outcome predictions for PSC patients on the waitlist for transplantation than other scoring tools. Our model further supports the conclusion that MELD scoring alone is not sufficient and that the incorporation of disease-specific variables contain valuable prognostic information”.
In year two, the team has continued to refine the machine learning algorithm and improve its prediction accuracy. This included accessing more data and testing the algorithm on a larger patient database.
PSC Partners Canada is proud to support Dr. Bhat and her team as they continue to pursue new ways to improve transplant outcomes for patients living with PSC.
You can download the full report at Final progress report: Machine learning evaluation of liver transplant wait-list prioritization for patients with PSC.
