Neural network model predicts liver transplant waitlist mortality
Researchers developed a prediction model using neural networks that outperformed the MELD-Na score in the identification of liver transplants waitlist mortality, according to research presented at The Liver Meeting Digital Experience.
In a press release, Shinji Nagai, MD, a transplant surgeon at Henry Ford Hospital, said that the MELD-Na score-based allocation model, although useful clinically, has limitations.
“We have seen many liver cirrhosis patients whose MELD scores that were low but suffered from life-threatening complications due to liver cirrhosis and actually could not have a chance of a liver transplant,” he said.
Nagai and colleagues sought to use neural networks to develop a model that more accurately predicted waitlist mortality.
The investigators collected data from the OPTN/UNOS registry comprising 194,299 patients who were listed for liver transplantation between 2002 and 2018. They used a data subset to create four neural network models constructed to predict mortality at 30, 90, 180 and 365 days. Researchers used 44 variables, including recipient characteristics, trend of liver and kidney function during wait time and registration year.
The developers split the data into training, validation and test datasets and assessed the models using area under receiver operating curve (AUC-ROC) and area under precision-recall curve (PR-AUC).
Nagai and colleagues found that the model showed the AUC-ROC for 30-day, 90-day, 180-day and 365-day mortality was 0.949, 0.928, 0.915 and 0.899, respectively, while the PR-AUC was 0.689, 0.73, 0.769 and 0.823, respectively.
The 90-day mortality model outperformed the MELD score for both AUC-ROC and PR-AUC. It also did better in recall, negative predictive value and F-1 score. Specifically, the 90-day mortality model identified more waitlist deaths with a higher recall of 0.833 vs. 0.308 (P < .001).
Additionally, the 90-day mortality model outperformed MELD scores across subsets separated based on ethnicity, sex, region, age, diagnosis group and year of listing.
“In the future, if these advanced technologies are introduced into the liver allocation system, liver waitlist ranking would better reflect patients’ medical urgency and this should lead to lower waitlist mortality,” Nagai said in the release.