Machine learning to predict waitlist dropout among liver transplant candidates with hepatocellular carcinoma.

Background: Accurate prediction of outcome among liver transplant candidates with hepatocellular carcinoma (HCC) remains challenging. We developed a prediction model for waitlist dropout among liver transplant candidates with HCC. Methods: The study included 18,920 adult liver transplant candidates in the United States listed with a diagnosis of HCC, with data provided by the Organ Procurement and Transplantation Network. The primary outcomes were 3-, 6-, and 12-month waitlist dropout, defined as removal from the liver transplant waitlist due to death or clinical deterioration. Results: Using 1,181 unique variables, the random forest model and Spearman's correlation analyses converged on 12 predictive features involving 5 variables, including AFP (maximum and average), largest tumor size (minimum, average, and most recent), bilirubin (minimum and average), INR (minimum and average), and ascites (maximum, average, and most recent). The final Cox proportional hazards model had a concordance statistic of 0.74 in the validation set. An online calculator was created for clinical use and can be found at: http://hccli verca lc.cloud medxh ealth.com/. Conclusion: In summary, a simple, interpretable 5-variable model predicted 3-, 6-, and 12-month waitlist dropout among patients with HCC. This prediction can be used to appropriately prioritize patients with HCC and their imminent need for transplant. [ABSTRACT FROM AUTHOR]