Development and validation of a prognostic risk score for hepatocellular carcinoma recurrence post–liver transplant: Insights from the UNOS database.

Abstract

4125 Background: For over two decades, the criteria for liver transplantation (LT) in hepatocellular carcinoma (HCC) have been well-established, yet recurrence remains a major clinical challenge. This recurrence contributes to inferior post-LT survival in HCC patients compared to those without HCC. Prognostic index could serve as a valuable tool to identify patients who may benefit from adjuvant therapies and guide standardized post-LT HCC surveillance, which currently varies across transplant centers. Methods: We developed and validated a predictive model using the United Network for Organ Sharing (UNOS) database, analyzing adult liver transplant recipients with hepatocellular carcinoma (HCC) from 2009 to 2024, including 4,970 patients. Univariable analysis identified variables associated with 1-year, 3-year, and 5-year post-transplant HCC recurrence, applying a strict p-value threshold ( < 0.01). Significant variables were selected for multivariable logistic regression to build the model. Internal validation was performed for each time point using Receiver Operating Characteristic (ROC) curve analysis and confusion matrix evaluation, with the best-performing model selected based on these metrics. Results: The most significant model was derived using 3-year recurrence as the outcome. The final model included the pre-transplant Model for End-Stage Liver Disease (MELD) score (p = 0.02), worst tumor histology grade (p < 0.001), and total tumor diameter (p = 0.03). The final logistic regression equation is as follows: log(1-p/p) = −3.9630 − 0.0621 × Initial MELD score + 0.7657 × Worst tumor histology grade + 0.1084 × Total tumor diameter. Internal validation results showed an AUC of 0.761 (95% CI: 0.718 - 0.804), accuracy of 0.769 (95% CI: 0.757 - 0.7807), sensitivity of 77.2%, and specificity of 65.0% for 1-year recurrence. For 3-year recurrence, the model demonstrated an AUC of 0.733 (95% CI: 0.702 - 0.763), accuracy of 0.6696 (95% CI: 0.6563 - 0.6827), sensitivity of 66.7%, and specificity of 70.9%. For 5-year recurrence, the AUC was 0.714 (95% CI: 0.685 - 0.743), with accuracy of 0.655 (95% CI: 0.641 - 0.668), sensitivity of 65.2%, and specificity of 68.6%. Conclusions: We have developed and validated a predictive model for HCC recurrence following LT using UNOS data. This model shows promising performance, particularly for predicting 1-year recurrence, and may potentially serve as a useful tool for guiding post-transplant management and surveillance strategies.