Stratifying post-transplant HCC recurrence risk: a two-centre externally validated predictive model

Abstract

Objective: HCC recurrence after liver transplantation (LT) happens in about 10-30% of the patients. A user-friendly, accurate predictive model allows individualized surveillance in high risk population for early detection of recurrence. Method: Retrospective cohort of consecutive HCC patients undergoing LT in Queen Mary Hospital (QMH) were recruited. Data randomization to training and validation set was performed. Independent factors identified by multivariate analysis in training were used to derive a predictive formula, which was subsequently validated internally using QMH validated set and externally by National Taiwan University Hospital (NTUH) with concordance statistics Results: There were 465 patients from QMH and NUTH recruited. Multivariate analysis in the QMH training set (183 patients) identified three independent factors associated with post-LT HCC recurrence, namely, alpha-fetoprotein (AFP) over 400 ng/ml (P=0.012, HR 2.92); sum of maximum tumour size and number (P=0.013, HR 1.15) and 3. salvage LT (P=0.033, HR 2.08). The derived scoring model showed good predictability to post-LT recurrence (c-stat: 0.75). Internal validation using 147 separate patient dataset demonstrated high discrimination ability (c-stat: 0.85). Validation set patients were classified into low (0 to 9), moderate (10 to 14) and high-risk groups (over 14) accordingly, and the risk of HCC was respectively 4%, 22%, 62% (c-stat 0.811). The total risk score continued to demonstrate satisfactory performance with the use of NTUH patient dataset (external validation) with c-stat of 0.75. Using the same stratification model, the recurrence risk in each NTUH group was 7%,31% and 75% respectively (c-stat: 0.70). The Chi-square goodness-of-fit test showed no significant discrepancy between the expected and observed recurrence risk with the stratification model (P=0.55). Conclusion: A reliable and user-friendly scoring model for the prediction of post-LT HCC recurrence was derived, internally and externally validated.