Abstract
BACKGROUND AND AIMS: Increasing utilization of extended criteria donor leads to an increasing rate of early allograft failure after liver transplantation. However, consensus of definition of early allograft failure is lacking. METHODS: A retrospective, multicenter study was performed to validate the Liver Graft Assessment Following Transplantation (L-GrAFT) risk model in a Chinese cohort of 942 adult patients undergoing primary liver transplantation at three Chinese centers. L-GrAFT (L-GrAFT(7) and L-GrAFT(10)) was compared with existing models: the Early Allograft Failure Simplified Estimation (EASE) score, the model of early allograft function (MEAF), and the Early Allograft Dysfunction (EAD) model. Univariate and multivariate logistic regression were used to find risk factors of L-GrAFT high-risk group. RESULTS: L-GrAFT(7) had an area under the curve of 0.85 in predicting 90-day graft survival, significantly superior to MEAF [area under the curve (AUC=0.78, p=0.044)] and EAD (AUC=0.78, p=0.006), while there was no statistical significance between the predicting abilities of L-GrAFT(7) and EASE (AUC=0.84, p>0.05). Furthermore, L-GrAFT(7) maintains good predicting ability in the subgroup of high-donor risk index (DRI) cases (AUC=0.83 vs. MEAF, p=0.007 vs. EAD, p=0.014) and recipients of donors after cardiac death (AUC=0.92 vs. EAD, p<0.001). Through multivariate analysis, pretransplant bilirubin level, units of packed red blood cells, and the DRI score were selected as independent risk factors of a L-GrAFT(7) high-risk group. CONCLUSIONS: The accuracy of L-GrAFT(7) in predicting early allograft failure was validated in a Chinese multicenter cohort, indicating that it has the potential to become an accurate endpoint of clinical practice and transitional study of machine perfusion.