Identifying candidates with favorable prognosis following liver transplantation for hepatocellular carcinoma: Data mining analysis.

Abstract

The optimal cutoff of each value in configuring selection criteria for pre-transplant assessment of hepatocellular carcinoma (HCC) remains uncertain. To build a predictive model for recurrent HCC, we performed data mining analysis on patients who underwent LT for HCC at University Health Network (n = 246). The model was externally validated using a cohort from the Scientific Registry of Transplant Recipients (SRTR) database (n = 9,769). Among 246 patients, 14.6% (n = 36) experienced recurrent HCC within 2.5 years post-LT. The risk prediction model for recurrent HCC identified two subgroups with low-risk (total tumor diameter [TTD] <4 cm and serum alpha-fetoprotein [AFP] <73 ng/ml, n = 135) and with high-risk (TTD >4 cm and/or AFP >73 ng/ml, n = 111). The reproducibility of the model was validated through the SRTR database; overall patient survival rate was significantly better in low-risk group than high-risk group (P < 0.0001). Using Cox regression model, this yardstick, not Milan criteria, was revealed to efficiently predict post-transplant survival independent of underlying characteristics (P < 0.0001). Grouping LT candidates with pre-LT HCC by the cutoffs of TTD 4 cm and AFP 73 ng/ml which were unearthed by data mining analysis efficiently classify patients according by the post-transplant prognosis.