External validation of the SNAPP score for hepatocellular carcinoma recurrence after living donor liver transplantation

Background: Binary morphologic criteria cannot capture biologic heterogeneity among living donor liver transplantation (LDLT) recipients. SNAPP score (tumor size, number, AFP, PIVKA-II, PET/CT) yields a graded recurrence-risk estimate but lacks independent validation. We externally validated SNAPP for hepatocellular carcinoma (HCC) recurrence after LDLT. Methods: SNAPP was applied without refitting in 230 LDLT recipients with HCC at two Korean centers and benchmarked against Milan, French-AFP, Metroticket 2.0, MoRAL, RETREAT, and three LDLT-derived selection systems (Kyoto, Medanta LDLT, and National Cancer Center Korea). Discrimination, calibration, Brier score, and reclassification were assessed using IPCW-weighted methods with bootstrap 95% CIs. Results: Among 230 patients (median follow-up, 91.5 mo), 49 (21.3%) recurred. SNAPP achieved discrimination comparable to the best pretransplant tool, MoRAL (Harrell C-index, 0.756; 95% CI, 0.690–0.821), and exceeded Milan (0.503; Holm-adjusted p =0.01). Discrimination was numerically higher than Metroticket 2.0 (0.605) and three LDLT-derived systems (C-index difference, 0.10-0.15), but none of these differences survived correction for multiplicity (Holm-adjusted p =0.24). RETREAT, a posttransplant benchmark, reached 0.807. Calibration-in-the-large was −0.068; slope, 0.59 (95% CI, 0.43–0.77). Within Milan-eligible patients (n=217), SNAPP resolved a monotonic, approximately 5.9-fold gradient in 5-year cumulative recurrence (11.9%, 48.8%, 70.0%). Reclassification favored SNAPP over Milan, French-AFP, and Metroticket 2.0; the center-adjusted hazard ratio per point 1.79 (95% CI, 1.51–2.14). Conclusions: SNAPP showed discrimination comparable to MoRAL and exceeded Milan, with numerically higher but not definitively superior discrimination compared with Metroticket 2.0, French-AFP, and three LDLT-specific selection systems, while resolving biologic heterogeneity within Milan-eligible candidates using only pretransplant variables. With near-zero calibration-in-the-large but a slope below unity, SNAPP supports risk stratification at listing, though individualized probability estimates require recalibration.