We retrospectively reviewed consecutive adult recipients (aged ≥18 years) who underwent LDLT for HCC at a single high-volume center between December 2005 and December 2022. Eligibility required viable HCC confirmed on the final explant specimen. Patients with a complete pathologic response, pediatric recipients, recipients of deceased-donor whole-liver grafts, patients undergoing 2:1 (dual-donor) LDLT, and those with incomplete follow-up data were excluded. After applying these criteria, the final cohort comprised 1,573 recipients, of whom 316 (20.1%) underwent ABOi and 1,257 (79.9%) ABOc transplantation. Clinical, operative, and pathologic variables were obtained from a prospectively maintained database and medical records. The study was approved by the Institutional Review Board of [Institution blinded for review], which waived the requirement for informed consent because of the retrospective design and the use of de-identified data, and was conducted in accordance with the Declaration of Helsinki and the Declaration of Istanbul.

ABOi LDLT was performed using an institutional rituximab-based desensitization protocol described previously [6]. In brief, a single dose of rituximab (375 mg/m²) was given 2–3 weeks before transplantation, and pretransplant plasma exchange was performed as needed to reduce isoagglutinin titers to the institutional target. Local graft infusion and splenectomy, used in earlier eras, were not part of the current protocol. Maintenance immunosuppression in both groups was based on tacrolimus, mycophenolate mofetil (MMF), and a corticosteroid taper, with basiliximab induction used selectively. However, ABOi recipients received more intensive early immunosuppression than ABOc recipients: in addition to higher early tacrolimus trough targets used to reduce the risk of AMR, MMF was added more frequently as part of the institutional ABOi protocol. Serial quantitative measures of immunosuppression exposure were not uniformly available across the study period.

Tumor burden was classified by the Milan criteria on explant pathology, and microvascular invasion (MVI) was assessed on the explant by dedicated hepatobiliary pathologists. Serum AFP and protein induced by vitamin K absence-II (PIVKA-II) were recorded at transplantation. Pre-transplant HCC treatment was defined as any bridging or downstaging locoregional therapy before transplantation. To capture tumor biology, tumor burden was additionally quantified using the ADV score, a pre-specified biomarker-based measure validated across hepatic resection and liver transplantation [19–23]. The ADV score is the product of serum AFP (ng/mL), des-gamma-carboxyprothrombin (PIVKA-II, mAU/mL), and tumor volume (mL), expressed on a base-10 logarithmic scale, capturing both biological activity and anatomic extent in a single continuous variable that complements the dichotomous Milan criteria. We applied the previously reported post-transplant cutoff of 5log (corresponding to an ADV product of 100,000), above which the risk of HCC recurrence is substantially increased [19, 20]. After transplantation, surveillance was protocol-driven and identical for ABOi and ABOc recipients. AFP and PIVKA-II were measured at every outpatient visit (monthly during the first year, then every 3 months). Chest and abdomen-pelvis computed tomography were obtained in all recipients at 3 months; thereafter, recipients within the Milan criteria were imaged every 6 months to 3 years, annually to 5 years, and every 2 years thereafter, and those beyond the Milan criteria every 3 months during the first year, every 4–6 months to 3 years, annually to 5 years, and every 2 years thereafter. Imaging frequency was therefore determined by tumor burden, not ABO compatibility. Recurrence was diagnosed primarily by cross-sectional imaging; histologic confirmation was obtained only when clinically indicated and was not routinely required. The primary endpoint, recurrence-free survival (RFS), was the interval from transplantation to first documented recurrence or last follow-up, with patients dying without recurrence censored at death. Overall survival (OS) was defined as the interval from transplantation to death from any cause.

Continuous variables are presented as medians with interquartile ranges (IQR) and were compared using the Mann-Whitney U test; categorical variables are presented as counts with percentages and were compared using the chi-square or Fisher exact test. Recurrence-free and overall survival were estimated using the Kaplan-Meier method and compared using the log-rank test. Hazard ratios (HR) with 95% confidence intervals (CI) were estimated using Cox models; the multivariable RFS model included ABO-incompatibility, recipient age and sex, MELD score, pre-transplant treatment, Milan status, tumor number and largest size, MVI, and log10-transformed AFP and PIVKA-II, with proportional hazards assessed by scaled Schoenfeld residuals. To address the baseline imbalance in tumor burden, 1:2 nearest-neighbor propensity score matching without replacement was performed with a caliper of 0.2 SD of the logit of the propensity score, estimated by logistic regression on tumor and clinical covariates; balance was assessed by standardized mean differences (SMD), with values below 0.10 acceptable. Pre-transplant treatment was included as a covariate rather than as a matching variable to preserve power in the small beyond-Milan subgroup. Stratified analyses by tumor burden were pre-specified. The Milan-based analysis was performed in both the whole and propensity-matched cohorts, whereas the continuous ADV score was analyzed in the whole cohort, where the larger number of high-burden events preserved statistical power. A multiplicative ABO-by-tumor-burden interaction term was fitted in the whole cohort for each measure. Given their exploratory nature, no correction for multiple comparisons was applied. An exploratory mediation analysis used the 1-month tacrolimus concentration as the mediator, with bootstrap confidence intervals. As sensitivity analyses, recurrence was additionally analyzed in a competing-risk framework in which death without recurrence was treated as a competing event, with cumulative incidence functions estimated using the Aalen-Johansen method and subdistribution hazard ratios using Fine-Gray models; alternative ADV cut-offs (3 to 6log) and stratification by transplant era were also examined. All tests were two-sided, and P < 0.05 was considered statistically significant. Analyses were performed using Python (version 3.11) with the lifelines and scikit-learn libraries.

Of the 1,573 adults with viable HCC on explant pathology, 316 (20.1%) underwent ABOi and 1,257 (79.9%) ABOc transplantation, and the median follow-up was 104 months. The groups were similar in recipient age, sex, underlying liver disease (predominantly hepatitis B virus), graft type, GRWR, AFP, PIVKA-II, and MVI (Table 1). However, ABOi recipients consistently had a lower baseline tumor burden, with a lower MELD score (median 9 vs. 10; P < 0.001), fewer tumors (median 1 vs. 2; P = 0.010), a smaller largest tumor (2.1 vs. 2.3 cm; P = 0.028), a smaller total tumor volume (7.9 vs. 11.2 cm³; P = 0.005), and a lower proportion of tumors beyond the Milan criteria (21.2% vs. 29.1%; P = 0.006); they were also more likely to have received pre-transplant locoregional therapy (83.5% vs. 77.3%; P = 0.020). This more favorable tumor profile in the ABOi group, probably reflecting more selective use of ABOi transplantation motivated the matched and stratified analyses.

HCC recurred in 329 of the 1,573 patients (20.9%) during follow-up, comprising 260 ABOc and 69 ABOi recipients. Crude recurrence was similar between the groups (21.8% vs. 20.7%), with 1-, 3-, and 5-year RFS of 89.8%/81.3%/78.2% versus 89.1%/82.2%/80.0%, respectively (log-rank P = 0.64; Figure 1A). In univariable analysis, ABO-incompatibility was not associated with recurrence (HR 1.06, 95% CI 0.82–1.39; Table 2). After adjustment for clinical and tumor factors, the point estimate increased toward higher risk (HR 1.19, 95% CI 0.91–1.55; P = 0.21), a shift that becomes apparent only after the more favorable baseline tumor burden of the ABOi group is taken into account. Beyond-Milan status, MVI, pre-transplant treatment, tumor number and size, AFP, and PIVKA-II were the dominant independent predictors of recurrence (Table 2).

To address the baseline imbalance in tumor burden, a 1:2 propensity-matched cohort was constructed (ABOi 315: ABOc 630), with excellent covariate balance (all SMD <0.05; Table 3). In the matched cohort, RFS again did not differ between groups (5-year RFS 78.1% vs. 80.5%; log-rank P = 0.50; HR 1.11, 95% CI 0.82–1.48), confirming the absence of an overall adverse effect of ABO-incompatibility (Figure 2A).

Overall survival was also similar between the groups. A total of 410 deaths occurred during follow-up (333 ABOc and 77 ABOi), and the 1-, 3-, and 5-year overall survival rates were 95.6%/85.7%/81.2% in the ABOc group and 95.3%/86.4%/81.4% in the ABOi group (log-rank P = 0.91). Thus, in the whole cohort, ABO-incompatibility compromised neither recurrence-free nor overall survival, consistent with contemporary ABOi LDLT being oncologically safe on average.

The effect of ABO-incompatibility differed according to tumor stage (Table 4). Among the 1,140 patients within the Milan criteria, 152 recurrences occurred (119 ABOc and 33 ABOi), and recurrence was nearly identical between groups (ABOi 13.3% vs. ABOc 13.4%; 5-year RFS 86.7% vs. 87.0%; log-rank P = 0.93; adjusted HR 1.00, 95% CI 0.68–1.47; Figure 1B), and this equivalence was maintained in the propensity-matched cohort (Figure 2B). Among the 433 patients beyond the Milan criteria, 177 recurrences occurred (141 ABOc and 36 ABOi), and recurrence was significantly higher in ABOi recipients (53.7% vs. 38.5%; 5-year RFS 45.7% vs. 63.0%; log-rank P = 0.008; adjusted HR 1.45, 95% CI 1.00–2.09; P = 0.049; Figure 1C). This difference persisted in the propensity-matched cohort (ABOi 53.7% vs. ABOc 40.2%; log-rank P = 0.028; Figure 2C). The multiplicative ABO-by-Milan interaction term was directionally consistent but not significant (HR 1.39; P = 0.23), reflecting the limited events in this subgroup (36 recurrences in 67 patients); the stratum-specific estimates are therefore best interpreted as a consistent gradient.

Because the Milan criteria reflect only tumor size and number, we next performed a stratified analysis using a complementary, pre-specified measure of tumor biology, the AFP-DCP-volume (ADV) score, which combines serum tumor markers with tumor volume, applying the established post-transplant cutoff of 5log [19, 20]. The results were directionally consistent with the Milan-based analysis (Table 5; Figure 3). Among patients with an ADV score below 5log, recurrence was nearly identical between groups (ABOc 14.0% vs. ABOi 13.9%; log-rank P = 0.94). Among patients with an ADV score of 5log or higher, recurrence was higher in ABOi recipients (64.0% vs. 44.6%; 5-year RFS 37.6% vs. 55.7%; log-rank P = 0.022), although this difference attenuated after adjustment for microvascular invasion and pre-transplant treatment (HR 1.31, 95% CI 0.88–1.93; P = 0.18). The continuous ADV-by-ABO interaction term approached significance (P = 0.088), and overall survival did not differ within either ADV stratum (ADV≥5log, log-rank P = 0.29). Thus, the tumor-burden-dependent effect of ABO-incompatibility was observed with an independent, biomarker-based measure of tumor burden as well as with the Milan criteria, although it remained statistically borderline. In a competing-risk analysis treating death without recurrence as a competing event, the findings were unchanged (beyond Milan: subdistribution HR 1.54, 95% CI 1.08–2.19; ADV ≥5log: 1.58, 95% CI 1.07–2.32; Supplementary Table S1). The direction of effect was consistent across ADV cut-offs of 3 to 6log, reaching significance only at the pre-specified 5log threshold (Supplementary Table S2), and the results did not differ by transplant era (ABO-by-era interaction P = 0.52; Supplementary Table S3).

Recurrences occurred early in both groups. Among patients who recurred, the median time to recurrence was similar (ABOi 12.6 vs. ABOc 11.7 months; P = 0.37), and approximately 77% of all recurrences occurred within 24 months in both groups (76.8% vs. 77.3%). ABO-incompatibility was therefore not associated with earlier recurrence. The absolute early-recurrence burden was highest in beyond-Milan ABOi recipients (24-month recurrence 46.3% vs. 30.9%), identifying the main target for intensified surveillance.

To explore the proposed mechanism, we examined tacrolimus trough concentrations at 1 and 3 months after transplantation, available in 1,532 and 1,526 patients, respectively. Consistent with the more intensive ABOi protocol, ABOi recipients had higher levels than ABOc recipients at both 1 month (9.5 vs. 9.1 ng/mL; P < 0.001) and 3 months (8.9 vs. 8.5; P < 0.001). A higher 1-month tacrolimus concentration was independently associated with recurrence in the whole cohort (adjusted HR 1.09 per ng/mL, 95% CI 1.01–1.19; P = 0.03). This association was largely confined to high tumor burden, being apparent beyond the Milan criteria (HR 1.12; P = 0.06) and at ADV ≥ 5log (HR 1.11; P = 0.09) but not within Milan (HR 1.05; P = 0.39), although the tacrolimus-by-Milan interaction was not significant (P = 0.54). In an exploratory mediation analysis, the 1-month tacrolimus concentration accounted for only a small, statistically imprecise fraction of the ABO-incompatibility effect, modestly attenuating its estimate when added to the model. The association was weaker in the smaller propensity-matched cohort (HR 1.09; P = 0.12).