Hepatocellular carcinoma (HCC) is the sixth most common cancer and the third leading cause of cancer-related death worldwide, with approximately 865,000 new cases and 758,000 deaths annually [1]. Liver transplantation (LT) removes both the tumor and the diseased liver. Since Mazzaferro et al. in 1996, the Milan criteria—a single tumor ≤5 cm or up to three tumors each ≤3 cm, without macrovascular invasion or extrahepatic spread—have been the benchmark for selection, with 5-year survival above 70% [2].

Strict adherence to the Milan criteria excludes many patients who might benefit from transplantation. Expanded criteria followed, such as the University of California San Francisco (UCSF) criteria, the Metroticket 2.0 model, and Asian criteria including Asan, Kyoto (5-5-500), and Seoul National University Hospital (SNUH) [3–5]. In parallel, downstaging—reducing tumor burden with locoregional therapy (LRT) to meet transplant criteria—has emerged as a way to expand candidacy [6, 7].

The evidence for downstaging comes predominantly from deceased donor liver transplantation (DDLT). The UCSF downstaging protocol, the 2017 United Network for Organ Sharing/Organ Procurement and Transplantation Network (UNOS/OPTN) national policy, and the XXL randomized trial all showed that successfully downstaged patients reach post-transplant survival close to that of patients within Milan criteria from the start [7–9], and meta-analyses of these mostly DDLT data report 5-year survival of 60%–80% [10–12].

Living donor liver transplantation (LDLT) is a different paradigm. It accounts for most liver transplants in Asia, where deceased donation remains low [13, 14]. LDLT does not depend on a shared allocation system, so centers have more freedom in selection and timing; many Asian centers accept patients beyond conventional limits based on alpha-fetoprotein (AFP), des-gamma-carboxy prothrombin (DCP), and response to pretransplant therapy [15, 16]. A healthy donor also assumes surgical risk, which makes a clear account of expected oncological outcome an ethical requirement.

We therefore reviewed and pooled post-transplant oncological outcomes after locoregional downstaging before LDLT for HCC beyond Milan criteria, with a pre-specified primary analysis restricted to studies targeting standard radiologic criteria (Milan or UCSF). We asked whether the oncological signal that justifies downstaging in DDLT survives the greater selection freedom of LDLT, and which pooled outcome is most comparable across this clinically diverse literature.

This review asks a narrow question: what happens after transplant when a patient beyond Milan criteria responds to locoregional therapy and then receives a living donor graft? Three features make it worth asking. First, a healthy donor accepts an operation for one named recipient, and no allocation rule constrains listing, so each center must judge the expected oncological return for itself. Second, response-based selection may work differently when centers are free to accept any responder; whether the signal that justifies downstaging survives that freedom is the central question. Third, recurrence varied less between studies (I² = 55.7%) than 5-year OS (I² = 77.2%) or 5-year RFS (I² = 74.7%). We therefore treat recurrence as the most comparable signal, not a stable one—in the smaller primary cohort it varied more (22.1%; I² = 81.4%)—and the pooled survival estimates need more caution. We analyzed 511 downstaged LDLT recipients in 17 studies, nine meeting the primary definition.

These numbers describe patients who had already succeeded. Downstaging is both a treatment and a filter: patients who progress, respond poorly, or decompensate are usually not transplanted, so they never enter the studies we pooled. Across the 1,186 patients in the 12 studies that reported how many started downstaging, only 47.5% succeeded. The good outcomes we report therefore reflect, at least in part, the selection of tumors with favorable biology. They are not proof that downstaging itself produced the benefit.

It matters which question a pooled number answers. What happens to everyone who starts downstaging requires an intention-to-treat design that counts the dropouts; our data cannot answer that. They answer a second question—what happens to a patient who has already responded and reached transplant—which is the answer a team needs when a responding beyond-Milan candidate has a living donor. The numbers should not be read as evidence that downstaging beats no downstaging. Twelve of the 17 studies reported how many patients started, but none reported what happened to those who failed, and none included beyond-Milan patients transplanted without successful downstaging. Future studies should enroll patients when downstaging begins and report every dropout with its cause and outcome.

Waiting time changes what a good response means, and LDLT has almost none. UNOS/OPTN policy requires at least 6 months of observation after successful downstaging—the “ablate-and-wait” principle—during which aggressive tumors declare themselves and leave the list. A deceased donor candidate therefore passes two tests, response to treatment and stability over time. LDLT removes the second: donor availability rather than a queue sets the timing, and transplant may follow within weeks. Response to downstaging then becomes the main, and often the only, window onto tumor biology. Two things follow. First, how a tumor responds matters more than the bare fact of response. Durability across serial scans, the AFP trend, and des-gamma-carboxy prothrombin (PIVKA-II) all carry information; a recent single-center series showed that both markers predict long-term outcome after LDLT for HCC [20]. Our own data fit this reading: neither the target criteria (p = 0.44) nor the downstaging modality (p = 0.55) separated recurrence risk. Second, a short deliberate wait may help when the response is recent, marginal, or achieved only after several treatments, because it would restore part of the biological filter that LDLT removes. This has not been tested and deserves prospective study, particularly in the Asian LDLT-predominant programs that produced most of our cohorts.

Our survival estimates sit close to the deceased donor downstaging literature. Yao et al. reported 5-year post-transplant survival of 77.8% in the UCSF experience [7]. Mehta et al. reported 3-year survival of 79% in the UNOS-DS group and 71% in the “all-comers” group, against 83% for patients always within Milan criteria [9]. Parikh et al. pooled 24 mostly DDLT studies and reported 5-year survival of 60%–80% [10], and Tan et al. reviewed 25 studies using the UNOS downstaging criteria [11]. Our pooled 5-year OS of 75.0% falls inside that range, whereas our recurrence rate of 25.0% sits above the 16%–20% those series report [7, 10]. These cohorts differ in selection, protocol, and follow-up, and we made no direct comparison; the data show broad similarity, not equivalence, between donor types. The overlap in survival fits the view that tumor biology and treatment response drive outcome more than graft type does [21].

Heterogeneity is the main objection to any downstaging meta-analysis, and we designed our subgroup analyses around it. By modality, recurrence was 30% for radiotherapy combined with LRT, 23% for LRT alone, and 17% for ICI-based regimens (p = 0.55); across very different treatments, outcome appears to be determined by the selection of tumors that respond rather than by the technique used [22, 23]. By target criteria, biology-based or expanded criteria showed 28% recurrence against 21% for Milan or UCSF (p = 0.44). Sensitivity analyses agreed with the main result: excluding studies under 10 patients gave recurrence 25.1%, 5-year OS 74.0%, and 5-year RFS 69.0%; excluding studies under 20 patients gave recurrence 24.2%; and excluding ICI-based protocols changed nothing (Supplementary Table S3). The pre-specified restricted analysis gave a recurrence rate of 22.1%, close to the overall 25.0%, so including higher-risk study types did not inflate the pooled estimate; heterogeneity for survival nevertheless stayed high even in that cohort, so survival appears to track center-level differences in patient selection more than modality or donor type. We did not pool by individual technique—chemoembolization, radioembolization, ablation, or radiotherapy—because in 11 of the 17 studies patients received combined or sequential treatments and outcomes were not reported by technique; labeling each cohort by a “main” modality would have required an arbitrary rule. Supplementary Table S4 lists the full treatment composition of every cohort.

These findings matter most where LDLT is the main route to transplant, as in South Korea, Japan, Taiwan, and India [13, 14]. DDLT has a standardized UNOS/OPTN downstaging framework; LDLT has none, and our results support an LDLT-specific consensus that uses biological markers alongside imaging criteria. ICI-based downstaging needs particular attention. The two included studies reported pathological response rates of 40%–50%, but one reported allograft rejection in half the recipients [24] and the other wound healing problems without rejection [25]; washout, immunosuppression, and long-term rejection risk remain unknown. AFP is built into several expanded LDLT criteria, but too few included studies reported it systematically for us to test its value as a selection marker.

Three features of the evidence limit how far these results travel. First, sixteen of the 17 cohorts were Asian, one was European, and none came from the Americas. LDLT practice, hepatitis etiology, tumor biology, and organ access differ, so Western programs should apply our estimates with caution. Second, no included study reported outcomes by graft type or operative detail—graft side and size, graft-to-recipient weight ratio, or extent of hilar dissection—so our finding that modality did not determine outcome (p = 0.55) says nothing about surgical factors. Third, the field needs standardization more than another single-center series: uniform mRECIST response assessment, an agreed definition of downstaging success, mandatory reporting of the intention-to-treat denominator and of dropout, routine explant pathology, and fixed-interval oncological endpoints.

This study has several limitations. First, and most important, every outcome is conditional on successful downstaging and transplant. No study reported what happened to patients who failed, progressed, or dropped out, so our estimates describe responders who reached transplant, not an intention-to-treat population, and both selection bias and immortal time bias apply. Second, every included study was retrospective and single-center. Third, downstaging criteria, modalities, and definitions of success varied widely; subgroup and sensitivity analyses suggest this did not meaningfully move the recurrence estimate, but heterogeneity for pooled survival stayed high, and individual locoregional techniques could not be compared because most cohorts received combined or sequential treatment (Supplementary Table S4). Fourth, the studies were small, with a median of 25 patients, although excluding studies under 20 and under 10 patients gave consistent results (Supplementary Table S3). Fifth, our main survival syntheses pooled proportions at fixed time points, which ignores censoring; reconstructed patient-level data gave closely matching estimates (5-year OS 75.7%, RFS 75.6%), but reconstruction was possible only where usable curves existed and relied on time-point rates where risk tables were missing. Sixth, explant pathology (microvascular invasion, differentiation, necrosis), AFP, and DCP were reported inconsistently, so radiologic response could not be linked to true tumor burden; systematic explant reporting would most directly validate downstaging. Seventh, one investigator performed selection and extraction, with pilot calibration and cross-checking. Finally, publication bias cannot be excluded, although Egger’s test was not significant.

After locoregional downstaging before LDLT for HCC beyond Milan criteria, pooled 5-year OS was 75.8% and recurrence 22.1% in the pre-specified primary analysis, and across all 17 studies recurrence (25.0%) varied least between studies, making it the most comparable oncological signal available. Successful downstaging is a filter as well as a treatment—fewer than half of the patients who attempted it succeeded—and LDLT removes the waiting interval that filters tumors a second time in the deceased donor setting. Selection should therefore rest on how durable and how biologically convincing the response is, rather than on meeting a size-and-number rule alone. Prospective multicenter intention-to-treat studies with standardized protocols and endpoints are needed before routine clinical application.