LEAP-012 bought 4.6 months of progression-free survival in liver cancer and returned a hazard ratio of 0.98 for death

What was tried

LEAP-012 (NCT04246177, protocol 7902-012) was a randomized, quadruple-masked, active-controlled phase 3 study run by Merck Sharp and Dohme LLC with Eisai Inc. It opened on 22 May 2020 across 137 sites in 33 countries and randomized 480 participants one to one. Eligible patients had unresectable, non-metastatic hepatocellular carcinoma not amenable to curative treatment but amenable to transarterial chemoembolization, with ECOG 0 or 1 and Child-Pugh class A liver function. The experimental arm received TACE plus oral lenvatinib (12 mg daily at body weight 60 kg or above, 8 mg below) plus intravenous pembrolizumab 400 mg every six weeks for up to two years, against TACE plus matched oral and intravenous placebo. Co-primary endpoints were progression-free survival by blinded independent central review using RECIST 1.1 modified for the study, and overall survival, both in the intention-to-treat population. Primary completion was 19 August 2025. The registry record was updated on 3 September 2026 with an overall status of TERMINATED, a stop reason of "Business Reasons", and posted results.

The biological hypothesis

TACE induces tumour ischaemia, and the resulting hypoxia drives VEGF release, neoangiogenesis and recruitment of immunosuppressive myeloid cells into the treated liver. Lenvatinib inhibits KDR, the VEGFR2 receptor tyrosine kinase, along with VEGFR1 and 3, FGFR1 to 4, PDGFRA, KIT and RET. The rationale for LEAP-012 was that blocking the post-embolization angiogenic rebound while releasing PD-1 mediated T-cell suppression would convert a locoregional debulking procedure into durable control. Open Targets scores the KDR and hepatocellular carcinoma association at 0.678, with a clinical datatype score of 0.980 and a literature score of 0.945. The genetic association datatype contributes zero. That profile describes a target validated by drug approvals and publication volume rather than by human genetics. KDR is heavily tractable, annotated with approved small molecule and approved antibody modalities, a high-quality pocket and structures with ligand.

What actually happened

The trial met one co-primary endpoint and missed the other. Median progression-free survival by RECIST 1.1 was 14.6 months (95% CI 12.6 to 16.7) against 10.0 months (95% CI 8.1 to 12.2) with TACE alone, hazard ratio 0.66 (95% CI 0.51 to 0.84), p equal to 0.0002. The mRECIST analysis agreed at a hazard ratio of 0.63 (95% CI 0.49 to 0.81). Objective response rate by mRECIST was 71.3% against 49.4%, a difference of 21.5 points (95% CI 12.7 to 30.0). Overall survival did not follow. At the January 2024 interim cutoff published in the Lancet, the overall survival hazard ratio was 0.80 (95% CI 0.57 to 1.11), one-sided p equal to 0.087. In the final posted analysis, median overall survival was 40.4 months (95% CI 34.4 to 47.0) with the combination and 40.8 months (95% CI 35.2 to 45.8) with TACE alone, hazard ratio 0.98 (95% CI 0.76 to 1.26), p equal to 0.4442. Merck and Eisai announced on 29 October 2025 that overall survival had not reached statistical significance, judged the likelihood of reaching it later to be low, and closed the study. Toxicity was substantial. Serious adverse events occurred in 122 of 237 treated participants in the combination arm against 62 of 241 on TACE alone, while deaths were 126 of 237 against 127 of 243. Serious hepatic encephalopathy occurred in 12 against 2, and among all-cause events hypertension occurred in 126 against 53 and proteinuria in 105 against 27.

Failure mechanism, best guess

The surrogate did what the pharmacology predicted, and it did not translate. VEGFR2 blockade after embolization suppressed radiographic regrowth, delaying progression by 4.6 months on the median and nearly doubling the confirmed response rate. What it did not change was how long patients lived. Two explanations fit the posted data and are not mutually exclusive. The first is competing risk. In Child-Pugh A hepatocellular carcinoma treated locoregionally, much of the mortality is driven by cirrhosis and hepatic decompensation rather than tumour progression, so a therapy that delays the radiographic event can leave the survival curve untouched. The six-fold excess of serious hepatic encephalopathy in the combination arm is consistent with that reading. The second is post-progression crossover to effective systemic therapy, since control-arm patients who progressed earlier had more time on subsequent lines. The 0.80 to 0.98 drift in the overall survival hazard ratio between the 2024 interim and the final analysis is the observable signature of this. This record is an efficacy failure, not a strategic termination, despite the registry stop reason.

How to prevent this next time

The posted endpoint data support one quantitative check that costs nothing. Compare the interim and final overall survival hazard ratios against the progression-free survival hazard ratio. Progression-free survival moved from 0.66 to a final 0.66 while overall survival moved from 0.80 to 0.98. A programme that treats an immature overall survival hazard ratio with a confidence interval crossing one as directional support is reading noise, and LEAP-012 puts a number on the error: the point estimate moved 0.18 with follow-up. Three levers follow. First, stratify and power for liver-related mortality separately from tumour-related mortality, because the two compete and only one is addressable here. Second, pre-specify post-progression therapy collection and a rank-preserving structural failure time analysis, so a null survival result can be attributed rather than reported. Third, weight the safety delta into the go decision earlier. A doubling of serious adverse events against an unchanged death count is a benefit-risk fact available long before the final analysis. The single highest leverage change would have been powering overall survival on a liver-function stratified population with pre-specified competing-risk analysis, so the trial could distinguish a therapy that fails to extend life from a population in which cirrhosis, not tumour progression, sets the survival ceiling.

What this means for similar programs

The mechanism risk score for KDR now stands at 75 of 100, in the red band, across all four KDR claims in the Claidex graph. Phase burden contributes 24.09 of a possible 30, archetype severity 19.87 of 25, recency 10.98 of 15, genetic deficit 4.83 of 15 and target saturation the full 15 of 15. Saturation is saturated in the literal sense, since Open Targets annotates 75 distinct clinical candidates against KDR, many of them approved. The genetic deficit term is the lowest of the five, which is the honest reading that this target is not undervalidated. It is overworked. The pattern across the four KDR claims is consistent. Lenvatinib failed in melanoma brain metastases at phase 2 and in metastatic oesophageal squamous cell carcinoma at phase 3, rivoceranib stopped in adenoid cystic carcinoma at phase 2, and now a phase 3 hepatocellular combination has separated on progression-free survival and not on survival. Sponsors running TACE combination trials should assume the same competing-risk ceiling.

Open questions

Was the survival null driven by liver-related death, and does the cause-of-death dataset support that partition. What proportion of control-arm participants received subsequent systemic therapy, and does adjustment recover any separation. Does the same dissociation appear in EMERALD-1, which shares the TACE plus antiangiogenic plus checkpoint architecture. Given that 347 of 480 participants were Asian, how much of the result reflects regional differences in TACE technique and subsequent care. Does any progression-free survival gain here justify a doubling of serious adverse events when survival is unchanged.

Sources

1. ClinicalTrials.gov, NCT04246177, LEAP-012 record and posted results retrieved 7 September 2026. https://clinicaltrials.gov/study/NCT04246177.

2. Kudo M, et al. Transarterial chemoembolisation combined with lenvatinib plus pembrolizumab versus dual placebo for unresectable, non-metastatic hepatocellular carcinoma (LEAP-012): a multicentre, randomised, double-blind, phase 3 study. Lancet. 2025. PMID: 39798578.

3. Merck. Merck and Eisai Provide Update on Phase 3 LEAP-012 Trial in Unresectable, Non-Metastatic Hepatocellular Carcinoma. 29 October 2025. https://www.merck.com/news/merck-and-eisai-provide-update-on-phase-3-leap-012-trial-in-unresectable-non-metastatic-hepatocellular-carcinoma/.

4. Open Targets Platform, KDR (ENSG00000128052) and hepatocellular carcinoma (MONDO_0007256), association, tractability and clinical candidate data retrieved 7 September 2026. https://platform.opentargets.org/target/ENSG00000128052.

5. ChEMBL, LENVATINIB CHEMBL1289601, retrieved 7 September 2026. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL1289601/.

6. openFDA drug event endpoint, generic name lenvatinib, retrieved 7 September 2026. https://api.fda.gov/drug/event.json.

7. Kudo M, et al. Lenvatinib plus Pembrolizumab Combined with Transarterial Chemoembolization in Intermediate-Stage Hepatocellular Carcinoma. Liver Cancer. 2025. PMID: 40144474.

8. Singal AG, et al. Impact of LEAP-012 and EMERALD-1 in the management of HCC. JHEP Reports. 2026. PMID: 41503570.

9. Claidex failure graph entries lenvatinib-kdr-vegfr2-metastatic-escc-leap-014-phase3-survival-failure, lenvatinib-pembrolizumab-kdr-melanoma-brain-metastases-phase2-efficacy-failure and rivoceranib-kdr-vegfr2-adenoid-cystic-carcinoma-rm-202-phase2-strategic-termination.