LEAP-014 added response in metastatic ESCC and no survival

What was tried

LEAP-014 (NCT04949256) was an open-label, randomised Phase 3 study run by Merck Sharp & Dohme in previously untreated metastatic oesophageal squamous cell carcinoma. Actual enrolment was 863 participants. A 13-patient safety run-in (Part 1) preceded the main randomised comparison (Part 2), which assigned 423 participants to pembrolizumab plus lenvatinib plus chemotherapy and 427 to pembrolizumab plus chemotherapy. Chemotherapy was either 5-fluorouracil with cisplatin or paclitaxel with cisplatin. The study opened on 28 July 2021, reached actual primary completion on 8 May 2025, and carried dual primary endpoints of overall survival and progression-free survival by blinded independent central review. The registry status is TERMINATED, with the posted reason recorded as "Trial was rendered as cancelled work ongoing in the book of business", administrative language that reflects close-out after the study had already read out rather than an early stop for harm or futility. Results were presented at the 2025 ESMO Congress in Berlin by Jong-Mu Sun and are posted in full on ClinicalTrials.gov.

The biological hypothesis

Lenvatinib is an approved small molecule recorded in Open Targets as a vascular endothelial growth factor receptor inhibitor acting on FLT1, FLT4 and KDR. The rationale, set out by the investigators in the published study design, was that antiangiogenic therapy converts an immunosuppressive tumour microenvironment into an immune-supportive one, and that preclinical models of lenvatinib combined with anti-PD-1 showed synergistic antitumour activity. The mechanistic chain has three links: VEGF signalling through KDR sustains abnormal vasculature and recruits suppressive myeloid and regulatory T cell populations, blocking that signalling normalises perfusion and lifts suppression, and a checkpoint inhibitor then converts the restored T cell access into durable tumour control.

The genetic support for that chain in this specific tumour is thinner than the pharmacology suggests. The Open Targets association between KDR and oesophageal squamous cell carcinoma is 0.4434, built from literature evidence (0.800), clinical evidence (0.598) and somatic mutation evidence (0.456). No human genetic association datatype contributes to that score. The association therefore reflects that KDR-directed drugs have been tested in this disease and written about, not that human genetics implicates KDR in causing it.

What actually happened

The lenvatinib arm did not improve overall survival. Median overall survival was 17.6 months (95% CI 15.5 to 19.1) with lenvatinib versus 15.5 months (95% CI 13.5 to 17.2) without, hazard ratio 0.92 (95% CI 0.77 to 1.10), p=0.1852. Progression-free survival was 7.2 months (95% CI 6.8 to 8.4) versus 6.9 months (95% CI 5.8 to 7.0), hazard ratio 0.89 (95% CI 0.75 to 1.04), p=0.0753. Both dual primary endpoints missed.

Response did separate. Objective response rate was 62.2% (95% CI 57.4 to 66.8) versus 54.8% (95% CI 49.9 to 59.6), a difference of 7.3 percentage points (95% CI 0.7 to 13.8), p=0.0152. The PD-L1 CPS 10 or greater subgroup showed the same pattern: median overall survival 18.0 versus 15.8 months, hazard ratio 0.89 (95% CI 0.71 to 1.11), p=0.1490.

Safety was not the limiting factor but it was not free either. Serious adverse events were reported in 224 of 421 treated participants in the lenvatinib arm and 201 of 426 in the control arm, which computes to 53.2% versus 47.2%, a difference of 6.0 percentage points. Grade 3 or higher treatment-related adverse events were reported at 81.2% versus 79.1%, and discontinuation due to an adverse event at 33.3% versus 39.0%, as presented at ESMO 2025.

Failure mechanism, best guess

This reads as an efficacy failure in which a real pharmacodynamic effect failed to convert into survival. Lenvatinib did something: response rate rose by a margin whose confidence interval excluded zero. What it did not do was deepen or extend that benefit enough to move the survival curves once a strong active control was already in place. Pembrolizumab plus platinum chemotherapy is an established first-line standard in this disease, so the trial asked lenvatinib to add value on top of a regimen that already produces high response rates and already reduces tumour bulk. Under those conditions, an incremental increase in early tumour shrinkage has little room to translate into longer life.

A second reading is that the vascular normalisation hypothesis is dose- and window-dependent in ways a fixed-dose combination cannot exploit. Normalisation is transient, and continuous VEGFR blockade at tolerated doses may push past the normalisation window into hypoxia, which is counterproductive for T cell function. The trial as designed could not distinguish between the two readings because no vascular or immune pharmacodynamic biomarker gated entry or dosing.

How to prevent this next time

Endpoint-level data here are strong enough to support a base-rate argument but not a formal Bayesian recalculation, so the levers below are the qualitative ones.

First, treat objective response rate as insufficient licence for Phase 3 in this setting. The surrogacy literature in advanced gastroesophageal cancer has repeatedly found weak trial-level correlation between response endpoints and overall survival, and LEAP-014 is now a direct instance: a response difference with p=0.0152 sat alongside a survival difference with p=0.1852 in the same 850 patients.

Second, require a marker of angiogenic dependency, not just PD-L1. The trial enriched on PD-L1 CPS for the checkpoint component and enriched on nothing for the lenvatinib component. The CPS 10 or greater subgroup behaved like the overall population, which is what one expects when the stratifier is unrelated to the mechanism being added.

Third, run the add-on question at Phase 2 with a randomised control and a survival-relevant readout before committing 850 patients. The LEAP programme produced the same signature in head and neck squamous cell carcinoma before this readout, which means the pattern was observable earlier.

The single highest leverage change would have been to gate Phase 3 entry on a prospectively validated pharmacodynamic marker of VEGF dependency rather than on response rate from a single-arm run-in.

What this means for similar programs

The Claidex graph now records three KDR failures, giving a Mechanism Risk Score of 70 out of 100 (orange band) at the target level. Open Targets lists 75 clinical-stage or approved agents against KDR, which puts the saturation component at its ceiling. The recurring shape across the KDR record is not that VEGFR inhibition does nothing, it is that VEGFR inhibition added to an active immunotherapy backbone keeps producing response gains that do not carry to survival. Any new programme proposing that architecture should state in advance what would have to be true for its combination to break the pattern, and how it will measure that before randomising.

Open questions

Did the response gain concentrate in an identifiable subgroup that the reported stratifiers do not capture, and are archival tissue and vascular imaging available to test that retrospectively? Would intermittent rather than continuous lenvatinib dosing preserve the normalisation window? Does the duration of response analysis, which is posted but not summarised here, show that lenvatinib responses were shorter, which would explain the survival result directly? And is there any biomarker-defined oesophageal squamous population where KDR blockade carries genetic rather than pharmacological support?

Sources

1. • ClinicalTrials.gov record and posted results for NCT04949256, retrieved 2026-08-26. https://clinicaltrials.gov/study/NCT04949256 - Study design and rationale: Future Oncology 2024, LEAP-014: first-line lenvatinib + pembrolizumab + chemotherapy in advanced/metastatic esophageal squamous cell carcinoma.https://- Open Targets Platform, KDR (ENSG00000128052) association with oesophageal squamous cell carcinoma (MONDO_0005580), score 0.4434; 75 clinical-stage or approved agents recorded against KDR; lenvatinib (CHEMBL1289601) mechanism of action. Retrieved 2026-08-26. https://platform.opentargets.org/target/ENSG00000128052 - ESMO 2025 presentation reporting, grade 3 or higher adverse event and discontinuation rates. OncLive, retrieved 2026-08-26. https://www.onclive.com/view/lenvatinib-plus-pembrolizumab-and-chemotherapy-fails-to-improve-os-in-advanced-escc - Surrogate endpoints in Phase 3 randomised trials of advanced gastroesophageal cancer: a systematic review.https://- openFDA FAERS, lenvatinib generic-name query, retrieved 2026-08-26: 32,614 total reports, 28,623 serious, 4,019 with a fatal outcome. https://api.fda.gov/drug/event.json.