Bemarituzumab in FGFR2b-overexpressing gastric cancer: an effect estimate that did not hold

What was tried

FORTITUDE-102 (NCT05111626) was an Amgen Phase 1b/3 trial in previously untreated, unresectable locally advanced or metastatic gastric and gastroesophageal junction adenocarcinoma with FGFR2b overexpression. Part 1 was an open-label safety lead-in to set the Phase 3 dose. Part 2 randomized participants, with participants and investigators masked, to bemarituzumab plus chemotherapy (mFOLFOX6 or CAPOX) plus nivolumab or to matched placebo plus the same backbone. The Phase 3 primary endpoint was overall survival among participants whose tumors showed FGFR2b staining of 2+ or 3+ on at least 10 percent of tumor cells. The trial opened on 14 March 2022, enrolled 515 participants, and records a primary completion date of 14 October 2025. No results are posted, and the registry gives the reason for stopping as "Study was terminated by sponsor" (ClinicalTrials.gov, NCT05111626).

Bemarituzumab is an afucosylated humanized IgG1 antibody against the IIIb splice isoform of FGFR2 (ChEMBL CHEMBL4802165, molecule type antibody, maximum phase 3), acquired by Amgen with Five Prime Therapeutics. Its selling point was isoform restriction: rather than inhibiting FGFR kinase activity across the receptor family, it binds only the FGFR2b ectodomain, which was expected to avoid the hyperphosphatemia and nail and skin toxicity that limit pan-FGFR small molecules.

The biological hypothesis

FGFR2b overexpression is a real and reasonably common feature of gastric cancer biology. In central immunohistochemistry prescreening of 3,782 tumors for the sister trial FORTITUDE-101, 37.8 percent were positive at any percentage of 2+/3+ tumor cells and 16.2 percent met the 10 percent threshold (Rha et al., JCO Precision Oncology 2025). Open Targets scores the FGFR2 association with gastric adenocarcinoma at 0.418, built from somatic mutation 0.623, animal model 0.541, literature 0.508 and clinical 0.186, with no germline genetic association contributing at all. The class is well drugged: Open Targets lists 27 clinical-stage molecules against FGFR2, eight of them approved.

The program therefore rested on a somatic, expression-level rationale rather than a human genetic one. That is legitimate in oncology, but it carries a specific liability: protein overexpression by immunohistochemistry is a continuous variable forced into a binary, and the cut point defining the treated population was chosen from earlier data rather than derived from a mechanism.

What actually happened

The effect estimate kept shrinking as the evidence base grew.

FIGHT, the randomized Phase 2 in 155 FGFR2b-selected patients, did not meet its primary endpoint. Median progression-free survival was 9.5 months with bemarituzumab plus mFOLFOX6 against 7.4 with placebo, hazard ratio 0.68 (95% CI 0.44 to 1.04, p=0.073). The authors called the result promising clinical efficacy despite the absence of statistical significance (Wainberg et al., Lancet Oncology 2022).

FORTITUDE-101, the Phase 3 in the same setting without nivolumab, initially looked confirmatory. At a 9 December 2024 cutoff and 11.8 months median follow-up, median overall survival was 17.9 against 12.5 months, hazard ratio 0.61 (95% CI 0.43 to 0.86, p=0.005). At a 20 June 2025 cutoff and 19.4 months follow-up, the same comparison gave 14.5 against 13.2 months, hazard ratio 0.82 (95% CI 0.62 to 1.08), per the ESMO 2025 presentation.

FORTITUDE-102 was halted on 4 November 2025 after an ad hoc internal data monitoring committee analysis. Amgen's head of research and development, Jay Bradner, said "the magnitude of observed efficacy did not meet our standard for an Amgen medicine." No FORTITUDE-102 effect estimate has been released.

Toxicity was never the stated reason, but it shaped the risk-benefit arithmetic. In FIGHT, all-grade corneal adverse events occurred in 51 of 76 bemarituzumab recipients (67 percent) against 8 of 77 on placebo (10 percent), and grade 3 corneal events in 18 bemarituzumab recipients (24 percent) and none on placebo. Amgen has stated that Phase 3 ocular events were consistent with Phase 2 but more frequent and more severe. FAERS holds only 7 serious reports naming bemarituzumab, led by pneumonia and sepsis, which reflects an investigational agent with no marketed exposure rather than a quiet safety profile.

Failure mechanism, best guess

This reads as an efficacy failure produced by an unstable effect estimate rather than by a wrong target. Three observations support that. First, the Phase 2 that licensed the program was formally negative, and its point estimate was carried forward as if it were positive. Second, the Phase 3 overall survival hazard ratio moved from 0.61 to 0.82 across roughly eight additional months of follow-up in the same randomized population, the signature of an early separation that does not hold. Third, the combination trial was stopped on efficacy magnitude by an internal committee rather than a prespecified futility boundary, implying the benefit was smaller than the program needed to carry a high burden of corneal injury.

An alternative explanation is that adding nivolumab diluted the FGFR2b effect. That cannot be evaluated without the FORTITUDE-102 data.

How to prevent this next time

Endpoint-level FORTITUDE-102 data was never released, so the available data supports three quantitative checks rather than a full Bayesian model.

Base-rate discipline on the screening funnel. At 16.2 percent prevalence, randomizing the 324 patients in the FORTITUDE-101 primary analysis set implies prescreening roughly 2,000 patients (324 divided by 0.162). A funnel that steep commits enormous fixed cost before any interim, which raises the bar for acting on an equivocal Phase 2.

Treat a p=0.073 Phase 2 as what it is. The FIGHT hazard ratio of 0.68 had an upper confidence bound of 1.04, and advancing on that point estimate without shrinking it toward the null propagated an optimistic prior into two Phase 3 trials.

Prespecify follow-up duration for the survival readout. The FORTITUDE-101 estimate changed materially between 11.8 and 19.4 months of follow-up. Locking the decision analysis to a minimum follow-up rather than an event count reached early would have exposed the instability before a second Phase 3 was fully enrolled.

The single highest leverage change would have been requiring FIGHT to be repeated or extended to a statistically significant progression-free survival result before committing two parallel Phase 3 trials to a point estimate whose confidence interval already crossed one.

What this means for similar programs

The Claidex graph now holds two FGFR2 failures, the earlier Phase 1b/2 strategic termination in FGFR2b-overexpressing solid tumors and this one, which moves the target Mechanism Risk Score to 54 (orange band). The saturation component sits near its ceiling at 14.93 of 15, reflecting 27 clinical-stage FGFR2 programs. Teams pursuing FGFR2b in gastric cancer should assume the expression cut point, not the antibody, is the open variable. A recent preprint argues the same point, framing FGFR2b as a target not yet turned into a reliable biomarker. Programs selecting patients on ligand-independent overexpression should also plan for on-target epithelial toxicity in tissues that depend on the same receptor, since corneal epithelium turnover is FGFR2b dependent.

Open questions

What were the FORTITUDE-102 survival estimates at termination, and did the nivolumab arm behave differently from FORTITUDE-101?

Does a higher FGFR2b cut point recover a durable effect, and how small does that population become?

Were corneal events dose limiting in a way that capped exposure below the level needed for tumor effect?

Is FGFR2 amplification by sequencing a better predictor than protein staining, given that amplification and fusion align with overexpression but do not fully overlap?

Sources

1. ClinicalTrials.gov. NCT05111626, FORTITUDE-102. https://clinicaltrials.gov/study/NCT05111626 Wainberg ZA, Enzinger PC, Kang YK, et al. Bemarituzumab in patients with FGFR2b-selected gastric or gastro-oesophageal junction adenocarcinoma (FIGHT): a randomised, double-blind, placebo-controlled, phase 2 study. Lancet Oncology 2022;23(11):1430-1440. https://Rha SY, Zhang Y, Elme A, et al. Prevalence of FGFR2b Protein Overexpression in Advanced Gastric Cancers During Prescreening for the Phase III FORTITUDE-101 Trial. JCO Precision Oncology 2025;9:e2400710. https://OncLive. Bemarituzumab/Chemo Combo Translates to OS Advantage in FGFR2b+ Gastric/GEJ Cancer. Reporting the FORTITUDE-101 presentation at ESMO Congress 2025. https://www.onclive.com/view/bemarituzumab-chemo-combo-translates-to-os-advantage-in-fgfr2b-gastric-gej-cancer The ASCO Post. Primary Analysis Shows Survival Benefit With Bemarituzumab Plus Chemotherapy in Gastric or Gastroesophageal Junction Cancer. 10 December 2025. https://ascopost.com/issues/december-10-2025/primary-analysis-shows-survival-benefit-with-bemarituzumab-plus-chemotherapy-in-gastric-or-gastroesophageal-junction-cancer/ Fierce Biotech. Amgen halts gastric cancer trial after antibody "did not meet our standard" of efficacy, exec says. 4 November 2025. https://www.fiercebiotech.com/biotech/amgen-halts-gastric-cancer-trial-after-antibody-did-not-meet-our-standard-efficacy Open Targets Platform. FGFR2 (ENSG00000066468) and gastric adenocarcinoma (MONDO_0005036). Association score 0.418. Queried 12 September 2026. https://platform.opentargets.org/evidence/ENSG00000066468/MONDO_0005036 ChEMBL. Bemarituzumab, CHEMBL4802165. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL4802165/ openFDA FAERS drug event endpoint, queried 12 September 2026 for medicinalproduct "bemarituzumab". Satala C, Gurau G, Patrichi G, et al. Fibroblast Growth Factor Receptor 2b (FGFR2b) in Gastric Cancer: The Challenge of Turning a Target into a Reliable Biomarker. Preprints.org, 6 May 2026. https://.