Pull up the randomized Phase 2 data on evorpacept in HER2-positive gastric cancer, published in Nature Medicine, and the number that jumps is the hazard ratio reported in the trial. That is not a noise signal. In a disease where second-line survival gains are measured in weeks, cutting the hazard of progression or death by half with a CD47 myeloid checkpoint inhibitor layered onto trastuzumab, ramucirumab, and paclitaxel is a result that should force a protocol design conversation across every oncology sponsor planning a gastric cancer program. Most of those sponsors are not having that conversation yet, because they are reading this trial as a drug story. It is actually a biomarker architecture story.

Three developments in oncology trial design over the past year are converging on a pattern the field has not yet named clearly: the emergence of myeloid checkpoint biomarkers as a second-tier precision medicine layer in HER2-driven tumors. The evorpacept Phase 2 result is the sharpest signal. But it does not stand alone, and understanding why it connects to earlier signals changes what sponsors should be doing in protocol design right now.

The Biomarker Layer Nobody Enrolled For

The RAINBOW trial established the baseline. In a subgroup analysis of patients with HER2-positive gastric or gastroesophageal junction cancer who had received prior trastuzumab therapy, ramucirumab plus paclitaxel produced outcomes that were numerically better than in the overall RAINBOW population, but the sample was 39 patients, too thin to drive clinical strategy. Sponsors filed RAINBOW away as a hypothesis. The evorpacept Phase 2 was designed, at least in part, to test whether adding a CD47 blockade on top of that backbone could push the survival curve further.

It did. The combination of evorpacept with trastuzumab, ramucirumab, and paclitaxel, presented at the 2025 ASCO Gastrointestinal Cancer Symposium, demonstrated favorable response rates and that hazard ratio of 0.50 compared to the trastuzumab-ramucirumab-paclitaxel triplet alone. What the headline results obscure is the mechanism doing the work. CD47, the “don’t eat me” signal expressed on tumor cell surfaces, blocks macrophage-mediated phagocytosis. Evorpacept strips that signal away, and in a tumor microenvironment already primed by anti-HER2 therapy and anti-VEGFR2 blockade, macrophage activation becomes a meaningful effector arm. The biology is elegant. The trial design problem it exposes is not.

The problem: none of the patients in this Phase 2 were prospectively enrolled based on CD47 expression levels. Research published in Frontiers in Immunology confirms that CD47 protein expression is demonstrably higher on tumor cells than on stromal cells across multiple solid tumor indications, including breast cancer, head and neck squamous cell carcinoma, and colorectal cancer, suggesting CD47 expression is not uniformly distributed and carries predictive signal. Running an unselected Phase 2 and observing a HR of 0.50 tells you the average treatment effect across a mixed CD47-expression population. It does not tell you what the effect is in a CD47-high subgroup, and it does not tell you what the effect is in a CD47-low subgroup. That gap will haunt the Phase 3 design.

What the Adaptive Design Playbook Gets Wrong Here

Here is the counterintuitive reading that most trial designers will resist: a strong Phase 2 result in an unselected population is sometimes worse for Phase 3 than a modest result in a biomarker-selected population. The unselected HR of 0.50 looks compelling enough to proceed without a biomarker enrichment strategy. Sponsors and their statisticians will power a Phase 3 on that effect size, enroll broadly, and then discover mid-trial that the signal is concentrated in the CD47-high tertile. By then, the trial is too large to pivot, the interim analysis has been pre-specified around an overall population estimate, and the Data Monitoring Committee has no adaptive rule to trigger enrichment.

The ICH E20 draft guideline on adaptive designs, which the FDA adopted and about which Advarra published a detailed review of the FDA’s 2024 DMC guidance, addresses exactly this failure mode. The guidance framework envisions pre-specified adaptive rules for population enrichment based on biomarker data accumulated during the trial. But those rules must be written into the protocol before the first patient is enrolled. A sponsor who reads the evorpacept Phase 2 data and moves straight to a confirmatory trial without embedding a CD47 biomarker enrichment rule is building a Phase 3 that the guidance architecture was specifically designed to prevent.

The DESTINY-Gastric01 trial is the cautionary comparison. Published in the New England Journal of Medicine, DESTINY-Gastric01 enrolled HER2-positive patients and produced a compelling response rate for fam-trastuzumab deruxtecan, but the trial still used a binary HER2-positive enrollment gate rather than a continuous HER2-expression enrichment strategy. The result: the drug works across the HER2-positive population, but the magnitude of benefit varies with HER2 expression levels in ways the trial was not powered to characterize cleanly. The evorpacept program has an opportunity to avoid that limitation by building a dual biomarker gate, HER2-positive confirmed plus CD47 expression level documented, into the Phase 3 protocol from day one. The window to make that design decision is closing as the program moves toward confirmatory study planning.

The Operational Reckoning for Sponsors, CROs, and Sites

For sponsors, the immediate operational implication runs through the companion diagnostic pathway. A CD47 expression enrichment strategy requires a validated assay that sites can run on archival tumor tissue or fresh biopsy at screening. No such assay is known to currently carry regulatory qualification for gastric cancer. Building one in parallel with Phase 3 planning adds substantial time to the pre-IND timeline and requires a biomarker development agreement conversation with FDA before the Phase 3 IND is submitted. Sponsors who treat this as a Phase 3 problem will find it has become a Phase 3 delay.

CROs inheriting this program face a site feasibility problem that the HR of 0.50 obscures. HER2-positive gastric cancer represents a minority of gastric cancer diagnoses, already a constrained enrollment pool., already a constrained enrollment pool. Layer a CD47 expression threshold on top and the screened-to-enrolled ratio climbs steeply, demanding more sites to achieve the same enrollment rate. CROs that built their gastric cancer site networks around HER2-only enrollment criteria will need to renegotiate site capacity estimates and screening budgets, and some of their legacy sites in community oncology settings lack the molecular pathology infrastructure to run a second biomarker at screening.

For technology vendors, specifically the EDC and RTSM providers supporting precision oncology trials, this trend creates a design requirement that most platform templates do not yet handle cleanly: dual biomarker stratification with an adaptive enrichment rule tied to interim biomarker readout. A trial that starts enrolling broadly across HER2-positive patients and then enriches toward CD47-high patients after an interim analysis requires a randomization algorithm that can shift stratum weights mid-trial, a feature that some legacy RTSM platforms implement only with custom development. Vendors who are not investing in adaptive stratification logic now will lose this class of program to competitors who are.

The 12-month forward view is uncomfortable for anyone who expected the evorpacept Phase 2 data to simply accelerate the path to a confirmatory trial. If the Phase 3 is designed without a CD47 biomarker strategy and the trial enrolles an unselected HER2-positive population, the probability of replicating a HR of 0.50 in a larger, more heterogeneous cohort drops materially. If the FDA reviewer opens the Phase 3 protocol and finds no pre-specified enrichment rule despite publicly available data suggesting CD47 expression is a potential effect modifier, that reviewer will have a legitimate basis to request a Type B meeting before the trial starts. The sponsors who move fastest on companion diagnostic development and adaptive enrichment protocol design will not just run better trials. They will run the only trials that FDA is likely to accept without a complete redesign request.

References

  1. Nature Medicine, “Evorpacept plus trastuzumab, ramucirumab and paclitaxel in HER2-positive gastric cancer: a randomized phase 2 trial”
  2. Cancer Network, “Evorpacept Elicits Favorable Responses in HER2+ Gastric/GEJ Cancer,” 2025 ASCO GI presentation
  3. Future Oncology, RAINBOW subgroup analysis, prior trastuzumab cohort, ramucirumab plus paclitaxel outcomes
  4. Frontiers in Immunology, CD47 protein expression on tumor cells vs. stromal cells in solid tumors
  5. PubMed Central / NEJM, DESTINY-Gastric01 trial, fam-trastuzumab deruxtecan in HER2-positive gastric cancer
  6. Advarra, “Understanding FDA’s 2024 Draft Guidance on DMCs,” including adaptive design and ICH E20 context
Website |  + posts

Moe Alsumidaie is Chief Editor of The Clinical Trial Vanguard. Moe holds decades of experience in the clinical trials industry. Moe also serves as Head of Research at CliniBiz and Chief Data Scientist at Annex Clinical Corporation.