The Checkpoint Combination Trap: Why Oncology Trials Keep Winning Battles the FDA Doesn’t Care About

An oncology medical director at a mid-size biotech pulls up her Phase III readout. The combination arm beat the control. Statistically significant. The hazard ratio looks clean. She sends the data package to her regulatory team expecting a clear path to an NDA. What comes back is a request for more information on the contribution of effect from each individual agent, a question the trial was never designed to answer.

This scenario has played out often enough across oncology portfolios that it now has the shape of a structural problem, not a one-off regulatory surprise. Three converging signals in checkpoint inhibitor development point toward what I’d call the Contribution Gap: a widening disconnect between how sponsors design combination immunotherapy trials and what FDA actually requires to establish approvability. The industry has spent a decade mastering the biology of dual checkpoint blockade. The regulatory science of how to prove each drug earns its place in the regimen has not kept pace.

The Approval Architecture Nobody Reads Carefully

Open the FDA’s draft guidance on combination cancer drug development and the central demand is unambiguous: sponsors must demonstrate the “contribution of effect” of each individual drug in the regimen. The guidance covers novel combinations specifically, describing an evidence standard for how much work each agent should demonstrably do before FDA will credit the combination as a whole.

Most sponsors read this as a pharmacology problem, something solvable with dose-escalation data and mechanistic rationale. FDA means it as a clinical evidence problem. The difference between those two readings is the difference between a successful BLA and a complete response letter.

The 2022 approval of Opdualag (nivolumab plus relatlimab-rmbw) for unresectable or metastatic melanoma illustrates the right model. Bristol Myers Squibb ran RELATIVITY-047 as a randomized trial that compared nivolumab plus relatlimab directly against nivolumab monotherapy, designed to isolate the LAG-3 inhibitor’s marginal contribution to progression-free survival. The design answered the contribution question before FDA had to ask it. That is not the design most dual checkpoint programs use.

Most programs run combination versus chemotherapy, or combination versus a historical control, generating an impressive headline hazard ratio that proves the combination works but leaves regulators unable to attribute how much of the benefit belongs to each agent. The trial succeeds on its own terms. The approval package then hits a question the data cannot answer.

The Signal BioNTech’s HPV Program Sent

BioNTech’s BNT113 program offers the clearest recent illustration of how combination trial design can carry hidden regulatory risk even when the biology is compelling. The Phase II/III trial combined BNT113, an mRNA immunotherapy targeting HPV16 antigens, with pembrolizumab for metastatic HPV16-positive head and neck squamous cell carcinoma. The scientific rationale was sound: a tumor-specific mRNA vaccine priming T-cell responses while pembrolizumab sustains them by blocking PD-1-mediated exhaustion. Two mechanisms, coordinated, potentially synergistic.

The design question was always whether the trial could isolate what the mRNA component added to pembrolizumab’s established single-agent activity in this indication. Pembrolizumab already carries an FDA approval in head and neck squamous cell carcinoma. That means any combination trial runs directly into the contribution-of-effect requirement from the other direction: FDA needs evidence that BNT113 adds something pembrolizumab alone does not deliver. Without a pembrolizumab monotherapy control arm, the combination’s hazard ratio over chemotherapy cannot answer that question cleanly.

This structural design gap recurs across the checkpoint combination landscape. A published meta-analysis comparing randomized controlled trials and real-world evidence studies on checkpoint inhibitor efficacy found no statistically significant differences in progression-free survival or overall survival between the two evidence types for advanced NSCLC and melanoma patients. That finding should reframe how sponsors think about trial design risk: if real-world outcomes track RCT outcomes this closely, the competitive advantage no longer comes from running the largest trial. It comes from running the trial whose architecture survives regulatory scrutiny on contribution of effect.

What the Industry Keeps Getting Wrong

Here is the counterintuitive read on where this trend leads. The conventional assumption inside oncology development is that stronger efficacy data insulates a program from regulatory problems. Post an overall survival benefit with a hazard ratio below 0.65, the thinking goes, and the approval follows. That logic holds when a single agent is seeking its first indication. It fails when a combination is seeking approval and FDA cannot attribute the survival benefit to the components.

The FDA’s posture on contribution of effect is not a new policy preference. It is an explicit requirement in the combination drug guidance. What has changed is the density of programs now hitting this requirement in the same review cycle. With thousands of active oncology combination trials registered globally, FDA review divisions are increasingly encountering submissions that run the same design error at scale: a combination versus chemotherapy architecture that generates impressive efficacy but cannot answer the contribution question for either agent.

For sponsors, the operational implication is a protocol amendment risk they are systematically underpricing. Adding a monotherapy control arm mid-program is not just a site burden. It resets the statistical analysis plan, triggers IRB amendments across potentially dozens of sites, and in some cases requires additional regulatory submissions that extend the development timeline. The cost of building contribution-of-effect evidence into the original protocol design is a fraction of the cost of trying to reconstruct it at the BLA stage.

CROs face a different version of the same problem. The standard oncology trial operating model optimizes for enrollment velocity and data quality against a fixed endpoint set. Few CROs’ standard protocol review processes are known to include a systematic check against the FDA combination guidance contribution-of-effect language. That means the design error often passes through three or four operational reviews before anyone raises the regulatory architecture question, and by that point, the first patient has already been dosed.

Technology vendors in the eClinical space are sitting on an underutilized asset here. Adaptive trial design platforms that allow real-time arm modification could, in principle, let sponsors add a contribution-of-effect evaluable arm after interim analysis without fully resetting the trial. The FDA’s existing framework for adaptive designs, articulated in the 2019 guidance “Adaptive Designs for Clinical Trials of Drugs and Biologics,” contemplates pre-specified adaptations of this kind. The combination of adaptive design methodology and contribution-of-effect planning represents the cleaner path forward, but very few oncology sponsors appear to be currently building trials at that intersection.

In the next 12 to 18 months, the first casualty of the Contribution Gap will be a high-profile dual checkpoint combination with a Phase III hazard ratio below 0.60 that receives a complete response letter asking for contribution-of-effect data the trial was never designed to generate. When that letter becomes public, the entire field of sponsors who built their regulatory strategy around the Opdualag approval without replicating its design will have to explain to investors why they are running a confirmatory trial that should have been the original trial. The sponsors who pre-specified their contribution analysis in 2024 and 2025 will have their approval, their label, and their market position. Everyone else will have a very clean hazard ratio and no clear path to use it.

References

1. Nature Reviews Drug Discovery, “Checkpoint inhibitor combinations: the trial-to-approval gap”
2. FDA, “Development of Cancer Drugs for Use in Novel Combination: Determining the Contribution of Individual Drugs to Effects”
3. Cancer Research Institute, “FDA Approves New LAG-3 Checkpoint Inhibitor Combo (Opdualag)”
4. MarketBeat, “BioNTech Q3 Earnings Call Highlights, BNT113 Program Update”
5. PMC / NCBI, “Meta-analysis: RCT vs. Real-World Evidence for Checkpoint Inhibitor Efficacy in NSCLC and Melanoma”