Whether the FDA will accept a marketing application supported by a pre-specified interim analysis of the primary endpoint — taken before every participant has completed follow-up — directly shapes trial design, timeline, and go-to-market strategy. For regulatory and clinical teams, knowing where the agency has actually accepted this approach, and under what statistical safeguards, is the difference between a defensible development plan and a costly late-stage surprise.
This analysis lays out the regulatory basis the FDA applies to prospectively planned interim analyses and Type I error control, then catalogs the concrete approvals that relied on such designs across therapeutic areas. It focuses on the design features and conditions the agency has repeatedly found acceptable, with each point tied back to source documents.
Want Rhizome's help on your own question? Try it for free.
FDA precedents for approving on a pre-specified interim analysis of the primary endpoint
A recurring question in trial design is whether the FDA will accept a marketing application built on a pre-specified interim analysis of the primary endpoint, taken before every participant has completed follow-up. The answer is yes, provided the interim look is prospectively planned and the study controls Type I error across all analyses. This design has repeatedly supported approvals, most heavily in oncology, and increasingly in large cardiovascular and metabolic outcomes trials. Below is how the FDA frames the practice and the concrete approvals that relied on it.
The regulatory basis
FDA guidance is explicit that prospectively planned interim analyses of a primary endpoint can support a regulatory decision or marketing application, but only when the design is pre-specified and the overall false-positive (Type I error) rate is controlled 4051. Repeatedly testing a primary endpoint at the conventional one-sided 0.025 level would inflate Type I error, so the interim and final decision boundaries must be planned in advance 4051. The accepted machinery comes from group sequential design:
- Group sequential designs with prospectively specified interim looks and stopping criteria for efficacy or futility 4062.
- O'Brien-Fleming boundaries, which demand very persuasive early results before a trial may stop for efficacy 404243.
- Pocock boundaries, which set a flatter bar and therefore stop early more readily 404243.
- Lan-DeMets alpha-spending functions, which define how the Type I error budget is spent over the course of the trial while allowing flexibility in the number and timing of interim analyses 404243.
The prospective plan should state the number and timing of interim analyses and the stopping rules, justified by how persuasive early evidence must be, the probability of early stopping, and the expected and maximum sample size or event count 4041. These principles trace back to ICH E9, Statistical Principles for Clinical Trials 94, and are developed in the FDA's 2019 final guidance, Adaptive Design Clinical Trials for Drugs and Biologics 90. The FDA also cautions that early stopping is most defensible where there are compelling ethical reasons, for example a survival primary endpoint, and warns that a sponsor who believes the product may outperform expectations should build in a group sequential design from the start, because stopping early in a non-adaptive design is not scientifically valid 414362.
One design nuance the guidance stresses: stopping early can leave too little information for safety, secondary endpoints, and subgroup analyses, so there is sometimes a practical floor on how early an interim analysis should occur, particularly when a minimum sample size or follow-up duration is needed to characterize safety 4159. As an illustration of the efficiency at stake, the guidance notes that a single interim analysis at half the maximum sample size with an O'Brien-Fleming boundary can cut expected sample size by roughly 15% when the drug truly works 40.
Oncology: the deepest body of precedent
Oncology accounts for most of the precedent, driven by time-to-event primary endpoints (overall survival, progression-free survival) and the ethical pull to stop early when a benefit is clear. Documented approvals supported by a pre-specified interim analysis of the primary endpoint include:
| Drug (brand) | Indication | Trial | Primary endpoint | Interim design |
|---|---|---|---|---|
| Everolimus (Afinitor) | Advanced renal cell carcinoma | Pivotal phase 3 | Progression-free survival | Two planned interim PFS analyses; stopped after the second for outstanding efficacy before final events accrued 1 |
| Lapatinib (Tykerb) | HER2+ metastatic breast cancer (with capecitabine) | EGF100151 | Time to progression | Pre-specified interim TTP analysis at 133 events triggered a recommendation to terminate early 2 |
| Ibrutinib (Imbruvica) | Relapsed/refractory CLL/SLL | PCI-32765CLL3001 | Progression-free survival | Pre-specified interim PFS analysis at 171 events with an overwhelming-efficacy boundary; crossing it supported stopping 5 |
| Cabozantinib (Cometriq) | Medullary thyroid cancer | Phase 3 | Progression-free survival | Single interim efficacy analysis after ~50% of expected PFS events, with early termination possible for overwhelming efficacy 6 |
| Eribulin (Halaven) | Metastatic breast cancer | Study 305 | Overall survival | Pre-planned interim OS analysis after half the deaths (206) accrued; conducted at 207 deaths before follow-up completed 8 |
| Nivolumab (Opdivo) | Squamous NSCLC | CA209017 (CheckMate-017) | Overall survival | Pre-specified interim OS analysis at ≥196 deaths; DMC recommended unblinding after the boundary was crossed 12 |
| Abiraterone (Zytiga) | Chemo-naive metastatic CRPC | COU-AA-302 | Radiographic PFS / OS | Statistically significant rPFS at the interim analysis; pre-specified OS interim at 56% information did not cross its boundary 1617 |
| Dinutuximab (Unituxin) | High-risk neuroblastoma | DIV-NB-301 | Event-free survival | Randomization stopped at the seventh interim analysis of EFS, before all participants completed follow-up 2124 |
| Ado-trastuzumab emtansine (Kadcyla) | HER2+ metastatic breast cancer | EMILIA | Overall survival | Second interim OS analysis crossed the pre-specified efficacy boundary; the randomized portion was closed for crossover 113118 |
| Idelalisib (Zydelig) | Relapsed CLL / lymphoma | GS-US-312-0116 | Progression-free survival | First pre-specified interim analysis crossed the alpha boundary; the blinded phase stopped for efficacy 115123 |
The common structure across these cases is a prospectively documented statistical analysis plan naming the interim event count, the alpha allocated to that look, and a pre-agreed action (typically a data monitoring committee recommendation to stop or unblind) if the boundary is crossed. The endpoint on which approval rested was the interim result, not a later, fully mature dataset.
Cardiovascular and metabolic outcomes trials
The same group sequential framework now appears in the large outcomes trials that anchor cardiometabolic labels, though these programs show more variation in whether the trial actually stopped:
- Semaglutide (Wegovy) — SELECT. The trial pre-specified a single interim analysis of the primary endpoint, time to first adjudicated major adverse cardiovascular event (MACE), using a Lan-DeMets group sequential stopping boundary of one-sided p < 0.00650 30. At the interim, conducted after 817 events (about two-thirds, later described as 68%, of the eventual 1,225 adjudicated events), the primary endpoint met the pre-specified boundary 30125126. Notably, meeting the boundary did not force a stop: the DMC recommended the study continue to accrue more data, and the trial ran to completion 124125. This is an instructive precedent that crossing an interim boundary permits, but does not obligate, early termination.
- Dapagliflozin (Farxiga) — DECLARE-TIMI 58. Two formal interim analyses were planned, with the option for the DMC to recommend early stopping if superiority was achieved for MACE and all-cause mortality; in the reviewed materials the trial was not stopped early 27.
- Insulin degludec (Tresiba) and Ryzodeg 70/30 — DEVOTE. An interim analysis after 150 MACE was described, with the upper confidence bound ruling out the pre-specified risk margin, though the review does not state the trial stopped early for efficacy 3536.
- Empagliflozin (Jardiance) — EMPA-REG OUTCOME and tirzepatide (Mounjaro) CV meta-analysis both used planned interim looks with alpha spending, again without documented early stopping for efficacy 34383139.
The heart-failure literature supplies a textbook example the FDA itself cites: PARADIGM-HF planned three interim analyses after one-third, one-half, and two-thirds of events and stopped at the third interim look after crossing the pre-specified boundary for compelling superiority 49.
What the precedents tell a sponsor
Three practical lessons emerge for regulatory strategy:
- Prospective specification is the price of admission. Every accepted case had the interim analysis, its event count or information fraction, its boundary type, and the decision rule written into the protocol and SAP before data were seen 404151. An unplanned interim look does not qualify.
- The boundary sets the evidentiary bar, and O'Brien-Fleming-style boundaries dominate for early efficacy stops because they require the early result to be very persuasive, protecting against stopping on a lucky look 404243.
- Crossing a boundary enables, but does not compel, filing on the interim. SELECT shows a DMC can let a trial continue past a met boundary 124125, while EMILIA, Halaven, Opdivo, and the everolimus program show interim data serving directly as the efficacy basis for approval 1812113118. The FDA also expects enough accumulated safety, secondary-endpoint, and subgroup information to support a benefit-risk decision, which can impose a floor on how early the interim is placed 4159.
For a program contemplating this route, the governing references to build against are ICH E9 94 and the FDA's Adaptive Design Clinical Trials for Drugs and Biologics guidance 90, with the specific boundary, timing, and DMC charter aligned to the endpoint's ethical and safety profile.