Master protocols have become a central fixture in modern drug and biologic development, yet the regulatory expectations governing basket, umbrella, and platform designs span multiple FDA guidance documents issued across different therapeutic contexts and years. For regulatory and clinical teams constructing or reviewing these protocols, understanding precisely how FDA defines each design type—and what infrastructure, statistical, and operational standards apply—is essential to avoiding late-stage review delays or partial clinical holds.
The analysis below synthesizes FDA's definitions and design expectations for each master-protocol archetype, drawing on the oncology-specific guidance, the 2021 COVID-19 master-protocol guidance, the animal-drug master-protocol guidance, and the broad draft guidance issued in June 2026. It addresses how the agency distinguishes the three designs from one another, what shared-infrastructure requirements apply across all types, and what design-specific expectations regulators apply to each.
Want Rhizome's help on your own question? Try it for free.
Basket, umbrella, and platform trials: how FDA guidance defines and evaluates master-protocol designs
Master protocols have moved from a niche oncology tool to a mainstream development strategy, and FDA's written expectations have kept pace. The agency's thinking is spread across several documents: the oncology master-protocol guidance, the 2021 final COVID-19 master-protocol guidance, the animal-drug master-protocol guidance, and, most recently, a broad draft, Master Protocols for Drug and Biological Product Development (draft, issued June 24, 2026) 104, which generalizes the concept beyond oncology 105. This overview pulls the definitions and the design expectations that recur across those documents, organized by master-protocol type.
What FDA means by "master protocol"
FDA defines a master protocol as a single protocol with multiple substudies designed to evaluate multiple hypotheses about one or more investigational products within an overall study structure 13. The defining feature is shared infrastructure: one trial design and protocol that can evaluate multiple drugs, multiple diseases, or both 42. Basket, umbrella, and platform designs are the three archetypes that sit under that umbrella term, differentiated by whether the multiplicity lies in the disease, in the drug, or in the time dimension.
Across guidance documents, the three types are defined as follows:
| Design | Core definition | What varies |
|---|---|---|
| Basket | Evaluates a single investigational drug (or combination) across multiple diseases, conditions, or disease subtypes 3031. The animal-drug guidance frames it as one drug in different populations defined by tumor type, stage, histology, prior therapy, or biomarker 13. | Multiple diseases/populations, one therapy |
| Umbrella | Studies multiple products in parallel for a single disease or condition under one master protocol 323113. | Multiple therapies, one disease |
| Platform | Evaluates multiple drugs for one or more diseases in an ongoing manner, with drugs entering or leaving the platform based on a prespecified decision algorithm 3031. | Multiple therapies against a common control, evolving over time |
The distinction matters because the statistical and operational risks differ by type: basket trials raise questions about pooling across heterogeneous populations, umbrella and platform trials raise questions about shared controls and multiplicity, and platform trials add the problem of comparisons drifting across calendar time.
Basket trials
FDA treats basket trials as a collection of substudies, where each substudy carries its own objectives, design, conduct, and analysis for one drug in one disease, condition, or subtype 58. The agency's expectations concentrate on how each substudy is specified and how results are interpreted across them.
- Substudy architecture. Basket-trial substudies are usually designed as single-arm, activity-estimating trials with overall response rate as the primary endpoint 91. Each substudy should state specific objectives, the scientific rationale for including each population, and a detailed statistical analysis plan with sample-size justification and stopping rules for futility or efficacy 91.
- A clear primary objective for the whole trial. Sponsors should specify and justify a clear primary objective for the overarching trial and explain how the design and analysis achieve it 81.
- Cohort expansion into a registrational path. If a substudy shows a strong response signal, FDA notes the substudy may be expanded to generate data that could potentially support a marketing application 91.
- Interpretation: separate analysis is the default. FDA describes two interpretation strategies: separate analyses confined to each substudy population, which is often most appropriate to minimize bias; or, where scientifically justified, leveraging information across related substudies qualitatively or quantitatively, including Bayesian borrowing 81. If sponsors pool or borrow across substudies, they must justify the degree of relevance, including similarities in disease pathophysiology, how the drug is expected to modulate that pathophysiology (including pharmacokinetics), and endpoint/estimand attributes 81.
- Guarding against overinterpretation. FDA warns that multiple study groups can lead to overinterpretation, including false apparent efficacy differences arising from multiple ad hoc between-arm comparisons 92.
- Submission structure. For basket trials it may be appropriate to submit separate INDs for new substudies; sponsors should discuss the planned approach with the review division, weighing the populations, design, and whether analyses will leverage information across substudies 62.
Umbrella trials
Because an umbrella trial evaluates several therapies within one disease 31, FDA's expectations center on preserving valid comparisons among those therapies.
- Randomization where feasible. FDA says comparisons among arms may be facilitated by randomization between the study arms, where feasible 32, and that randomization ratios may be modified in adaptive designs, though this must be handled carefully 19.
- Shared control groups. Umbrella trials may share a control group, which improves efficiency and simplifies study management 32; each drug is generally compared against a control arm 19.
- Eligibility-matched control. To preserve valid comparisons, the control group for a given drug should generally include only participants who were concurrently eligible and could have been randomized to that drug 56.
FDA's retrieved guidance is lighter on biomarker-stratification specifics for umbrella designs; the documents reference disease subtypes and eligibility criteria rather than prescribing a biomarker-stratification scheme 30565834.
Platform trials
Platform trials are the most operationally complex master protocol: multiple therapies evaluated concurrently against a common control, running continuously as therapies enter and leave over time under prespecified rules 3487.
- Common control. Platform (and umbrella) designs study multiple interventions concurrently against a common control group, with the comparison typically each individual drug versus that control arm 3419.
- Adding arms. The design allows different drugs to enter the study at different times in a continuous manner 87.
- Dropping arms. Drugs may leave the platform based on prespecified decision rules; more broadly, master protocols may use interim analyses to stop enrollment in a substudy for efficacy or futility 8719.
- Concurrent vs. nonconcurrent controls. This is the design issue FDA emphasizes most for platform trials. Because a platform control arm can contain both participants enrolled concurrently with a given drug and participants randomized earlier (nonconcurrent controls who could not have been randomized to that drug) 56, the primary comparison for a given drug should generally include only concurrently eligible controls and exclude nonconcurrent controls, as well as concurrently enrolled participants who could not have been randomized to that drug 5627. FDA's rationale is that concurrently eligible controls preserve randomization and avoid bias from temporal changes in participant characteristics, trial conduct, or standard of care 27. Nonconcurrent controls may be justified only in rare circumstances, with scientific justification addressing temporal shifts, the amount of nonconcurrent data, separation in calendar time, and statistical methods to mitigate bias; this should be discussed with FDA early and agreed before the trial starts 27.
Statistical design expectations across master-protocol types
FDA's statistical expectations apply across all three designs and are where many master protocols succeed or fail on review.
- Type I error and multiplicity. Master protocols should control type I error probability across the relevant multiple comparisons and control familywise type I error for each individual drug in its intended population across other multiplicity sources such as multiple endpoints 18. Type I error control is especially important for closely related comparisons (multiple doses, administrations, or formulations of the same drug) 18. For shared-control designs, the expected total number of type I errors may match separate trials, but the distribution of those errors differs, so sponsors should consider the probability of at least one error, the probability of multiple errors, power, and other operating characteristics 18. In the COVID-19 umbrella/platform setting, FDA did not require multiplicity adjustment for multiple drug-versus-comparator comparisons to control overall familywise error, but still expected sponsors to weigh the possibility of multiple correlated erroneous findings rather than relying on a single p-value 22.
- Shared and common control arms. Master protocols may use one common control arm or multiple control groups 13, and sponsors may consider greater-than-equal allocation to control to reduce correlated erroneous findings and potentially improve power 1822. When some participants are eligible only for some arms, the analysis for a given arm should compare against only the eligible control participants 22.
- Bayesian methods. Bayesian methods may be used in adaptive and master-protocol settings, but the rationale should be clear and the conclusions sufficiently robust 12. FDA notes Bayesian methods can inform adaptations while the primary analysis still uses a frequentist hypothesis test to control type I error 12, and sponsors should justify that the prior, decision criteria, and adaptive elements achieve targeted operating characteristics (power, expected sample size, reliability of adaptation decisions) while maintaining type I error control 12.
- Adaptive features and interim analyses. Master protocols may include interim analyses to stop enrollment for efficacy or futility, modify sample size, or modify the randomization ratio 19. FDA's general adaptive-design principles apply: adaptations should be prespecified, supported by a detailed protocol and statistical analysis plan, and evaluated through simulation of operating characteristics including type I error, power, expected sample size, and bias 1917. Interim analyses can also create confidentiality and interpretability issues when information from one substudy reveals information about another 1922.
Safety monitoring, oversight, and IND logistics
FDA's operational expectations reflect the fact that one master protocol may touch many drugs, sponsors, and sites at once.
- Centralized safety reporting. Safety reports from all clinical investigators flow to the master-protocol sponsor, who must report unexpected serious adverse events with a reasonable possibility of causation to FDA and all participating investigators, supported by a process for rapid communication of serious safety issues and rapid protocol amendment when needed 6061.
- Independent oversight. FDA recommends a central IRB for the master protocol and an independent, external data monitoring committee (DMC) or comparable body to oversee accumulating safety and efficacy data 655768. The DMC charter should permit prespecified and ad hoc efficacy and futility reviews and recommendations for protocol modifications, including sample-size adjustment or stopping or modifying a substudy for futility or overwhelming efficacy 68.
- Protocol amendments. Amendment cover letters should be labeled "Protocol Amendment-MASTER PROTOCOL"; substantive amendments affecting safety, quality, or scope should be submitted at least 30 days before changes begin, with the regulatory project manager notified at least 48 hours before submission 6061. New investigational drugs or new substudies are added as amendments to the master-protocol IND 6364.
- IND structure. Sponsors should request a pre-IND meeting early; master-protocol trials should generally be conducted under a single master-protocol IND, using Study Tagging Files and proper eCTD organization, and cross-referencing appropriately where separate substudy INDs are used 62636466.
Early engagement and the path to substantial evidence
FDA repeatedly stresses that master protocols are complex enough to warrant early and often iterative interaction. Sponsors should request a pre-IND meeting to discuss the protocol and submission logistics, and FDA notes multiple meetings may be needed; early engagement is particularly important when contemplating features such as nonconcurrent controls 627627. For rare-disease programs, innovative designs should be discussed in advance, ideally at the pre-IND meeting 78.
On whether master-protocol data can support approval, FDA's position is conditional: master protocols may generate proof-of-concept, dose-ranging, effectiveness, and safety data, but whether a master protocol is adequate to contribute to a demonstration of substantial evidence of effectiveness depends on the design, conduct, and persuasiveness of the results 6729. Development programs often combine master protocols with stand-alone trials because different data types are needed, and the master protocol typically forms part of the overall safety database and benefit-risk assessment rather than the entire evidentiary basis 672974.