Umbrella, Basket, and Platform Trials: What the FDA's Master-Protocol Guidance Actually Changes

The FDA's revised draft guidance on master protocols, published in the Federal Register on June 24, 2026, does not invent umbrella, basket, and platform trials; it tidies up how one governing protocol can ask many questions at the same time. The revision adds a dedicated section on basket trials in response to public comments and tightens what the agency expects around shared controls, randomization, choice of comparator, and multiplicity. It revises and replaces the December 22, 2023 draft, and the comment window stays open through August 24, 2026.

Key points#

• A master protocol is one framework that runs several substudies under shared infrastructure, so investigators do not have to build a separate trial for every drug-and-disease pairing.
• The three families differ by what is held fixed: umbrella (one disease, many drugs), basket (one drug, many diseases), and platform (arms enter and leave over time).
• The June 2026 revision is a refinement, not a reinvention: more detail on basket trials plus minor clarifications on randomization, control choice, and informed consent.
• Two design choices decide how much a result can be trusted: how the shared control group is used, and how the many comparisons are counted.

Three ways to share one protocol#

The efficiency argument is straightforward. Screening, data management, and often a single control group are expensive to stand up, so a master protocol reuses that machinery across related questions instead of rebuilding it each time. Where the three designs diverge is in what stays fixed and what varies. A helpful anchor is the 2019 systematic review by Park and colleagues in Trials, which cataloged how these studies are actually built and reported rather than how they are described in press coverage.

The cleanest way to keep the labels straight is to ask yourself a single question: is the disease fixed, or is the drug fixed?

Umbrella: one disease, many drugs#

An umbrella trial fixes the disease and varies the treatment. Patients with a single condition are sorted into arms, usually by a biomarker or other characteristic, and each subgroup receives a therapy matched to its features. Park and colleagues framed this as stratifying one disease into molecularly or clinically defined subgroups and then testing a targeted intervention inside each. Oncology has driven most of the examples, where one tumor type is partitioned by mutation and every subgroup gets its own matched agent under the same umbrella.

Basket: one drug, many diseases#

A basket trial flips the arrangement. Here the treatment is fixed and the disease varies: a single agent is studied across several conditions or subtypes that share a common predictive marker. The recurring motivation is a drug aimed at a molecular alteration that turns up in otherwise unrelated cancers. This is the design the revision spends the most new attention on, precisely because sponsors asked for it. The added recommendations cover how to judge a drug's effect when the same agent is tested across dissimilar populations, ranging from a combined analysis that pools across substudies to separate analyses within each individual basket.

Platform: arms that come and go#

A platform trial relaxes the assumption that every arm is chosen at the start. Its adaptive structure lets treatment arms enter or leave as evidence accumulates, overlapping with multi-arm, multi-stage designs. Park and colleagues described these as studies that can drop intervention arms and introduce new ones mid-course. I-SPY 2 in breast cancer, STAMPEDE in prostate cancer, and RECOVERY during the pandemic are the examples most often cited for this perpetual, evolving format.

What the June 2026 revision actually adds#

The headline is narrower than some coverage suggests. The FDA states that the differences from the original draft amount to more detailed recommendations on basket trials plus minor clarifications on topics including randomization, choice of control, and informed consent. The core message carries over intact: a master protocol can improve efficiency by sharing a control arm across several drugs and by borrowing information across related conditions. That is a sharpening of expectations, not a new category of trial. Two appraisal tradeoffs deserve emphasis, because they are exactly where efficiency and rigor pull in opposite directions.

Where rigor can slip#

The shared control group#

A shared control arm is the main efficiency lever, and also the main place a comparison can go wrong. If several experimental arms are each measured against one common control group, fewer participants overall are assigned to placebo or standard care, and the study runs faster. The tradeoff is that every comparison now rests on a control group that may have been recruited at a different time, or under slightly different conditions, than a given experimental arm. Concurrent randomization, in which each experimental arm is compared only against controls enrolled during the same window, guards against drift in patient mix and standard of care over a long-running platform. When a comparison leans on non-concurrent controls instead, the interpretive burden climbs, and that is a fair question for you to put to any master-protocol result.

Counting the comparisons#

Testing many arms or many populations under one protocol multiplies the statistical comparisons, and each added comparison raises the chance of a false positive somewhere in the family of tests. The guidance treats multiplicity as something to plan for at the design stage rather than patch afterward. If you are appraising a published master-protocol trial, the practical questions are whether the analysis was pre-specified, whether Type I error was controlled across the substudies or only within each one, and whether a pooled claim across baskets rests on a genuinely consistent effect or papers over real differences between diseases. A drug can look effective in a combined basket analysis while actually working in only one or two of the constituent conditions, which is why the choice between a pooled and a substudy-level analysis is never a mere technicality.

Reading the guidance as a clinician#

Draft guidance describes the FDA's current thinking. It is not a regulation, it does not bind sponsors or the agency, and the open comment window means the text may still shift before it is finalized. Read as a neutral map, the document mostly codifies habits that careful trialists already keep: name the design honestly, justify the control structure, pre-specify the analysis, and account for every comparison being made. The efficiency of master protocols is real, and so is the extra interpretive care their results ask of you. Both can hold at once, and the revision is largely an effort to keep the two in balance.