Basket, Umbrella, and Platform Trials: What Master Protocols Actually Do

When a cancer drug's label reads "for any solid tumor carrying an NTRK gene fusion" instead of naming a specific organ, that wording traces back to how the drug was studied. It was almost certainly tested inside a master protocol. A master protocol is a single overarching study framework that runs several substudies at once, so that investigators can evaluate one or more therapies across one or more diseases under shared infrastructure and oversight. Understanding the three main forms it takes tells you a great deal about how much the resulting evidence can bear.

Key points#

• A master protocol builds one study skeleton (sites, screening, control group, safety monitoring) and attaches new questions to it as substudies.
• A basket trial tests one drug across many diseases that share a molecular feature; an umbrella trial tests many drugs within one disease, sorted by biomarker; a platform trial stays open over time, adding and removing arms against a common control.
• Tissue-agnostic approvals, where a drug is cleared by a genetic marker rather than the organ, grew out of the basket design and often rest on single-arm response data.
• The design a result came from is a clue to how much weight it can carry.

The problem master protocols were built to solve#

Run a separate trial for every drug and every disease and you rebuild the same machinery each time: a new protocol, a new network of sites, a fresh control group, another regulatory submission. That repetition is slow and expensive, and it is especially wasteful when the diseases in question are rare molecular subsets with few eligible patients to go around.

A master protocol assembles that machinery once. The shared skeleton (common screening, a common control arm, unified safety monitoring) is built at the outset, and individual research questions plug into it as substudies. The U.S. Food and Drug Administration frames these designs as a way to expedite development for exactly this reason. The efficiency is structural rather than a statistical shortcut. That same shared structure also sets the terms of caution: because many substudies lean on one framework, a weak control arm or a loose analysis plan can compromise several comparisons at once rather than just one.

The FDA set out this taxonomy in its final oncology guidance, "Master Protocols: Efficient Clinical Trial Design Strategies To Expedite Development of Oncology Drugs and Biologics," issued in early 2022. In December 2023 the agency published a broader draft guidance, "Master Protocols for Drug and Biological Product Development," extending the approach beyond cancer, with a public comment period that ran into 2024. A draft is proposed thinking, not settled policy, so the 2023 document is best read as the direction of the agency's reasoning rather than a fixed rule.

Three shapes, three questions#

The three common designs share the skeleton but ask different questions. A useful way to keep them straight is to notice what stays fixed and what varies.

Basket trials: fix the target, vary the disease#

A basket trial starts from a molecular target instead of an organ. If a particular mutation, gene fusion, or protein turns up in lung, colon, thyroid, and pancreatic tumors alike, a basket trial enrolls all of them into parallel cohorts, the "baskets," and gives each the same targeted drug. The scientific wager is that the target matters more than where the tumor sits.

This is the design that produced the first tissue-agnostic approvals. In 2017 the FDA approved pembrolizumab for solid tumors that are microsatellite instability-high or mismatch-repair deficient, the first time it cleared a cancer therapy defined by a biomarker rather than a body part. In late 2018 larotrectinib followed, granted accelerated approval for solid tumors carrying an NTRK gene fusion, on the strength of multi-cohort single-arm studies that enrolled patients across roughly a dozen tumor types. Responses concentrated in the biomarker-positive population, which is what justified a label untethered from any single cancer site.

The hazard specific to baskets is that the same target can behave differently in different tissues. A drug that shrinks tumors in one organ may barely move another despite the identical mutation. This is why the FDA's guidance emphasizes adequate cohort sizes and pre-specified rules, so a genuine signal in one basket is neither diluted by pooling nor mistaken for a broad effect.

Umbrella trials: fix the disease, vary the drug#

An umbrella trial flips the logic. It takes a single disease, non-small cell lung cancer is a classic example, and divides patients by molecular subtype, then matches each subtype to the drug aimed at it. One umbrella can hold several experimental arms under a shared eligibility screen that routes each patient toward the arm that fits their biomarker. For a disease that splits into many rare molecular subsets, this replaces a dozen small, slow trials with one structure that screens everyone and directs them. The 2023 draft guidance pays particular attention to umbrella and platform designs, because these are where questions about shared controls and comparisons between experimental arms get most delicate.

Platform trials: keep the structure running#

A platform trial is an umbrella that never formally closes. Arms enter as new candidates mature and leave once a decision is reached, all measured against a common control group that persists across the life of the study. The design proved its worth during the COVID-19 pandemic, when platform trials assessed one treatment after another in sequence without tearing down and rebuilding each time. The cost is governance. A structure that stays open for years has to manage shifting standards of care, evolving comparators, and the statistical work of comparing arms that entered at different moments, which is why the guidance dwells on control-arm choice and analysis planning.

From basket to a tissue-agnostic label#

The line from design to approval usually runs through the basket. When a biomarker-defined population responds consistently across tumor types, the regulatory case for approving by biomarker rather than by organ starts to hold together. The supporting evidence is often a single-arm response rate rather than a randomized survival comparison, which is why several of these clearances arrived through the accelerated approval pathway, with confirmatory studies expected to follow.

That is a real trade-off, not a free win. A biomarker-defined label reaches patients with rare cancers who would otherwise have no matched therapy, but it rests on smaller, non-randomized datasets that reward careful reading. For a generalist tracking oncology from a distance, the practical takeaway is simple: knowing whether an approval came from a basket cohort, an umbrella arm, or a randomized platform arm tells you most of what you need to know about how firm the ground underneath it is.