Choosing a master protocol
Use basket and umbrella to describe the treatment-population structure; use platform to describe a trial that can evolve over time.
Fit
When it matters, and when it does not
It matters when
- You must organize several linked treatment questions under one protocol.
- You need to distinguish biomarker groups, diseases and changes to trial arms.
It matters less when
- For one fixed treatment question, a conventional randomized trial may be simpler. Read: Randomized controlled trial
Before you start
What to check in your study
- Treatment and population structureState which treatments cross which diseases or biomarker groups.
- Arm changesDecide whether arms can enter, stop or replace others over time.
- Control strategySpecify randomization, concurrent controls and which groups may share information.
- Each comparison has a credible control and eligible population.How to check: Check calendar time, eligibility, standard care and overlapping control recruitment.
- Information sharing has a justified model.How to check: Test performance when effects differ across diseases or biomarker groups.
- Adaptation rules preserve the intended error control.How to check: Prespecify decisions and simulate plausible trial scenarios before recruitment.
Plain language
How it works
A basket design studies a treatment across disease groups linked by a shared feature. An umbrella design studies different treatments within one disease, often assigned by biomarkers. A platform can add or drop arms under standing rules. A platform may also have basket or umbrella features.
Concurrent controls enroll while a treatment arm is open. Earlier controls are nonconcurrent; changes in care can bias the contrast. Shared controls make test results correlated. Define the family of claims and control false positives accordingly. A screening signal is not automatically confirmatory evidence. The single-design pages explain each trial structure.
Worked example
A worked example
- Setting
- A standing trial initially has four experimental arms.
- Question
- Does adding another arm change the protocol label?
- Approach
- The trial permits entry and stopping under prespecified rules. One arm stops and two arms enter.
- Result
- The active arm count changes from 4 to 5, a change of 1: 4 minus 1 plus 2 equals 5.
- What it means
- That evolution is a platform feature. Whether it is also basket or umbrella depends on its disease and treatment structure, not the arm count.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
Platform, basket and umbrella are overlapping descriptions, not guarantees of randomization or causal evidence. Shared controls, borrowing across groups and changing arms require prespecified validity and error-control rules.
- Treating the three labels as mutually exclusive.Instead: Describe both population structure and evolution over time.
- Assuming every master protocol randomizes.Instead: Read the allocation and control rules.
- Sharing historical controls without checking time trends.Instead: Justify comparability and distinguish concurrent recruitment.
- Calling a fixed shared-control test a platform engine.Instead: A Dunnett comparison addresses correlated tests against one control. It does not handle arm entry, stopping or changing care.
Inside Mallard
How Mallard uses this
Mallard includes this comparison in Learn search and the Atlas. Plans do not link to this comparison automatically.
concept.choosing_master_protocol_comparison, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- Master Protocols to Study Multiple Therapies, Multiple Diseases, or Both (opens in a new tab) · N Engl J Med · doi 10.1056/NEJMra1510062
- Vemurafenib in Multiple Nonmelanoma Cancers with BRAF V600 Mutations (opens in a new tab) · N Engl J Med · doi 10.1056/NEJMoa1502309
- Adaptive Randomization of Veliparib-Carboplatin Treatment in Breast Cancer (opens in a new tab) · N Engl J Med · doi 10.1056/NEJMoa1513749