Multi-Arm and Master-Protocol Trials

A multi-arm parallel-group randomized trial allocates participants concurrently among three or more groups. A master protocol coordinates multiple substudies within an overall trial structure. These ideas can overlap, but they are not synonyms: a fixed three-arm trial need not use a master protocol, and a master protocol can contain substudies with different populations, objectives, or opening dates.[1][3][4]

The efficiency of a shared structure does not turn many questions into one answer. Every treatment comparison still needs a defined population, a clearly identified control set with eligibility and concurrency stated, an endpoint, an analysis population, a statistical method, an effect estimate, and uncertainty.

A multi-arm parallel trial is not a factorial trial

The 2019 CONSORT extension defines the multi-arm parallel design as concurrent randomization to three or more treatment groups.[1] Imagine one control and three experimental groups. The primary questions might compare each experimental group with the common control. Participants are assigned to one group.

In a 2×2 factorial trial, by contrast, participants are randomized with respect to two factors, and the four cells represent combinations of those factor levels. Main analyses may combine cells to estimate each factor’s effect. Four visible groups can therefore encode different randomized questions.

Feature	Multi-arm parallel trial	Two-factor factorial trial
Basic allocation	One of three or more parallel groups	A combination of levels across two randomized factors
Typical main questions	Several treatment-versus-control or other arm comparisons	Main effect of each factor, interaction, or selected cell comparisons
Shared control	One control group may contribute to several arm comparisons	Control levels are embedded across factor combinations
Key reading task	Map the comparison family and control contribution	Map cells, factor comparisons, and interaction assumptions

Calling every four-group study “factorial” can produce the wrong denominator and effect interpretation. Reconstruct the randomization and prespecified hypotheses.

Shared controls can improve efficiency

When several experimental groups are eligible to use the same control, a common control group can avoid duplicating control enrollment. More participants may be allocated to investigational groups without losing a randomized comparator for each question. FDA’s June 2026 guidance recap describes this as one potential efficiency of umbrella and platform trials, alongside shared infrastructure and oversight.[4]

Efficiency is conditional. For experimental arm A, the relevant control contribution should meet the eligibility rules for comparison A and be enrolled during an appropriate concurrent period. For arm B, those details may differ. A single displayed control total can therefore conceal comparison-specific control subsets.

The analysis should report:

which control participants contributed to each primary comparison;
the arm and control entry and exit dates;
randomization ratios and any changes over time;
comparison-specific eligibility and endpoint rules;
the number analyzed and safety denominator for every group; and
the effect estimate and interval for each comparison.

A participant’s appearance in a common control arm does not make that person relevant to every substudy automatically.

Umbrella, platform, and basket structures answer different questions

FDA’s 2026 recap distinguishes three examples of trials that can use master protocols.[4]

Umbrella trial: evaluates multiple products concurrently for a single disease.
Platform trial: evaluates multiple products for one or more diseases on an ongoing basis, with products able to enter or leave over time.
Basket trial: evaluates a product across multiple diseases, conditions, or disease subtypes.

These are structural descriptions, not quality labels. A platform is not simply any trial with three arms or any trial using interim analysis. An umbrella trial need not be perpetual. A basket trial does not assume that the treatment effect is identical across every included disease or subtype.

The protocol should describe the overall structure and each substudy’s objective, population, intervention, comparator, endpoint, and analysis. Results from one substudy cannot be transferred to the others merely because they share governance or infrastructure.

Concurrent and nonconcurrent controls are different evidence

A concurrent control is eligible and randomized during the period when the experimental arm is recruiting. A nonconcurrent platform control comes from an earlier period when that arm was not available. It remains an internally randomized control; FDA treats external-control data as a separate category outside this draft’s scope.[3] Calendar time can change prognosis, diagnostic criteria, background care, site participation, follow-up, outcome measurement, data completeness, and participant selection.

Randomization protects a comparison among options available at that randomization point. It does not automatically remove bias between an arm opened later and control participants enrolled months or years earlier. Modeling calendar time may be useful, but it depends on measured trends and assumptions. It cannot guarantee removal of unmeasured changes.

Readers should ask for arm-specific timelines, recruitment volumes, baseline patterns, outcome rates over time, contemporaneous eligibility, site overlap, and prespecified methods. A nonconcurrent analysis should be identified as such rather than blended invisibly into a randomized concurrent-control estimate.

The comparison family determines multiplicity

Three experimental arms versus one control create several potential claims. More may arise from secondary outcomes, subgroups, interim looks, dose levels, or newly added arms. The report should identify the primary comparison family and explain whether and how false-positive risk was controlled.

The 2019 multi-arm CONSORT extension asks authors to state objectives and hypotheses referring to all groups, identify primary comparisons, report planned and unplanned comparisons transparently, and describe the rationale and method when multiplicity adjustment is used.[1] The extension improves reporting; it does not prescribe one universal correction.

Multiplicity choices depend on which conclusions are intended jointly, the relationship among tests, and the decision framework. A platform-wide statement that “one arm succeeded” and an arm-specific decision may require different error considerations. Readers need the prespecified strategy, not just an adjusted p value after the fact.

Adding or dropping arms changes the information map

In an ongoing platform, arms may enter after launch or leave for prespecified benefit, harm, futility, operational, or strategic reasons. Each change affects who could have been randomized to what and when. Reports should disclose the decision rule, information available to decision makers, timing, operational consequences, and any resulting analysis changes.

Stopping one substudy does not automatically stop the master protocol. Nor does a successful arm validate the remaining arms. An early-stopped comparison can have different follow-up, information size, and estimation properties from a comparison that continues.

Adaptive features form a separate layer. FDA’s December 2019 final Level 1 guidance discusses prospective adaptations in drug and biologic trials and the information needed for review.[5] A master protocol need not be adaptive, and an adaptive trial need not be a platform. When both are present, the report should explain each adaptation and how it interacts with the shared control, multiplicity, blinding, and trial integrity.

Blinding and information access matter across substudies

Interim results from one arm can influence enrollment, care, site behavior, or expectations elsewhere in a shared platform. The protocol should identify who sees comparative data, what summaries are released, and how the integrity of ongoing comparisons is protected.

Common committees can improve consistency, but common oversight does not eliminate operational bias. If control allocation changes, sites learn that an arm is under review, or public results alter recruitment, later participants may differ from earlier ones. Calendar-time diagnostics and transparent reporting become especially important.

Safety reporting also needs arm-specific denominators and exposure periods. Pooling all participants under one master-protocol label can hide which intervention, substudy, or follow-up window produced an observation. Shared infrastructure should increase clarity, not erase attribution.

FDA’s June 2026 document is a revised draft

FDA’s Master Protocols for Drug and Biological Product Development, issued in June 2026, is marked Draft, Not for implementation, and as containing non-binding recommendations.[3] It addresses design and analysis of master-protocol trials and documentation for regulatory review. FDA’s recap states that the revised draft replaces the December 2023 draft and adds recommendations on basket trials.[4]

“Revised” does not mean final. The document should not be presented as binding regulation or a completed final guidance. The FDA snapshot and recap are communication aids; the draft guidance itself controls its stated scope and status.

The SPIRIT–CONSORT overview separately lists the 2019 multi-arm extension under extensions to CONSORT 2010.[2] Reporting against that extension can improve transparency but does not certify that a shared-control or master-protocol analysis is unbiased.

A multi-arm and master-protocol reading checklist

Draw the arms, substudies, and primary comparisons.
Distinguish fixed multi-arm, umbrella, platform, and basket features.
Record arm-specific eligibility, endpoints, and analysis populations.
Mark each arm’s entry and exit dates and every randomization-ratio change.
Identify the exact control participants contributing to each comparison.
Separate concurrent from nonconcurrent control evidence.
Review multiplicity, adaptation, and stopping strategies.
Identify who saw interim information and how ongoing integrity was protected.
Extract comparison-specific effect estimates, intervals, follow-up, and safety denominators.
Keep each conclusion within its substudy and tested intervention.

Important limitations

A common protocol does not make every arm comparable with every control participant.
Shared controls improve efficiency only when comparison-specific eligibility and timing are defensible.
Nonconcurrent controls can differ because of calendar-time changes.
“Platform” is not a generic term for all multi-arm or adaptive trials.
One positive arm does not validate other arms or the platform as a whole.
Stopping one substudy does not automatically stop the master protocol.
Reporting-guideline adherence does not establish unbiased analysis.
Master-protocol evidence cannot authenticate a separately sourced material.

Frequently asked questions

What makes a randomized trial multi-arm?

In the 2019 CONSORT extension, it is a parallel-group trial that concurrently randomizes participants among three or more treatment groups.[1] The primary comparisons must still be specified.

Is every multi-arm trial a platform trial?

No. A fixed trial with three groups is multi-arm but may have no ongoing entry or exit of products. A platform has an ongoing structure in which products can enter or leave over time.[4]

Why use a shared control group?

It can reduce duplicate control enrollment and share infrastructure. Each comparison still requires an eligible, relevant control set and its own effect estimate, uncertainty, and safety denominator.

Can a later arm use all earlier control participants?

Not automatically. Earlier participants are nonconcurrent with that arm and may differ because of changes in population, care, sites, conduct, or measurement. Any use requires explicit justification and analysis.

Are umbrella and basket trials the same?

No. FDA’s recap describes umbrella trials as evaluating multiple products for one disease and basket trials as evaluating a product across multiple diseases or subtypes.[4]

Does dropping one arm end the whole master protocol?

No. An arm or substudy can stop while others continue, depending on the protocol and decision rules. The report should explain what stopped, why, and what remained active.

Is FDA’s June 2026 master-protocol guidance final?

No. It is a revised draft, marked not for implementation and non-binding. It replaces the December 2023 draft but is not final guidance.[3][4]

Are you over 21?

Multi-Arm Trials, Shared Controls, and Master Protocols