Two Different Questions About Trial Structure
Platform and umbrella trials are often discussed together because both can evaluate several treatments under a coordinated protocol. Their defining features, however, describe different aspects of a trial. A platform structure emphasizes a continuing trial framework in which treatment arms may change over time. An umbrella structure emphasizes how participants with one disease are divided into groups, often by a disease characteristic, and considered for different treatments within those groups.
These labels describe structure, not a guarantee of a particular statistical method or a claim that every participant receives the same kind of comparison. To classify a design, start with its protocol: who can enter, how participants are grouped, which treatments are available to each group, and whether the set of treatment arms changes during the trial. The introductory material on the anatomy of a platform trial provides the basic platform concept; here the emphasis is on comparing that concept with the umbrella structure.
The distinction is not simply “many arms” versus “few arms.” A trial with several treatments is not automatically a platform trial: it may have a fixed set of arms and no continuing framework that accommodates changes. Nor is every trial that records a biomarker an umbrella trial. The participant groups need to be part of the design structure in a way that affects the treatment options or comparisons.
Compare the Participant Structure and the Arm Timeline
Two questions help keep the concepts separate. First, how are participants grouped? In an umbrella design, participants with the same overall disease are organized into distinct groups, such as groups defined by a tumour characteristic. Those groups may have different treatment pathways. Second, how does the set of interventions behave over time? In a platform design, the trial framework can continue while arms enter or leave.
These are separate dimensions. A trial may divide participants into disease subgroups but keep every treatment arm fixed throughout the study. That is an umbrella feature without necessarily being a platform. Conversely, a platform can evaluate several treatments in a broad eligible population without making subtype-based treatment allocation its defining feature. A design can also have both features: it can use subgroup-specific treatment groups and allow its set of arms to change over time.
| Structural question | Platform feature | Umbrella feature |
|---|---|---|
| What chiefly distinguishes the structure? | A continuing framework evaluates multiple interventions, with the set of arms able to change over time. | Participants with one disease are organized into distinct groups, and treatment options can differ by group. |
| What should be checked in the protocol? | Whether the framework continues as interventions are introduced or discontinued. | Whether participant grouping is built into eligibility or treatment assignment. |
| What is not enough on its own? | A large number of treatment arms. | Collecting or reporting a biomarker that does not determine participant grouping or treatment options. |
| Can the feature occur alongside the other? | Yes. A platform can also organize participants into disease subgroups. | Yes. An umbrella design can also have a changing set of arms over time. |
For a descriptive classification, assume that the protocol and its amendments accurately state the eligible population, participant groups, and available treatments over time. The classification follows those design features; it does not follow from the trial’s title, the number of publications, or the fact that several investigators or sites are involved.
Worked Example: A Changing Set of Treatments in One Broad Population
Consider a hypothetical trial for adults with a specified advanced cancer. Participants are eligible regardless of tumour-marker group, and the protocol evaluates regimens A and B. One year later, regimen C is added under the continuing protocol. Later, recruitment to A ends while recruitment to B and C continues. In a simplified enrollment summary, 240 participants have entered: 80 in A, 80 in B, and 80 in C.
The structure has a clear platform feature: the framework continues while the set of available arms changes. The participant population is broad with respect to the marker in this example, so marker-based grouping is not what defines the structure. On these facts, calling it an umbrella trial would require additional evidence that participants are organized into disease subgroups with treatment options linked to those groups.
The arithmetic describes the enrollment distribution, not the definition: \(80+80+80=240\). Each arm accounts for \(80/240=0.333\), or about 33.3% of the participants in this simplified summary. Equal enrollment across arms does not create the platform feature; the changing arm set within a continuing framework does.
Worked Example: Treatments Organized Around Disease Groups
Suppose a hypothetical trial enrolls 180 people with one cancer type. Before treatment assignment, the protocol places each participant into one of three groups using a specified disease characteristic. The protocol provides a distinct treatment approach for each group, and all three approaches are in place for the period described. There are 60 participants in each group.
This has an umbrella feature: participants share the overall disease, but the protocol organizes them into groups that have different treatment options. The arm set being fixed in this example does not, by itself, establish a platform feature. More detail about the subgroup-to-treatment mapping is needed to describe an umbrella design fully, but it is not needed to identify the structural distinction here.
The group proportions are \(60/180=0.333\), or about 33.3% each, and \(60+60+60=180\). Those equal counts make the example easy to summarize but are not required for an umbrella structure. Unequal group sizes would not change the classification if the subgroup-based organization remained part of the design.
Worked Example: A Trial With Both Features
A hypothetical trial evaluates treatments for one disease through two protocol-defined participant groups. At the start, group 1 has treatment A and group 2 has treatment B. During the trial, treatment C is added for group 2, while the overall protocol and recruitment to group 1 continue. At a given snapshot, the trial has enrolled 120 participants: 50 in group 1 and 70 in group 2.
The participant structure gives the trial an umbrella feature: the groups are part of the design, and their treatment options differ. The changing set of treatments within a continuing framework gives it a platform feature as well. The two labels are therefore compatible, not mutually exclusive.
As a check on the snapshot counts, \(50+70=120\). Group 1 represents \(50/120=0.417\), or 41.7%, and group 2 represents \(70/120=0.583\), or 58.3%. These percentages describe who enrolled in the snapshot; they do not establish either structural feature. The protocol’s grouping rules and treatment timeline do that.
A Practical Classification Procedure
The following sequence makes the classification explicit and helps prevent a label from resting on a single superficial feature.
State the disease or clinical population shared by participants. Do not infer subgroup structure merely because the protocol measures participant characteristics.
Ask whether protocol-defined disease groups affect eligibility or the treatment options considered. If so, this is evidence of an umbrella feature.
Check whether the trial framework continues while interventions can be added or discontinued. If it does, this is evidence of a platform feature.
Describe the trial as platform, umbrella, both, or neither according to the evidence. If the protocol does not provide enough detail, state what remains unclear.
This procedure follows from the definitions: the umbrella label depends on subgroup organization, whereas the platform label depends on the continuing framework and its capacity for a changing set of interventions. Because those criteria concern different design properties, checking one cannot substitute for checking the other.
Common Classification Errors and Diagnostics
- Equating multiple arms with a platform. A fixed, multi-arm trial can have several interventions without a framework designed to continue as arms change. Check the protocol’s structure over time rather than counting arms at one snapshot.
- Equating biomarker measurement with an umbrella design. A biomarker may be measured for description or analysis without determining participant groups or treatment options. Check whether it affects the design’s allocation pathways.
- Treating the labels as mutually exclusive. A trial can be organized around disease subgroups and also have treatments enter or leave within a continuing framework. Assess both dimensions separately.
- Classifying from the trial name alone. Names may use “platform” or “umbrella” loosely, or may emphasize only one feature. Verify against the protocol and amendments.
- Confusing enrollment counts with structure. Counts and percentages summarize participants; they do not show whether a group is protocol-defined or whether arms change over time.
A useful diagnostic is to sketch two simple records from the protocol: a participant-group diagram showing whether distinct groups affect treatment options, and a dated list of which arms are available. If the diagram is unclear, the grouping may not be a defining umbrella feature. If the dated list shows only a fixed set of arms, a platform claim needs other justification. These checks classify the design; they do not establish that treatment comparisons are unbiased or that results can be interpreted in a particular way.
What the Labels Do—and Do Not—Tell You
The labels summarize trial organization, but they do not determine the treatment-effect estimand, the validity of a comparison, or the appropriate analysis. In particular, knowing that a framework changes over time does not by itself say how participants enrolled in different periods should be compared. Likewise, knowing that participants belong to different disease groups does not specify the treatment-to-group mapping or justify combining results across those groups.
A fixed set of arms can still be multi-arm without being a platform, and a trial with subgroup-specific treatment pathways can be an umbrella without changing its arms over time. Conversely, the combination of subgroup organization and a changing arm set is entirely possible. When describing a design, state the observed features rather than forcing a single label to carry more information than it can.
Check Your Understanding
Answer each question by identifying the relevant structural feature. Full solutions follow.
- A protocol measures a biomarker in every participant, but biomarker status does not affect eligibility or treatment options. Is this enough to call the design an umbrella trial?
- A trial has four treatment arms, all fixed for the entire study, and no subgroup-based treatment pathways. Does the number of arms alone establish a platform or umbrella structure?
- A disease trial organizes participants into two protocol-defined groups with different treatment options. The protocol later adds another treatment while recruitment continues. Which features are present?
- A hypothetical enrollment snapshot has 90 participants in group A and 60 in group B. Calculate the total and each group’s percentage. Do these numbers alone classify the trial?
Solutions. 1. No. Measuring a biomarker alone is insufficient; the groups must be part of the design in a way that affects eligibility or treatment options. 2. No. Four fixed arms alone establish neither defining feature. 3. Both umbrella and platform features are present: subgroup organization supports the former, and an added treatment within a continuing framework supports the latter. 4. The total is \(90+60=150\). Group A is \(90/150=0.600\), or 60.0%, and group B is \(60/150=0.400\), or 40.0%. These are descriptive enrollment summaries; classification requires the protocol’s grouping rules and treatment timeline.