Why a Shared Vocabulary Matters
“Adaptive trial” describes a broad family of designs, not one particular change. In one trial, accumulating information may lead to a change in the number of participants; in another, it may lead to a change in which treatments are studied. These are different adaptations, with different implications for the trial’s design and interpretation. A consistent taxonomy makes it possible to say precisely what changes and when it changes, before considering the details of how a change is implemented.
Here, an adaptation means a modification to a trial’s design or conduct during the trial, informed by information that becomes available as the trial proceeds. The earlier material on defining adaptive clinical trial designs distinguishes adaptive designs from designs that remain fixed. This tutorial supplies a vocabulary for describing the kinds of modifications that can occur. A category identifies the feature that changes; it does not, by itself, establish that a change is appropriate, statistically valid, or beneficial.
The “What Changes” Categories
The categories refer to different features of the design. They are useful labels, not mutually exclusive boxes: one adaptation can affect several features at once. When a scenario mentions a change, first identify the object being changed rather than guessing its category from the reason for changing it.
| Category | What changes | Illustrative description |
|---|---|---|
| Sample size | The planned number of participants or the amount of enrollment. | Increase the planned total enrollment from 300 to 360 participants. |
| Treatment arms | Which treatment groups are included in the trial. | Stop assigning new participants to one study arm. |
| Target population | Which participants are eligible or intended to be studied. | Revise an eligibility criterion so that a defined group may enroll. |
| Dose or regimen | The amount, schedule, or manner of delivering a treatment. | Change the assigned dose or dosing schedule in a study arm. |
| Allocation | How eligible participants are assigned among the available arms. | Change the relative assignment rates for two arms. |
| Outcomes or follow-up | What is measured, how it is defined, or when participants are followed up. | Change the planned follow-up duration or measurement schedule. |
These distinctions prevent common category errors. Changing the assignment ratio among existing arms is an allocation change; removing an arm changes the set of treatment arms. Altering a dose within an existing arm is a dose or regimen change, even if it also changes what participants receive. Revising eligibility changes the target population; changing the number of participants alone does not. A change to a measurement schedule belongs under outcomes or follow-up even when the motivation is to reduce participant burden.
An adaptation’s reason is not its category. For example, a team might seek greater precision, but that aim does not tell us whether the adaptation changes sample size, allocation, or another feature. Likewise, the label for what changes does not specify the decision rule or the analysis that will be used. Those are separate questions.
Worked Example: Separating Sample Size From Allocation
In a hypothetical trial, 240 participants are planned for two arms, with equal assignment. At an interim review, the design is revised to plan for 300 participants, while the equal assignment rule remains unchanged. Classify the change.
The planned number of participants changes from 240 to 300, so this is a sample-size adaptation. The allocation rule remains equal between the two arms, so this scenario does not describe an allocation adaptation. The fact that a larger trial will yield more participants in each arm does not itself mean the assignment ratio changed.
For contrast, if the total planned enrollment stayed at 240 but the assignment rule changed from equal allocation to a 2:1 ratio, the category would be allocation. The total sample size and the allocation ratio describe different design features.
Classifying When the Change Occurs
Timing adds a second description to the category. A useful account distinguishes the decision time—when the choice to modify the design is made—from the effective time—when the revised design begins to apply. They need not be identical. For example, a decision may be made at a review point and then apply only to participants enrolled afterward.
For a high-level description, use the trial’s stage when the decision is made: before enrollment begins, during enrollment, or after enrollment has ended but before the planned analysis is completed. If information accumulated during enrollment informs a decision made then, describe it as a within-trial adaptation. This timing label does not by itself tell us which data were reviewed, what rule was used, or whether the change applies to people already enrolled; those details require a fuller description.
Timing alone is not enough to identify an adaptive decision. A protocol may be amended during a trial for operational reasons without the change being informed by accumulating trial information. Conversely, naming the change as “sample size” does not reveal when it was decided or when it took effect. State both dimensions when the information is available, and say when it is not.
Worked Example: Decision Time and Effective Time
In a hypothetical trial, 180 participants have enrolled when a review informs a decision to add a new treatment arm. The decision is made during enrollment, and only participants enrolled after the change can be assigned to the added arm. Classify the adaptation.
The what category is treatment arms, because the set of available treatments changes. The when decided label is during enrollment. The when effective description is for participants enrolled after the change; it does not apply to the 180 already enrolled as described. Keeping decision time and effective time separate makes the account precise without assuming how the earlier participants will be followed or analyzed.
If the scenario said only that an arm was added, the category would still be identifiable, but the timing and the participants to whom the change applies would remain unspecified.
Compound Adaptations and a Classification Method
A single decision can modify multiple features. For example, adding an arm at a revised dose may change both the set of treatment arms and the dose or regimen. If the same decision also changes assignment rates across arms, allocation changes too. Naming only the most visible feature can conceal important parts of the design.
A simple representation makes this point explicit. Let \(c\) be a six-entry description of a scenario, with one entry for each category in the table. Set an entry to 1 if that feature changes and 0 if it does not. This is a descriptive device, not a statistical model or a decision rule.
Here the entries refer, in order, to sample size, treatment arms, target population, dose or regimen, allocation, and outcomes or follow-up. A value of 1 flags a category that applies. More than one entry may equal 1, so the notation avoids forcing a compound adaptation into a single label. It also does not claim that every scenario can be fully represented by six yes-or-no answers: the timing and details of each change still need to be stated in words.
Use the following sequence when classifying a scenario:
Ask what is different: the number enrolled, the available arms, eligibility, treatment dose or schedule, assignment, or measurement and follow-up.
Check whether one decision changes more than one feature, and record each category that applies.
State when the decision is made and, if known, when it takes effect and which participants it applies to.
Do not infer a decision rule, rationale, or analysis plan from a category label alone.
This method follows from the structure of the classification: the six categories answer different “what changes?” questions, while timing answers a separate “when?” question. A description is incomplete if it supplies only one dimension when the other is relevant or available.
Worked Example: A Change With Several Categories
In a hypothetical trial, 120 participants have enrolled in two arms using equal assignment. Following a review during enrollment, the design adds a third arm at a different dose from the existing active-treatment arm. The revised design also assigns new participants to the three arms in a 2:1:1 ratio. The original total enrollment target remains unchanged.
The set of available treatments changes, so treatment arms applies. The new arm has a different dose, so dose or regimen also applies. The assignment ratio changes, so allocation applies. Because the total enrollment target is unchanged, this description does not indicate a sample-size change. No eligibility or outcome change is stated.
The decision is made during enrollment and applies to new participants under the scenario. It would be misleading to call this only an “arm adaptation”: that label would omit the dose and allocation changes. It would also be unsupported to claim that the target population changed, because no eligibility change is described.
Diagnostics, Alternatives, and Limits of the Taxonomy
A classification is going wrong when its label does not match the feature that changed, when a compound change is reduced to one category without explanation, or when timing is assumed rather than reported. A useful check is to rewrite the scenario as “before the change, the design specified __; afterward, it specified __.” If the blanks refer to different features, more than one category may apply.
A second check is to distinguish the six categories from nearby descriptions. The purpose may be improved precision or a better treatment comparison; the category identifies which design feature changes. The method may specify how a decision is reached; the category does not reveal it. Likewise, a change in assignment probabilities is not the same as changing the list of arms, and changing eligibility is not the same as changing enrollment size.
This taxonomy is more useful than calling every modification simply “an adaptation,” because it preserves distinctions needed to communicate a design. It is also more flexible than forcing each modification into exactly one class, because real decisions can affect several features. But it is not a substitute for a complete protocol description. The labels alone do not establish the statistical consequences of a change, whether it was specified in advance, how interim information was handled, or how the resulting data should be analyzed. Those questions require additional design and analysis detail.
Check Your Understanding
Classify each scenario by what changes and, where stated, when the change occurs. Solutions follow the questions.
- A trial retains two treatment arms and its planned total enrollment but changes the assignment ratio from 1:1 to 3:2. Which category applies?
- During enrollment, a review informs a change to an eligibility criterion. Name the category and the timing.
- A study changes the planned follow-up from 12 months to 18 months, with no other stated change. Which category applies?
- A new treatment arm is added at a dose different from the dose in the existing active arm, and new participants are assigned across the arms using a revised ratio. Name all categories indicated.
- A scenario says that the total enrollment target increased, but gives no information about treatment arms, eligibility, allocation, or outcomes. What can and cannot be classified?
Solutions. (1) Allocation: the assignment ratio changes while the number of arms and planned total enrollment stay fixed. (2) Target population, decided during enrollment; the eligibility criterion changes who may participate. (3) Outcomes or follow-up, because the planned follow-up duration changes. (4) Treatment arms, dose or regimen, and allocation: the available treatment set, a dose, and assignment ratios all change. (5) Sample size can be classified because the planned enrollment target changes. The other categories cannot be inferred from the information given, and the timing is unspecified.