Why classify an adaptation scenario?
A proposal to change a trial can sound appealing while leaving important questions unanswered. “We will adjust the study if the interim results are disappointing” does not specify what will change, what information will prompt the change, or why the change is useful. Classifying the proposal makes those parts visible. It also helps identify what could go wrong before the trial begins.
This tutorial uses the distinctions established in the earlier material on trial adaptations and their rules. The task here is to apply them to short scenarios: name the change, locate its trigger, state its intended benefit, and identify a main practical or interpretive risk. A risk is a question to investigate, not proof that an adaptation is unsuitable.
A useful classification is specific enough to separate these four elements. For example, “increase the sample size” names a change, not its trigger. “Increase the sample size if a planned review finds greater-than-expected outcome variability” adds a trigger. “So the trial can retain adequate power” states an intended benefit. “The review and final analysis must preserve the planned error control” identifies an interpretive concern.
A consistent way to classify proposals
Identify the trial feature that would be different after the decision: for example, the sample size, active arms, allocation ratio, eligible population, dose, outcome, or follow-up.
Specify the data, review, or operational event that prompts the action. Distinguish a measured trigger, such as a blinded estimate of variability, from a general intention to respond to “promising results.”
Say what the change is expected to improve, such as avoiding unnecessary enrollment, learning more about a useful treatment, or reducing exposure to a concerning treatment.
Ask whether the change complicates interpretation or error control, requires new operational arrangements, or makes the resulting evidence less relevant to the intended patients or care setting.
Describe what would happen without the adaptation. This makes the benefit and the cost of changing the design easier to assess.
This sequence is a classification aid, not a substitute for reviewing the actual rule. A complete protocol would need to state how the trigger is measured, who reviews it, when the decision occurs, and what action follows. It would also need an analysis plan appropriate to the design. Merely calling a change “adaptive” does not establish that its statistical or operational consequences are acceptable.
Worked scenarios
Worked Example: Reassessing the Sample Size
Hypothetical scenario. A two-group trial plans to enroll 240 participants and compare a continuous symptom score at 12 weeks. After 120 participants have completed the 12-week assessment, an independent statistician estimates the outcome variability without using the treatment-group difference. The protocol allows enrollment to rise to at most 360 if the estimated variability exceeds a prespecified threshold.
Adaptation. The proposed change is a sample-size increase, with the maximum changing from the original plan of 240 to 360 participants.
Trigger. The trigger is the blinded estimate of outcome variability at the scheduled review, not a finding that one treatment appears more effective. In this scenario, suppose the estimate exceeds the protocol’s threshold, so the rule increases the planned total to 360.
Intended benefit. Greater variability makes it harder to distinguish a given treatment difference from random variation. Increasing enrollment is intended to preserve the planned ability to detect a clinically relevant difference, rather than completing a smaller trial that may be inconclusive for that reason.
Main risk and fixed-design comparison. The review, stopping point, maximum sample size, and analysis must be prespecified and appropriate to the rule. Otherwise, using interim information to alter the design can complicate the interpretation of the final result and its error properties. Without adaptation, the trial would stop at 240 even if the variability were unexpectedly high; that is operationally simpler but could leave the study less informative than intended. The fact that the review is blinded does not, by itself, establish that every inferential concern is resolved.
Worked Example: Dropping a Treatment Arm
Hypothetical scenario. A randomized trial compares two experimental treatments, A and B, with a control. At a scheduled interim review, an independent committee examines the accumulated response and safety data using a prespecified rule. Suppose the rule indicates that B has little prospect of meeting the trial’s success criterion and no compensating safety advantage. The proposal is to stop assigning new participants to B while continuing A and control.
Adaptation. The treatment-arm structure changes: B is discontinued for future enrollment. This is not simply a change in the number of participants; it removes one treatment option from the ongoing randomized comparison.
Trigger. The trigger is the interim committee’s assessment under the prespecified rule. In the hypothetical outcome, the rule is met for B. The trigger is not just “the results look poor”: the protocol must define what evidence is reviewed and how the decision is made.
Intended benefit. Discontinuing B is intended to avoid assigning additional participants to a treatment with little expected value, while allowing enrollment and learning to continue for A and control.
Main risk and fixed-design comparison. The trial now contains participants enrolled under different sets of available arms and possibly different calendar periods. The final comparisons and their uncertainty must account for the design and interim decision process; the unadjusted comparison that would have applied in a fixed three-arm trial cannot automatically be assumed appropriate. The protocol must also specify how recruitment, allocation, and participant communication change. Without dropping B, the trial would preserve its original arm structure but continue assigning participants to B. The proposed adaptation trades that continuity for the possibility of focusing resources on the remaining comparisons.
Worked Example: Expanding Eligibility to a Broader Population
Hypothetical scenario. A trial initially enrolls adults aged 50 to 70 with a specified diagnosis. After the first 80 participants, a planned review of recruitment logs shows that enrollment is much slower than expected. The proposal is to allow adults aged 40 to 70 with the same diagnosis to enroll, subject to the same safety requirements.
Adaptation. The eligibility criteria change by lowering the minimum age from 50 to 40. This is a change to the target population, not a sample-size increase, even if it also makes recruitment faster.
Trigger. The trigger is the observed recruitment shortfall at the planned review. The scenario does not use treatment-response differences to justify widening eligibility.
Intended benefit. The broader eligibility range is intended to improve recruitment and make it more feasible to complete the trial on schedule.
Main risk and fixed-design comparison. Younger participants may differ from older participants in prognosis, treatment response, or relevant care. Combining their results could change what population the overall evidence describes. The protocol would need to explain why the added ages are appropriate and how their data will be handled in the analysis. Without the change, the trial would retain its original population definition but might recruit more slowly or fail to meet its enrollment target. Faster recruitment is not, on its own, evidence that the expanded population remains suitable for the study question.
Compare changes by what they alter
Several proposals can share a trigger or a hoped-for benefit but still be different adaptations. For example, slow recruitment might motivate increasing the sample size, broadening eligibility, adding sites, or changing the duration of recruitment. Those choices alter different features of the trial and carry different risks.
| Proposed change | Feature altered | Example of a possible trigger | Risk to examine |
|---|---|---|---|
| Increase sample size | Number of participants planned | Prespecified reassessment of blinded variability | Whether the decision rule and final inference remain appropriate |
| Drop a treatment arm | Set of treatment options | Interim evidence under a prespecified stopping rule | Selection effects, changed comparisons, and operational transition |
| Change allocation ratio | Relative probability of assignment to groups | Accumulating safety information or a prespecified efficiency rule | Unequal group sizes and possible changes to precision or conduct |
| Broaden eligibility | Population permitted to enroll | Recruitment shortfall at a planned review | Whether the expanded population supports the intended interpretation |
| Change outcome or follow-up | Measurement or timing of assessment | Feasibility or data-quality information | Comparability of measurements and potential for outcome-driven choices |
The table gives possible triggers, not universally appropriate rules. A safety concern, for instance, could prompt a dose change, a change in allocation, or stopping an arm. Classification depends on the action actually proposed, rather than the reason given for considering it.
Diagnostics, common mistakes, and limits
A classification is incomplete when it cannot answer practical questions. Check whether the trigger can be observed at the planned time; whether the rule says what action follows; whether the action is feasible for sites and participants; and whether the analysis is consistent with the design that may result. For changes to eligibility or outcomes, ask whether the revised definition still supports a clear interpretation. For changes based on interim comparisons, ask whether the decision process and final inference have been considered together.
One common mistake is to treat the intended benefit as if it were the trigger: “We will drop an arm to save resources.” Saving resources is a rationale, not a rule for deciding when to drop the arm. The correction is to state the trigger separately, such as a prespecified assessment of interim evidence, and then describe the expected resource benefit.
Another mistake is to assume that any planned adaptation is automatically harmless to interpretation, or that any adaptation automatically invalidates a trial. Neither conclusion follows just from the label. Consequences depend on the specific change, the information used to choose it, the timing, implementation, and analysis. A well-classified proposal makes these dependencies explicit, but classification alone does not prove validity or feasibility.
Finally, do not infer a causal or clinical benefit from the fact that an adaptation was made. A change may improve recruitment or concentrate enrollment, but whether the treatment comparison answers the intended scientific question requires a separate design and analysis assessment. The classification exercise identifies what should be assessed; it does not replace that assessment.
Check Your Understanding
For each prompt, identify the adaptation, trigger, intended benefit, and a main risk. Suggested answers follow the questions.
- A trial plans a blinded review of pooled outcome variability. If variability is higher than its prespecified threshold, it will increase the maximum sample size. What is the adaptation, and what is not the trigger?
- A committee uses a planned interim rule to stop enrollment to one dose because of a safety concern, while other groups continue. Name the adaptation and one practical risk.
- Recruitment is slower than expected, so investigators propose adding clinics but keeping the eligibility criteria and treatment comparisons unchanged. What feature changes, and what benefit is intended?
- A team proposes changing the primary outcome after seeing that another measured outcome has a more favorable treatment result. Identify the proposed change and one major interpretive risk.
- Why is “improve efficiency” not a sufficient description of an adaptation scenario?
Solutions. (1) The adaptation is an increase in planned sample size. The trigger is the prespecified blinded variability threshold; the expected power benefit is the rationale, not the trigger. A risk is whether the rule and final analysis adequately address consequences of the interim decision. (2) The treatment-dose structure changes because one dose is removed from future assignment. A practical risk is implementing the change consistently across sites and communicating it to participants; interpretation and analysis also need to reflect the design. (3) The trial’s site structure changes, not its eligibility criteria or treatment comparisons. The intended benefit is faster or more reliable recruitment. Added sites may introduce differences in conduct or measurement that need management. (4) The outcome definition changes. Selecting an outcome because its observed result is more favorable can make the final evidence difficult to interpret and can undermine the planned inferential framework. (5) Efficiency is an intended benefit, not a complete classification: the proposal must also state what changes, what triggers it, and what risks arise compared with a fixed design.