Why an Adaptation Needs Its Own Decision Record
An adaptive design can be prespecified and statistically justified yet still be hard to review if the trial file does not show what information was available when a decision was made, which rule was applied, who recommended the action, who authorized it, and when it took effect. A decision record makes that sequence traceable. It is not just a summary of the outcome: it shows how the study moved from the planned decision process to an operational action.
The record should be made for each adaptation decision, including a decision to retain the current design when the prespecified process evaluated an adaptation and selected no change. A later reader should be able to distinguish that deliberate outcome from a decision that was never considered or documented. The record complements, rather than replaces, the protocol, analysis plan, committee documentation, source data, and controlled trial documents.
What a Complete Record Connects
A useful record is organized so that a reviewer can follow the decision in order, without having to infer missing steps from scattered emails or later summaries. At a minimum, record the following.
- Decision point and information cutoff. Identify the planned or triggered review, the date and time of the data cutoff, the time zone, the data extract or report used, and the population or records included. State what information was available to each decision-making role, including whether treatment assignments or other confidential information were available.
- Rule applied. Name the protocol and analysis-plan versions in force, the relevant section or rule identifier, the exact decision criterion, and the inputs used to evaluate it. Record the result of the rule, including an explicit “no change” outcome where applicable. Do not substitute a broad description such as “the interim criteria were met” for the criterion and result.
- Recommendation. Record the recommended action, the person or committee making it, and the date and time of the recommendation. If the recommendation differs from the rule’s direct output, document the reason and the applicable authority for that discretion; do not make a discretionary departure appear to be an automatic rule output.
- Authorization. Identify who had authority to approve the action, the decision and rationale, and when authorization occurred. A committee recommendation and a sponsor or other authorized party’s approval are different events and should be recorded separately.
- Implementation. State exactly what changed, who implemented it, and the effective date and time. Identify operational artifacts affected, such as a randomization configuration, enrollment limit, or trial instruction, with their controlled version identifiers. If approval and implementation occur on different dates, preserve both dates.
- Supporting documentation. Link or reference the source report, committee record, recommendation, authorization, relevant controlled documents, and evidence of implementation. Use identifiers that allow a reviewer to retrieve the actual versions, not just generic names such as “minutes” or “final protocol.”
A clear chronology matters. The record should not imply that an action was authorized before the authorization occurred, or that an implementation date was the date of the committee meeting when the operational change happened later. If a record is entered after an event, preserve the original event date and the actual entry date; never backdate the record to make the sequence appear contemporaneous.
Traceability as a Practical Test
The central test is whether the record supports a complete chain from evidence to action:
This is a process test, not a statistical theorem. Its logic is simple: if any link is missing, a reviewer may be unable to determine whether the action followed the applicable plan, whether the right authority approved it, or whether the trial actually implemented the authorized action. A decision record is sufficiently traceable when another qualified reviewer can identify the same relevant inputs, locate the governing rule, understand the recorded result, and verify the subsequent approvals and operational change.
That test depends on several conditions. The governing documents must be identifiable by version and effective date; the information cutoff and data source must be specified; roles and decision authority must be clear; and records must be stored with suitable access controls and retained in a way that preserves their integrity. If a key report is revised, the record should point to the version actually considered, not silently replace it with a later version.
Worked Example: Recording a Blinded Sample-Size Adaptation
Hypothetical trial. A protocol specifies that, at a planned review, a blinded pooled standard deviation of 12 points or less retains the total sample size of 240; a value greater than 12 and no greater than 15 increases it to 300. The rule uses the blinded pooled estimate and does not authorize a treatment-effect-based decision. The trial’s independent committee reviews the report and makes a recommendation; the designated sponsor official authorizes the action.
The cutoff report includes 80 participants with evaluable outcome data. It reports a blinded pooled standard deviation of 13.2 points. Applying the specified bands gives \(12 < 13.2 \leq 15\), so the rule’s output is an increase to 300 participants. The record should capture the result and the input, not merely say “sample size increased.”
Record the decision point. Enter the review identifier, the data cutoff date and time with time zone, the report identifier and version, and that the committee received blinded pooled information. Note that the available report contained 80 evaluable outcomes and did not disclose treatment-specific estimates to the committee.
Record the rule and recommendation. Identify the protocol version and section containing the two thresholds, reproduce the applicable criterion, and record the reported value as 13.2 points. The committee’s recommendation is “increase total sample size to 300,” with the recommendation date, time, and committee record identifier.
Record authorization and implementation. Separately identify the sponsor official who authorized the increase and the authorization date and time. Record the date the enrollment system was updated, the responsible implementer, the revised target, and the controlled configuration version. Attach or link the cutoff report, committee recommendation, authorization, and implementation evidence. This preserves the distinction between the rule’s output, the recommendation, the approval, and the operational change.
Worked Records Should Show the Actual Decision Path
The examples below use invented trial details. They illustrate how to record the process; the hypothetical criteria are not general recommendations for choosing adaptation rules.
Worked Example: Documenting a Committee Recommendation to Stop an Arm
Hypothetical trial. A protocol states that an experimental arm is recommended for discontinuation if fewer than 25% of its first 100 evaluable participants meet a defined response endpoint at the interim cutoff. An independent committee reviews unblinded arm-level results. The protocol assigns the committee the recommendation role and the sponsor’s designated medical officer the authorization role.
The cutoff report identifies 100 evaluable participants in Arm B, of whom 22 meet the endpoint. The observed proportion is \(22/100 = 0.22\), or 22%. Since \(0.22 < 0.25\), the prespecified condition for recommending discontinuation is met. This calculation establishes the rule result; the record should not imply that the arithmetic alone authorized stopping the arm.
The decision record identifies the report version, cutoff, Arm B population, and the committee’s access to unblinded results. It cites the exact protocol version and rule, records the numerator, denominator, calculated proportion, and threshold, then states the committee’s recommendation and the date of its meeting. A separate entry records the medical officer’s authorization, with date and rationale. Finally, the record identifies when recruitment to Arm B was disabled, who made that system change, and the configuration version that took effect. The report, committee record, authorization, and system-change evidence are all referenced. The sequence makes clear which information supported the recommendation and who had authority to approve the action.
Worked Example: Recording an Allocation-Ratio Change
Hypothetical trial. A prespecified rule permits changing a 1:1 allocation ratio to 2:1 after a scheduled review if a specified operational threshold in the protocol is met. The trial’s independent committee evaluates the required report and recommends whether to make the change; the authorized sponsor representative approves it.
At the review, the report states that the protocol-defined threshold is met. The record should identify the exact report and version, the cutoff and included participants, the threshold as written in the applicable protocol version, and the reported result. It should state that the committee recommended a change from 1:1 to 2:1, who made that recommendation, and when. It should then identify the sponsor representative’s separate authorization and its date.
Suppose authorization occurs on 3 September and the randomization system is changed on 5 September. The record must give both dates, name the person responsible for the system update, and identify the configuration version that implements 2:1 allocation. It should also link evidence that the configuration was checked against the authorized setting. Recording only “ratio changed to 2:1 on 3 September” would be inaccurate if that date was approval rather than implementation. The separate dates preserve the actual chronology.
Diagnostics, Failure Modes, and Alternatives
Several warning signs indicate that a record may not support reliable review:
- The rule is described only in general terms, or the protocol version and section are absent.
- The record states a conclusion without identifying the cutoff, report, inputs, or population used.
- A recommendation is presented as authorization, or the authorized decision-maker cannot be identified.
- The implementation date is missing, conflicts with system or enrollment records, or is confused with the meeting date.
- Supporting documents are named but cannot be retrieved, or the linked files do not match the versions available at the decision time.
- The record is written retrospectively from memory, or later edits obscure who entered or changed information and when.
A structured decision form is often easier to review than free-form minutes because it prompts the same required fields at each decision point. Minutes can still be important evidence of deliberation, questions, and rationale, but they may not reliably capture cutoff metadata, version identifiers, or implementation details. The two formats can work together: the structured record provides the consistent traceability index, while minutes and reports supply supporting detail.
A brief narrative prepared later may be useful for explaining the decision to a broader audience, but it is not a substitute for contemporaneous evidence of what was known and authorized at the time. Conversely, a collection of source documents without a decision record may contain all the pieces but leave the reviewer to reconstruct their order and relationship. The practical aim is not duplication for its own sake; it is a clear, controlled path from the governing rule and available information to the action actually taken.
The record also has limits. It cannot make an incomplete or ambiguous prespecified rule adequate, establish that a statistical procedure controls error, or replace appropriate data governance and access controls. Nor does a complete record by itself establish that an adaptation was scientifically justified. It documents the decision process and its evidence so those questions can be reviewed against the relevant design and regulatory rationale.
Common Mistake: Treating the Decision as One Event
Why it is inadequate: It combines the review, recommendation, authorization, and implementation into one vague event. It does not identify the information cutoff, rule, authorized decision-maker, or effective system change.
Correction: Record each step separately, with its date, responsible role, source documents, and applicable version. If a decision is not implemented immediately, preserve the interval and the actual effective date rather than implying that approval and implementation were simultaneous.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A committee recommends an adaptation on 10 May, the sponsor authorizes it on 12 May, and the randomization system is changed on 14 May. Which date should the record call the implementation date?
- Why is “the interim criterion was met” weaker than recording the criterion, its input, and its result?
- A decision record cites “final interim report” but gives no version identifier. What review problem does this create?
- Why should a decision to make no design change still be recorded when a planned adaptation review has taken place?
Solutions. (1) The implementation date is 14 May; 10 May and 12 May are the recommendation and authorization dates and should also be recorded. (2) The short statement does not show which rule, data cutoff, input, or version was used, so a reviewer cannot verify the result. (3) The reviewer cannot establish which report was considered or whether the cited report is the version available at the decision time. (4) Recording the outcome distinguishes a considered, rule-based no-change decision from a review that was not performed or documented.