Workshop Scenario: Two Adaptations, Different Information Needs
An access plan is easiest to assess when the trial’s decisions and the information needed for each decision are made explicit. This workshop uses a hypothetical, two-arm randomized trial with a continuous primary outcome. The protocol begins with a planned total of 240 participants, allocated equally between treatments A and B. At a prespecified point, an adaptation team may increase the total sample size, up to a maximum of 360, using a blinded estimate of outcome variability. Separately, an independent efficacy review may consider treatment-specific interim results and recommend stopping for efficacy under a prespecified rule.
These are distinct tasks. A blinded sample-size update is intended to use nuisance information, such as outcome variability, without exposing the treatment contrast to the adaptation team. An efficacy review needs treatment-specific information to assess the evidence for a difference. It is not enough to call a process “blinded” or “independent”: the plan must say which data each role receives, through which system, for what purpose, and what it may communicate.
The scenario assumes that randomization is concealed from site staff and participants, and that the sample-size rule, review timing, maximum sample size, and efficacy decision rule are specified before the interim analyses. It also assumes that the efficacy review is conducted by an independent committee supported by a statistician who is not responsible for routine trial operations or for preparing the blinded adaptation report. These assumptions define the access problem; they do not, by themselves, establish that either adaptation is statistically valid or that the trial’s error rate is controlled.
Specify the Information Before Assigning Roles
Start by distinguishing the information products, not merely the names of the analyses. For the sample-size update, the adaptation team needs the prespecified blinded summary used by the rule, such as a pooled estimate of variability, together with the current recruitment and follow-up counts needed to interpret it. The team should not receive treatment-specific means, an estimated treatment difference, conditional power based on that difference, or an unmasked listing from which the contrast can readily be reconstructed.
For the efficacy review, the independent committee needs the treatment-specific interim results and the statistical summaries required by its charter and prespecified efficacy rule. That information may include arm-specific outcome summaries, the estimated contrast, its uncertainty, and the analysis population definition. The sponsor’s operational team, investigators, site staff, and other personnel responsible for recruitment or participant care do not need those comparative results to carry out the blinded sample-size update.
A blinded pooled summary is not necessarily free of all treatment-effect information. For example, if outcomes in each arm have the same within-arm variance \(\sigma^2\), the treatment means differ by \(\delta\), and the arms are equally represented, the variance in the combined outcome distribution is:
The pooled variance can therefore reflect both within-arm variability and some contribution from the difference between arm means. This does not make every blinded variance-based update inappropriate, but it does mean that the protocol must define and justify the estimator and its use. Calling an analysis blinded is not a substitute for checking what its output reveals or how the adaptation rule behaves.
Worked Access Plan
The following role-by-information matrix is a compact starting point. “Yes” means the role receives the information for the stated task; “No” means the proposed plan does not provide it. A role’s access should be judged across reports, meetings, electronic systems, and informal communications—not only by whether a person can open a particular file.
| Role | Blinded pooled update summary | Unblinded arm-level efficacy results | Permitted communication |
|---|---|---|---|
| Blinded adaptation statistician | Yes, as specified in the rule | No | Validated blinded summary and rule-based recommendation to adaptation team |
| Blinded adaptation team | Yes | No | Sample-size recommendation within protocol limits |
| Independent efficacy statistician | Not needed for the efficacy review | Yes | Confidential analysis outputs to the independent committee |
| Independent efficacy review committee | Not needed for its efficacy decision | Yes | Recommendation under its charter, with only the permitted operational message released |
| Sponsor operations and site staff | Only the operational instruction needed to implement the approved sample size | No | Implementation information, without comparative efficacy findings |
This matrix is not a universal allocation of access. A particular protocol may assign decisions differently. Its value is that it exposes mismatches: a role may be listed as blinded but still receive a report or meeting presentation containing treatment-specific information; or a role may be described as independent while also making operational decisions affected by the interim results.
Worked Example: Tracing the Blinded Sample-Size Update
At the scheduled review, 120 participants have been enrolled and the protocol’s blinded update rule is triggered. The data system contains outcomes and randomized treatment assignments. The adaptation rule uses a prespecified blinded variability summary to determine whether the total target remains 240 or increases, subject to the cap of 360.
Step 1: Identify the input. The blinded adaptation statistician receives the eligible outcome data in a form that does not display treatment assignments or treatment-specific summaries. The statistician also receives the prespecified data-status information required to apply the rule, such as the number of participants with an evaluable outcome.
Step 2: Check what the input can reveal. The statistician verifies that the analysis file, code, and output do not expose arm labels, treatment-specific means, or an easily reconstructable contrast. The statistician calculates only the prespecified summary and documents the dataset version and analysis procedure.
Step 3: Apply the rule. Suppose, hypothetically, that the prespecified rule maps the observed blinded summary to a target of 300 participants. The team receives the target and the information needed to approve it under the protocol, not arm-specific outcomes or the efficacy statistician’s findings.
Step 4: Communicate implementation. Once the target is approved, sponsor operations and sites receive the revised recruitment target and implementation instructions. They do not receive the interim efficacy report. The decision record should show that the target arose from the specified blinded rule, not from informal discussion of the treatment contrast.
Worked Example: Tracing the Independent Efficacy Review
At the same scheduled interim point, the independent efficacy statistician prepares an efficacy analysis for the independent committee. The committee’s charter permits a recommendation to stop early for efficacy if the prespecified criterion is met.
Step 1: Identify the analysis team. A statistician authorized to access treatment assignments receives the data needed for the efficacy analysis. The statistician’s access is justified by that review role, not by the blinded sample-size task.
Step 2: Restrict the report route. The treatment-specific results and associated interpretation go to the committee through its confidential channel. They are not placed in the routine trial-management workspace or attached to a general meeting agenda.
Step 3: State the decision message. Suppose, hypothetically, that the committee concludes that the efficacy criterion is met and recommends stopping. It sends the authorized recommendation through the agreed channel. The operational message might instruct the sponsor to implement a stop, without distributing the arm-level estimates to staff who do not need them.
Step 4: Preserve the boundary. The committee’s unblinded report is retained in the restricted record. The blinded adaptation team is not given the report to “help interpret” the pooled variability estimate. If the two processes need to coordinate timing or an operational action, that coordination should use a prespecified message that does not disclose the comparative findings.
Why the Separation Matters: A Short Justification
The access separation follows from the different functions of the two decisions. The blinded rule is designed to change the sample-size target using a defined summary, while withholding the treatment contrast from the adaptation team. The efficacy committee, by contrast, cannot assess a treatment-specific efficacy criterion without treatment-specific information. Giving both products to the same decision-makers can defeat the intended masking even if each analysis is separately described as “independent” or “blinded.”
A useful audit question is whether an output contains enough information to infer a protected quantity. For example, if the adaptation team receives both the pooled mean and separate means under anonymous labels, the labels may be easy to map back to treatments using known coding or other reports. Likewise, a sequence of reports with small subgroup counts can reveal more than any single report. Therefore, the boundary concerns the information content and combination of outputs, not just the file’s title.
Write down the sample-size decision and the efficacy decision separately, including who is authorized to make each one.
For each decision, specify the exact data, summaries, and interpretation needed. Exclude outputs that are not needed for that role’s task.
Follow the information from source data through analyst, system, report, meeting, decision record, and operational communication.
Check whether shared staff, system permissions, report contents, or combined outputs could expose treatment-specific information to a role intended to remain blinded.
Record who approved access, what is communicated, how exceptional disclosures are handled, and how access and distribution logs are reviewed.
Worked Example: Finding and Repairing an Access Conflict
A proposed plan assigns one statistician to prepare both the blinded pooled update and the unblinded efficacy report. The statistician sends the adaptation team a pooled variability summary but also attends the efficacy committee meeting, where arm-specific results are discussed. The statistician says the adaptation team will see “only the pooled number.”
Conflict. The adaptation statistician has access to the unblinded findings through the committee meeting. The team’s formal report may be pooled, but the person preparing it is not insulated from the efficacy results. That creates a risk that choices about data handling, timing, or explanation of the update are influenced by information unavailable to the rest of the team. It also makes the claim that the adaptation workstream is separated from the efficacy review difficult to verify.
Repair. Assign the blinded update to a statistician who does not receive the unblinded report or attend the efficacy discussion. Keep the efficacy analysis and committee materials in a restricted workspace with separate permissions. Define a limited operational message for any decision that must be communicated across workstreams. If staffing constraints make complete separation impossible, document the specific compromise, the safeguards, the affected decisions, and the reason it is acceptable under the protocol and governance arrangements; do not describe the workstream as fully separated when it is not.
The repair does not prove that the statistical procedures are valid. It addresses who can know what and who can act on it. Statistical review must separately check that the blinded update rule is prespecified and appropriate, and that the efficacy review follows its prespecified analysis and decision criteria.
Diagnostics, Alternatives, and Limits
The plan should be tested against the actual information environment before the interim review. Useful checks include reviewing sample reports with treatment labels and contrasts removed; verifying that access permissions match the role matrix; confirming that data exports cannot be joined to a treatment-allocation file by the blinded team; and checking meeting invitations, minutes, email distribution lists, and shared workspaces. A tabletop rehearsal can reveal whether a seemingly harmless operational question would require disclosure of an efficacy result.
- Warning sign: one person crosses workstreams. Check whether anyone who sees unblinded results also prepares, validates, or explains the blinded adaptation output.
- Warning sign: reports reveal more than their stated purpose. Check whether arm-specific counts, subgroup breakdowns, or repeated summaries make the protected comparison inferable.
- Warning sign: unrestricted data-system permissions. A written plan is insufficient if the role can open unblinded listings or dashboards in practice.
- Warning sign: informal disclosure. Calls, slides, meeting discussions, and status updates can leak interim findings even when formal reports are restricted.
The main alternative to a blinded update is to base adaptation on an unblinded estimate. That approach may be scientifically appropriate in some designs, but it changes the information access required and may affect the statistical properties of the adaptation. It should be assessed and governed as an unblinded adaptation, not treated as equivalent to the blinded route. Conversely, the blinded route may not use information needed to detect an efficacy effect. That is why an independent efficacy review, if planned, requires its own protected channel.
No access matrix can guarantee that information will never be inferred or disclosed. Small samples, unusual outcome patterns, staff turnover, shared software accounts, and urgent decisions can create edge cases. Nor does restricted access alone protect trial integrity if the adaptation rule is unclear, the data are unreliable, or the authorized recommendation is communicated inaccurately. The plan is a control structure that needs operational testing and oversight, not a substitute for sound statistical design.
Common mistake: “The adaptation team sees only a pooled summary, so it is blinded, even though the same statistician attends the unblinded efficacy meeting.” The correction is to assess access across people, systems, and communications. A pooled report does not preserve separation if its preparer has already seen the treatment-specific findings or if other outputs allow the team to reconstruct them.
Check Your Understanding
For each question, decide what the access plan should permit, then compare your reasoning with the solutions.
- The blinded adaptation analyst receives a file with outcomes and anonymous labels “Group 1” and “Group 2,” but no treatment names. Is this necessarily a blinded pooled input?
- Who should receive the efficacy statistician’s arm-specific interim report in the workshop scenario?
- A sponsor operations lead receives the revised sample-size target and asks for the treatment difference to explain the change to sites. What should the access plan provide?
- Give one access conflict in a plan where the same analyst prepares both reports, and state one repair.
- Why should an audit check system permissions and meeting materials as well as formal reports?
Solutions. (1) No. The two anonymous groups are still treatment-specific if the labels separate outcomes by arm; if the labels can be mapped to treatment assignments, the information is unblinded in substance. The blinded input should contain only the prespecified information needed for the update and should be checked for inferential disclosure. (2) The report should go to the authorized independent efficacy committee and its designated unblinded support, through the restricted channel specified by the plan. It should not routinely go to the adaptation team, sponsor operations, or sites. (3) The lead should receive the approved operational target and implementation instructions, not the treatment difference. The sample-size change should be explained by reference to the prespecified blinded rule, to the extent that such an explanation is authorized and accurate. (4) The analyst could carry knowledge of the efficacy result into the blinded update or disclose it in explaining the report. A repair is to assign separate analysts and restrict the efficacy report and meeting to the unblinded workstream. (5) Formal reports are only one route. System access, slides, meeting discussion, or linked outputs can expose the same information, so the audit must trace all realistic routes and verify that permissions and communications match the intended roles.