Why Interim Decisions Need Governance
An interim analysis can affect enrollment, treatment assignment, follow-up, or whether a trial continues. Those consequences depend not only on the statistical result but also on who may review it, who has authority to decide, and how the decision is put into effect. A prespecified threshold does not answer those governance questions by itself.
Interim governance is the set of roles, permissions, decision pathways, and records established for responding to interim evidence. Its purpose is to make decisions consistent with the protocol and the trial’s regulatory commitments, while preserving appropriate confidentiality and allowing concerns to reach the people responsible for acting on them. The details should fit the trial, its jurisdiction, and its oversight arrangements; no single decision structure is appropriate for every trial.
The plan should align with the protocol, any applicable committee charter, the analysis plan, and the sponsor’s regulatory commitments. Prespecified adaptive rules and change control determine which trial adaptations are planned and how changes are governed. Interim governance makes explicit who applies those arrangements and responds to the results. If a planned rule is ambiguous, or the planned analysis cannot be completed as specified, the decision makers should not improvise a new rule without following the agreed escalation and change-control procedures.
Assigning Authority and Restricting Access
A useful governance plan distinguishes four functions that may be assigned to different people or groups: producing the analysis, interpreting the evidence, making or recommending a decision, and implementing the resulting action. Separating these functions can reduce conflicts of interest and prevent a person from acquiring decision authority merely because they can view a result. The appropriate separation depends on the trial’s design and oversight structure.
Authority should be stated in operational terms. For example, the plan can identify who may recommend continuation or stopping, who makes the final decision under the charter, who can authorize a protocol-defined operational step, and who must be notified. It should also specify the quorum or voting rule when a group decision is required, how a tie is handled, and what happens if a required decision maker is unavailable. A role label such as “the committee decides” is incomplete if the charter does not say how the committee reaches a decision and how that decision is communicated.
Access permissions should likewise be defined by role and purpose. A person who needs a treatment-level interim result to evaluate a decision may not need access to participant-level data or unrelated operational details. Conversely, a staff member responsible for maintaining the blind should not receive information that could inadvertently reveal treatment allocation. The plan should identify the permitted recipients, the information each role needs, and the authorized route for communicating a recommendation or decision. These controls reduce unnecessary disclosure; they do not replace safeguards for participant safety or required reporting.
| Governance function | Question to settle in advance | Example of a clear specification |
|---|---|---|
| Analysis | Who prepares the planned interim summary? | A designated statistical group runs the locked interim program and documents its output. |
| Review | Who may interpret the relevant interim evidence? | Only roles named in the protocol or charter receive the specified summary. |
| Decision | Who recommends or authorizes a trial action? | The charter names the decision-making body and its quorum and voting rule. |
| Implementation | Who carries out an authorized action? | The sponsor’s named operational lead communicates and implements the decision. |
These assignments should not obscure responsibility. A group may make a recommendation while the sponsor or another designated authority remains responsible for deciding or implementing a permitted action. The plan should say which is the case. For urgent safety concerns, it should also identify an appropriate route for prompt escalation; an efficacy decision pathway must not delay a response required to protect participants.
From Emerging Evidence to an Authorized Action
A well-governed process begins with the planned analysis and ends with a clearly communicated, authorized action—or a documented decision not to act. The steps below are a practical sequence, not a substitute for the protocol or applicable oversight requirements.
Check that the planned data cut, analysis population, model, and decision rule are the ones specified for the interim review. Identify missing inputs, deviations, or limitations that could affect interpretation.
Provide the prespecified summary only to recipients with a defined role in reviewing or deciding. Use the agreed channel and include any required context about uncertainty or data limitations.
The authorized person or group considers the result under the prespecified rule and charter. If the rule is not met, unclear, or impossible to evaluate, follow the escalation procedure rather than silently replacing the rule.
Document the decision, its basis, any conditions or dissent, and who approved it. Communicate the operational instruction to those who need it, with no more sensitive information than their roles require.
Confirm that the authorized action was carried out, identify any required notifications, and set the next review or reassessment point when applicable.
The process also needs a route for evidence that is important but does not fit the planned efficacy trigger. A safety concern, data-integrity issue, or unexpected pattern may warrant urgent review under a separate procedure. The person who observes such a concern should know whom to contact, how promptly to do so, and what information to provide. A signal that does not satisfy an efficacy stopping boundary is not, by that fact alone, evidence that no other action is warranted.
Worked Example: A Prespecified Efficacy Boundary
Suppose a hypothetical trial’s approved interim rule says to recommend stopping for efficacy if the posterior probability that the treatment effect exceeds a clinically relevant margin is at least 0.95. The charter assigns the review to a named committee and states that a recommendation requires at least four of five members to support it. The final action is authorized by the sponsor’s designated decision maker.
The planned analysis produces a posterior probability of 0.972. Applying the statistical rule gives:
The efficacy boundary is met. Assume that four of the five committee members vote to recommend stopping. The quorum is five, so the decision is valid under the hypothetical charter, and the required support count is \(4\) of \(5\). The recommendation is then sent to the sponsor’s designated decision maker, who records and authorizes the applicable action under the trial’s governance arrangements. Meeting the statistical boundary does not itself mean that the committee or sponsor has taken an action; each step in the authority chain still has to occur.
Worked Example: A Boundary Is Not Met
In a different hypothetical interim review, the same kind of prespecified rule requires a posterior probability of at least 0.95. The analysis reports 0.91, and the charter says that failure to meet the boundary means the trial continues without an efficacy-based adaptation at this review.
The comparison is:
The efficacy stopping boundary is not met. Under the stated rule, the authorized outcome is continuation without that efficacy-based adaptation, not a claim that the treatment is ineffective. The posterior probability is below the decision threshold; it does not establish a zero treatment effect. The record should identify the analysis and rule applied, the result, the authorized outcome, and any other concern considered under a separate procedure. If a decision maker believes the result raises an issue not covered by the rule, the appropriate response is to use the escalation pathway, not to change the threshold informally.
Worked Example: An Emerging Concern Outside the Efficacy Rule
Consider a hypothetical trial with an interim review scheduled after 100 participants have completed a defined safety follow-up period. The charter says that a specified serious safety concern must be escalated promptly to the designated safety contact, regardless of whether an efficacy boundary is met. At review, one such concern is identified in the data available for the planned review. It does not meet the trial’s efficacy stopping rule.
The efficacy result and the safety route answer different questions. Even if the efficacy calculation is, for example, \(0.91 < 0.95\), that comparison only indicates that the efficacy boundary is not met. It does not cancel the separate safety instruction. The responsible reviewer sends the concern through the named safety route, records when it was recognized and escalated, and follows the authorized process for assessing whether immediate trial action is needed. No numerical efficacy threshold can be substituted for this required escalation.
Justifying the Separation Between Rule and Authority
The distinction between a statistical trigger and an authorized decision can be expressed simply. Let \(E\) represent the planned interim evidence, \(R\) the prespecified rule, and \(G\) the governance arrangements. The rule maps evidence to a recommendation; governance maps that recommendation, along with any required contextual review, to an authorized action:
This separation is necessary because the same recommendation can lead to different permissible actions under different protocols or charters. For example, a rule might call for a recommendation to stop, while the named authority, voting requirement, and participant-protection procedures determine how that recommendation is reviewed and implemented. The statistical rule does not specify who may see the result or who may sign off.
The separation also helps identify a governance failure. If a result is received by someone who is not authorized to review it, the issue is not repaired by showing that the result was computed correctly. If an authorized reviewer receives a result but the charter gives no clear route to a decision, a numerical threshold alone cannot resolve the missing authority. Both the evidence process and the decision process must be defined.
Documentation, Diagnostics, and Common Pitfalls
A decision record should make it possible to understand, later, what was considered and what was authorized. Depending on the trial’s plan, it may include the interim review date and data cut; the analysis version and result; relevant limitations or deviations; the decision rule applied; the roles involved; the recommendation, vote, or approval; the rationale and any dissent; the action communicated; and the date implementation was confirmed. The record should also capture when escalation was required and to whom the concern was sent. The exact recordkeeping system and technical audit-trail requirements depend on the trial’s procedures.
Governance is going wrong when responsibilities are unclear, information reaches an unintended recipient, a decision is communicated without required approval, or an action occurs without a record of its authority. Other warning signs include a committee interpreting a rule that is not in its charter, a threshold being changed after results become known without an authorized process, or a safety concern being held until a routine efficacy review. These are reasons to pause the affected decision pathway, notify the designated responsible roles, preserve the relevant records, and follow the established escalation process. The appropriate response depends on the nature and seriousness of the issue.
- “The boundary was crossed, so the trial has stopped.” A boundary may generate a recommendation or trigger a review. The actual authority and required approvals depend on the protocol and charter.
- “The efficacy rule was not met, so nothing needs to be done.” The rule addresses its stated efficacy decision. Separate safety or operational concerns may require escalation.
- “Everyone on the trial team should see the interim result.” Access should match defined responsibilities. Broad access can expose sensitive results to people who do not need them and can undermine the intended safeguards.
- “A good analysis is enough documentation.” The analysis shows what was calculated, not who reviewed it, what authority was used, or whether the resulting action was implemented.
The main alternative to an explicit governance plan is informal, case-by-case decision making. That may appear faster, but it leaves authority, access, and escalation open to interpretation precisely when interim evidence can create pressure to act. A prespecified process is more predictable and reviewable, though it cannot anticipate every unexpected event. For unusual situations, the plan needs a clear route to authorized review rather than a pretense that every circumstance can be reduced to one numerical rule.
Check Your Understanding
Answer each question using the governance principles in this tutorial. Brief solutions follow the questions.
- A prespecified efficacy threshold is met. Why does that fact alone not establish that the trial has stopped?
- Name two items that should be settled in advance if a committee is responsible for an interim recommendation.
- A planned efficacy boundary is not met, but a separate urgent safety concern is identified. What should happen?
- Why should access to interim results be assigned by role rather than given to all trial staff?
Solutions. (1) The threshold generates a statistical recommendation or trigger; the protocol or charter specifies who reviews it, who authorizes an action, and how it is implemented. (2) Examples include who may receive the summary, the committee’s quorum or voting rule, how ties are handled, and who is authorized to act on its recommendation. (3) The concern should follow the separate, prespecified safety escalation route; failure to meet an efficacy boundary does not cancel that obligation. (4) Role-based access limits unnecessary disclosure, helps preserve confidentiality and appropriate blinding, and ensures that people receive information consistent with their responsibilities.