Why trace an interim result?
An interim comparison can affect a trial even when nobody outside the review group receives a table of treatment results. A changed recruitment target, a vendor’s revised instruction, or a comment made at a site may allow people to infer that the arms are performing differently. The integrity question is therefore not only who saw the report, but also what information moved, through which role, and what could be inferred from it.
This workshop follows one hypothetical result through a trial’s roles. The aim is to map access and communication controls, not to decide whether a treatment works or to prescribe a particular adaptive design. The protocol, analysis plan, review charter, and documented access plan govern the actual trial. They should specify which roles may receive which outputs and who may authorize an operational change.
Set up the hypothetical trial and result
Suppose a randomized trial compares Treatment A with Treatment B. At a scheduled interim review, the database contains a binary outcome for 160 participants: 80 assigned to each arm. The outcome is recorded as a response or no response. For this illustration, assume the data are sufficiently complete for the planned interim summary, assignments are correctly linked to outcome records in the restricted analysis environment, and the planned review is authorized. These assumptions allow us to trace the information; they do not establish that the interim estimates are unbiased or that the observed difference is meaningful.
A restricted analyst calculates response proportions by assigned treatment:
The observed response proportions are 55.0% for Treatment A and 43.75% for Treatment B, a difference of 11.25 percentage points in this interim sample. This arithmetic is descriptive. It is not, on its own, a test of treatment efficacy, a causal conclusion, or a decision to change the trial. The designated review body must interpret the prespecified review information under the applicable plan and charter.
Map the result across roles
A useful trace starts at the data and follows each handoff. The table gives a plausible pattern, not a universal assignment of responsibilities. A trial may use different structures, but its documented access plan should make the boundaries explicit before the interim review.
| Role | Information that may be needed | Control to check |
|---|---|---|
| Restricted data and analysis personnel | Participant-level data, treatment assignment, and the analysis specifications needed to produce the interim output | Role-based permissions, a controlled analysis environment, authorized code and data versions, and a record of access and output release |
| Independent review body | The arm-specific report and context required to review the interim result under its charter | Named membership, secure distribution, restricted discussion, and documented recommendations or decisions |
| Sponsor decision-makers | Only the information their defined decision role requires; some designs permit an unblinded decision group, while others route decisions through an independent body | A documented boundary between authorized decision-makers and sponsor personnel responsible for blinded trial operations |
| Blinded sponsor functions | Operational direction needed to implement an authorized decision, without the comparative result unless specifically authorized | Separate distribution lists, masked instructions, and communication that does not reveal arm-specific performance |
| Investigators and site staff | Site instructions needed to continue the trial or implement an approved change | No routine access to arm-specific interim summaries; consistent, need-to-know site communications |
| External vendors | Only the data or task instructions needed for their contracted function | Defined vendor permissions, approved transfer routes, and checks that files, dashboards, and support channels do not expose comparative outputs |
The analyst’s access does not make that person a decision-maker. Likewise, a sponsor’s responsibility for trial conduct does not automatically justify broad access to interim comparisons. The access plan should identify who produces the report, who reviews it, who has authority to decide, and who implements the decision. If one person has multiple roles, the plan should address how those roles are separated or controlled.
Confirm the scheduled review, the approved data cut, the planned outputs, and the people authorized to receive each output. Do not treat a convenient spreadsheet or dashboard as an authorized report merely because it contains the same data.
Restricted personnel verify the data version, treatment coding, outcome counts, and recipient list. The report should be limited to the prespecified purpose and released through the approved channel.
The authorized review body considers the result under its charter. Record the recommendation or decision and its rationale without copying sensitive comparative detail into broad operational correspondence.
Where permitted, send blinded sponsor functions, investigators, and vendors an actionable instruction that omits arm-specific results and unnecessary clues. Verify recipients and confirm the instruction matches the authorized decision.
Review access records and communications, and look for indirect signals such as unusual changes in site activity or vendor behavior. An audit trail helps establish what happened; it cannot show by itself what recipients inferred.
Worked examples: follow specific handoffs
Worked Example: From Analysis Output to Review Body
The restricted analyst has verified the hypothetical data cut: 80 participants per arm, 44 responses in Treatment A, and 35 in Treatment B. A review charter names an independent committee as the recipient of the arm-specific interim report. What should happen next?
Check the reported values. The proportions are \(44/80=0.550\) and \(35/80=0.4375\). Their difference is \(0.550-0.4375=0.1125\), or 11.25 percentage points. The report should label the denominator, outcome, time point, and treatment arms clearly enough for the authorized committee to interpret the numbers.
Check the handoff. The analyst sends the report through the approved restricted channel to the committee members named in the access plan. Before release, the analyst checks that no blinded operational distribution list, general sponsor mailbox, or site contact is included. The committee receives the comparison because its defined review role requires it; the calculation does not justify sending it to every person involved in trial conduct.
Check the interpretation. The result is a descriptive difference at this interim data cut. Without the planned inferential framework and the committee’s review, it is not a declaration that Treatment A is superior. The access decision follows from the charter, not from how large or appealing the observed difference appears.
Worked Example: Translating a Decision Into a Masked Instruction
Suppose, hypothetically, that the authorized committee recommends continuing recruitment without changing the current operational plan. The relevant sponsor authority confirms that this is the decision. A blinded operations lead must communicate next steps to sites. What information should the lead receive?
The lead needs the authorized instruction: recruitment and visit procedures continue under the current plan, with no change to site targets. The lead does not need the counts 44 and 35, the proportions 55.0% and 43.75%, or the 11.25-percentage-point difference. Sending those figures would provide comparative information without helping the lead implement this particular decision.
The communication should go only to the roles that need to act and should not imply an arm-specific reason. For example, it can state that the scheduled review is complete and current procedures remain in effect, if that statement is accurate and permitted by the communication plan. The decision record held in the restricted governance channel can retain the rationale and relevant review documentation.
This separation reduces unnecessary exposure, but it cannot ensure that every recipient draws no inference. If other operational changes occur around the same time, recipients may still speculate. The team should therefore keep timing and wording consistent with the communication plan and avoid suggestive explanations.
Worked Example: A Vendor Receives a Clue
A contracted data vendor supports trial dashboards. Its routine task requires checking that incoming outcome forms have been processed, not comparing outcomes by treatment. A sponsor employee asks the vendor to “prioritize the Treatment A dashboard because the interim response looks promising.” The vendor’s ordinary permission allows viewing treatment-coded outcomes. Trace the problem.
Information disclosed. The message reveals both comparative direction and a treatment-specific priority. Even if the vendor never sees the full interim report, the wording communicates a clue that is unnecessary for the stated task.
Control check. The vendor’s permitted data view should match its function. If treatment-specific outcomes are not needed, the access configuration should not expose them. The message should be handled through the trial’s established communication and exception process, not followed by informal discussion or additional treatment-specific requests.
Operational consequence. A priority change can also be visible to other vendor staff or later appear in service records. That may spread the clue beyond the original recipient. The appropriate response is to use the established breach and documentation procedures, restrict further access as authorized, and assess who received or could infer the information. A claim that “the vendor only saw a dashboard” does not resolve whether the dashboard disclosed comparative information.
Why the controls work—and where they do not
The justification for these controls is a simple information-flow argument. If a role’s task can be completed with an operational instruction \(I\), but an exact interim result \(R\) is not needed for that task, then giving the role \(R\) adds exposure without an operational benefit. Restricting the output to \(I\) limits the information available to that role. This does not prove that the recipient cannot infer anything about \(R\); it reduces direct disclosure and makes any remaining inference less certain.
A trace should therefore examine not just files but also channels and behavior:
- Access: Are permissions tied to current roles, and do they allow more than the task requires?
- Output: Do reports, dashboards, filenames, subject lines, or small-group summaries reveal arm-specific patterns?
- Communication: Are recipients, timing, and wording consistent with the approved plan, including in meetings and informal channels?
- Implementation: Could a change in recruitment targets, monitoring, site support, or vendor workload signal the comparative direction?
- Evidence: Do access records and decision records show who accessed or authorized information, and are they reviewed alongside operational signals?
A log can show that a user opened a report; it cannot establish whether the user understood an indirect clue, discussed it elsewhere, or changed behavior. Conversely, a change in site recruitment after an interim review is a signal to investigate, not proof that a particular person was informed. Interpretation requires context, timing, and comparison with the documented plan.
Alternatives, limitations, and common mistake
The main organizational alternatives are to let a defined sponsor decision group receive unblinded results or to route the interim review and recommendation through an independent body, with sponsor operations receiving only a masked decision. Neither structure is automatically correct for every trial. The governing design and charter should specify authority, and the chosen structure should avoid unnecessary overlap between access to comparative results and responsibility for routine blinded conduct.
A fully restricted approach can make implementation slower if the authorized decision pathway is unclear or if the masked instruction omits information an implementer genuinely needs. Broad access can make decisions easier to coordinate but creates more opportunities for direct or indirect leakage. The access plan must balance the specific decision need against the risk of wider exposure; a role label alone is not a sufficient justification.
Common mistake: “The sites never received the interim table, so the trial remained blinded.” This reasoning overlooks indirect disclosure through target changes, treatment-specific priorities, explanations, meeting comments, or vendor outputs. The correction is to trace both the exact result and downstream signals, identify what each recipient could see or infer, and compare actual communications with the authorized plan. A lack of direct report access is relevant evidence, not proof that no leakage occurred.
This trace is a practical review aid, not a substitute for the trial’s formal access plan, communication protocol, or response process. It also cannot determine from an observed imbalance alone whether leakage caused a change in trial conduct. Its value is in making roles, information, handoffs, and potential inference points explicit enough to check.
Check Your Understanding
Answer each question, then compare with the brief solutions.
- An interim report gives exact response counts by treatment. Which kinds of roles might legitimately need those values, and what should determine access?
- A blinded site team is told to increase recruitment to a particular target immediately after a review. Why might that instruction be a leakage signal even without treatment labels?
- A vendor can view arm-specific outcomes, although its task is only to check whether forms arrived. Identify the mismatch and one control to examine.
- Does an access log showing that only committee members opened the report prove that no information leaked? Explain.
- In the hypothetical example, calculate the difference in response proportions and state what it does—and does not—show.
Solutions. (1) The restricted analyst and authorized review body may need exact results; other decision-makers may receive them only if the documented governance structure authorizes that access. The task, charter, and access plan should determine recipients. (2) The target change may reveal that a review prompted an operational response, and its direction or timing could allow recipients to infer comparative information. (3) The vendor has broader access than its stated task requires; examine role-based permissions and restrict the view to the necessary data if authorized. (4) No. The log records access events, not all conversations, indirect clues, or inferences. (5) \(44/80-35/80=0.1125\), a descriptive difference of 11.25 percentage points in the hypothetical interim data. It does not by itself establish treatment efficacy or justify a decision.