Why Site Access Requires a Separate Decision
Investigators and site staff need information to run a trial well: they may need to know whether recruitment is on schedule, whether data are being collected completely, and whether a protocol procedure needs clarification. But access to interim treatment comparisons can change how staff and participants behave while outcomes are still being collected. The central question is therefore not whether information is useful in the abstract. It is whether a particular summary is needed by a site role, and whether sharing it could disclose treatment differences or influence the trial.
In an adaptive trial, interim information can also contribute to decisions about changing the trial. The people responsible for a site’s routine conduct do not ordinarily need comparative results to carry out their duties. A sound access plan separates information needed for those duties from information that could affect recruitment, treatment delivery, outcome assessment, or an adaptation decision.
These labels describe what a report contains, not whether it is automatically safe to share. A pooled report can still reveal treatment information if group assignments are known locally, the report is very small, or other information allows a treatment-specific result to be inferred. Conversely, some comparative information may need to be disclosed under a protocol or applicable safety process. The intended audience, context, and purpose all matter.
Information That Can Support Site Work
A site may have a legitimate need for operational summaries that do not compare treatment outcomes. Examples include overall recruitment against a target, the number of participants with overdue visits, the completeness of required forms, and pooled measures of data variability when those measures are relevant to trial operations. Such reports should be limited to the detail needed for the task. A site does not need outcome results merely because it needs to know whether its data are complete.
Pooled summaries can also be appropriate when a planned adaptation uses a nuisance parameter estimated without treatment labels. For instance, the earlier material on blinded variance re-estimation distinguishes a pooled variance estimate from an unblinded treatment-effect estimate. A site may be told a result only if it has a defined operational reason to know it, and if the report cannot reasonably be used to reconstruct a treatment comparison. The fact that a statistic is pooled is helpful, but it is not by itself a guarantee of masking.
Information about safety requires particular care. Sites must receive information needed for participant care and for the safety responsibilities established by the trial’s procedures. That does not mean routine site distribution of comparative efficacy summaries is necessary. Safety communication should follow the appropriate protocol and oversight processes; it should not be treated as a reason to circulate unrelated interim treatment comparisons.
Worked Example: A Pooled Event Summary
Suppose a hypothetical trial has enrolled 120 participants across its sites. At an interim data cut, 18 participants have experienced a specified event. A site asks whether the pooled event frequency can be reported without revealing treatment results.
The pooled proportion is calculated using all participants in the denominator:
The summary says that 15.0% of participants experienced the event by that data cut. It does not, on its face, state how the events divide between treatment groups. For example, the same total of 18 events could arise from 9 events in each of two groups of 60, or from 4 and 14 events. The pooled count alone cannot identify which split occurred.
Before sharing it, the trial team should still check whether site staff know the treatment assignments for some participants, whether the number of participants is so small that the summary is revealing, and whether other available reports could be combined with it to recover a comparison. If these checks are satisfactory and sites have a defined need for the pooled figure, it may be suitable for sharing. It is not suitable merely because it is labelled “pooled.”
Why Comparative Results Are Ordinarily Withheld
A comparative interim result can influence trial conduct even when no formal adaptation decision is made at the site. Investigators who learn that one treatment appears more effective may unintentionally recruit or retain different kinds of participants, communicate expectations differently, or make different judgments about symptoms and outcomes. Staff may also change how closely they monitor particular participants. Such changes can affect the data being collected and can make the interim comparison less representative of the planned trial.
Comparative information can also compromise masking. If a site learns which treatment is performing better, staff or participants may try to infer individual assignments from the overall result. Beliefs about assignment can affect reporting, care, follow-up, and assessment. Even when the comparison is uncertain, an interim numerical difference can shape expectations. Withholding the result is not a claim that the observed difference is true; it is a way to reduce avoidable influence while the trial continues.
There is a further governance reason: adaptation decisions should be made through the trial’s specified process by roles authorized to use the relevant information. Broad site circulation can create informal pressure to change enrollment or conduct before a decision is made through that process. The information tiers established for trial roles and the earlier material on sponsor and monitoring access provide the general principle: access should be limited to what each role needs, with comparative results confined to appropriately authorized functions.
Worked Example: A Site Requests an Interim Comparison
Consider a hypothetical two-group trial with 60 participants in each group. At a planned interim data cut, 4 of 60 participants in Group A and 14 of 60 in Group B have experienced an outcome. A site investigator asks to see the group-specific counts to decide whether to focus recruitment on a particular patient profile.
The observed proportions are:
The observed risk difference, defined here as Group A minus Group B, is:
In this hypothetical sample, the observed event proportion is 16.7 percentage points lower in Group A than in Group B. This is a descriptive interim difference, not proof of a treatment effect. It is nevertheless comparative information: revealing it could influence recruitment decisions and staff expectations. The investigator’s proposed use—changing recruitment according to an interim pattern—is precisely why routine site access is ordinarily withheld. Recruitment criteria should follow the protocol, not an unplanned response to a site’s view of comparative interim outcomes.
If a result is needed for a formal trial decision, it should be handled through the authorized decision process rather than circulated to sites for local interpretation. The site can instead receive relevant operational guidance, such as a reminder about eligibility criteria or information about recruitment progress, without being given the treatment contrast.
Justification: Why Pooling Does Not Give the Contrast
Pooling removes the treatment labels from a summary, so a pooled total generally cannot determine the separate group results. This is a direct consequence of adding the group counts. Let \(e_A\) and \(e_B\) be the event counts, and \(n_A\) and \(n_B\) the group sizes. The pooled event proportion is a weighted average of the two group proportions:
The pooled value gives one equation involving two group proportions. Unless additional information is supplied, it does not determine their difference. With equal group sizes of 60 and 18 total events, for example, both a split of 9 and 9 and a split of 4 and 14 produce the same pooled proportion, \(18/120=0.150\). Their risk differences are not the same: the first is zero, while the second is \(-1/6\). Thus, a pooled count alone does not disclose the contrast.
This argument has limits. If someone receiving the pooled total also knows one group’s event count, they can subtract it from the total to recover the other group’s count. Small denominators, sequential reports, or outside knowledge can create similar opportunities. The relevant question is not only whether each report is pooled in isolation, but whether the full set of information available to the recipient permits a treatment-specific inference.
Worked Example: Deciding Whether an Operational Report Is Appropriate
In a hypothetical trial, a site is expected to enroll 40 participants by a specified date. It has enrolled 36. The data team also reports that 3 of those participants have at least one overdue required form. The site requests a report by treatment group to determine whether the overdue forms are concentrated in one group.
The recruitment progress is:
The site is at 90.0% of its enrollment target. The overall proportion with an overdue form is:
Thus, 8.33% of the enrolled participants have at least one overdue form, based on the information given. These overall figures could help the site address recruitment and data-completion tasks. The requested treatment-group breakdown is different: it could reveal a group-specific pattern and is not necessary to identify or resolve the overdue forms. A suitable response is to provide the overall operational measures, identify the records needing follow-up through the authorized workflow, and withhold the treatment-stratified summary unless an approved role-specific need justifies its disclosure.
The denominator and the time point should accompany any such report. Without them, a percentage can be difficult to interpret; for example, “8.33% overdue” does not indicate whether it is 3 of 36 participants or a different count and population.
Making the Access Decision and Recognizing Failure Modes
A practical site-access decision starts with the task the recipient must perform. The report should answer that task with the least revealing useful information. A recruitment problem may need an enrollment count, not an efficacy comparison. A data-quality problem may need a list of incomplete records, not a treatment-stratified outcome table. An adaptation decision may require comparative information, but that is a reason to route the information to the authorized decision function, not to distribute it broadly.
State what action the investigator or staff member must take, such as correcting missing data or addressing recruitment progress.
Prefer operational or pooled information when it is sufficient for the stated task. Avoid treatment labels and contrasts unless they are needed and authorized.
Consider group sizes, prior reports, treatment knowledge at the site, and whether combining information could reveal a comparison.
Use established safety and adaptation processes for information that must be disclosed or used by authorized roles; do not substitute informal site circulation.
Warning signs include a report that unexpectedly presents results by treatment group, a pooled statistic with very small denominators, repeated reports that allow changes to be calculated, or a request to share results so sites can alter enrollment or clinical behavior. Another failure mode is treating “not statistically significant” as equivalent to “not revealing.” A numerically small or uncertain comparison can still shape expectations and conduct.
The main alternative to restricting access is broad disclosure of interim comparisons. That can be appropriate in some trial designs or when disclosure is required, but it should be an intentional feature of the design and procedures. For routine site operations in a trial that depends on maintaining masking, the usual approach is narrower: provide the operational information sites need, while withholding comparative efficacy results unless a defined, authorized purpose requires otherwise.
No single rule makes every pooled summary safe or every comparative result prohibited. The judgment depends on the trial’s design, the information already known to recipients, the purpose of disclosure, and the procedures governing safety and adaptation. Access decisions should be made before a particular interim result creates pressure to share it, and the rationale should be clear enough that staff can apply it consistently.
Check Your Understanding
Answer each question, then compare your reasoning with the brief solutions.
- A site receives a pooled count of 12 events among 100 participants. Does that count by itself identify the difference between treatment groups? Explain.
- A site asks for a treatment-specific efficacy table so it can recruit more participants who seem likely to benefit. What is the main access concern?
- A site’s overall enrollment is 45 of a target of 50. What proportion of its target has been reached, and why might this be a suitable operational summary?
- Why can a pooled report still disclose treatment information in some circumstances?
- Does withholding comparative efficacy results mean sites should be denied information needed for participant safety? Explain.
Solutions. (1) No. The total of 12 events does not specify how they are divided between groups, so it does not by itself identify the treatment contrast. (2) The requested comparison could influence recruitment and trial conduct; sites should ordinarily follow protocol eligibility criteria rather than adapt local recruitment in response to interim efficacy patterns. (3) \(45/50=0.900\), or 90.0% of the target. This can support recruitment planning without presenting a treatment comparison. (4) Other available information, such as a known group count, small denominators, or successive reports, may let a recipient infer the group-specific results. (5) No. Safety information needed for participant care should be communicated through the applicable trial procedures; that is distinct from routine disclosure of comparative efficacy results.