Workshop: Resolving a Dispute Over Shared Controls

What Is Being Disputed?

In a platform trial, an existing control group may be proposed as the comparator for a new experimental arm. Sharing can reduce the number of participants assigned to control, but the proposal raises a distinct governance question: which control participants, if any, are eligible to contribute to this particular comparison? Sponsors may disagree because they have different interests in timelines, costs, confidentiality, or access to data. Those interests matter to governance, but they do not replace the criteria needed to decide whether sharing is permitted and scientifically appropriate.

The key distinction is between a participant’s eligibility to enroll in the platform and the eligibility of that participant’s control data for a particular arm comparison. A person may have been correctly randomized to control and remain in the trial, while their data are not eligible for a proposed comparison because, for example, the person enrolled outside the relevant calendar period or did not provide the required permission for the proposed use.

A defensible decision begins with criteria agreed before the dispute, rather than criteria selected to favor one sponsor’s preferred answer. The agreement should identify who decides, which documents govern, how conflicts or recusals are handled, and where the rationale is recorded. The earlier material on multi-sponsor platform agreements, conflicts of interest in platform decisions, and decision records and traceable governance provides the relevant governance context.

Use Decision Gates, Not a Sponsor Scorecard

A practical decision framework separates non-negotiable gates from considerations that may be managed through conditions. A strong operational case cannot compensate for missing permission to use the data. Likewise, a sponsor’s willingness to pay for additional analyses does not make nonconcurrent controls scientifically interchangeable with concurrent controls.

These gates have to be defined for the particular comparison. For example, concurrent enrollment is usually important when the design relies on contemporaneous controls to address changes over calendar time. If a design explicitly allows nonconcurrent controls, that use needs its own pre-specified scientific justification and governance authorization; a sponsor dispute is not a reason to silently relax the design. Similarly, a shared control is not automatically suitable just because both arms use the same nominal endpoint name. The measurement schedule, definition, ascertainment, and follow-up must support the intended comparison.

The criteria should be applied in a stated order. First check authority and permissions: if data use is not authorized, do not treat the data as eligible while other issues are debated. Next assess whether the participant and data fit the comparison’s target population, time period, and design. Then consider whether any unresolved operational differences can be addressed without changing the estimand or undermining the design. Record which gate failed, rather than merely labeling a set “ineligible.”

A Compact Justification for the Rule

The gate structure follows from what eligibility means. Let \(E_i\) indicate whether control participant \(i\) is eligible for a specified comparison, and let \(G_{ij}\) indicate whether that participant satisfies required gate \(j\). If all \(J\) gates are mandatory, eligibility is their conjunction:

If any required gate is zero, the product is zero. This is not a statistical weighting scheme: it expresses that a mandatory condition cannot be traded against another. For example, excellent endpoint completeness does not remedy a lack of authorization for the proposed use. If the governing documents instead allow a condition to be addressed—for instance, by obtaining an approved data transfer or completing a defined validation—the record should say that eligibility is conditional and specify what must happen before the data are used.

This rule is only as sound as its gates. The platform protocol, statistical design, consent materials, sponsor agreements, and applicable approvals may impose additional requirements. Governance decision-makers should not create a new analysis population or revise the protocol through an informal meeting. If the proposed resolution would materially change the design, the relevant protocol, statistical, ethics, or regulatory process must be followed before implementation.

Partial Overlap and Conditions

A disagreement is not always an all-or-nothing choice between sharing every control and sharing none. If a new arm targets a narrower population than the platform’s overall population, a prespecified subset may be appropriate. But selecting that subset must preserve the intended comparison: the target population and subgroup definition should be clear, eligibility should be determined using appropriate information, and the same rule should apply regardless of which sponsor benefits.

Conditions can address remediable issues, but they should be explicit and completed before the affected data are used. Examples include verifying that a data field was collected consistently, completing an authorized data transfer, or obtaining required approval for a proposed use. A condition is not a way to conceal a failed gate. If a condition cannot be satisfied, the record should state that the relevant data are ineligible for this comparison.

When Sharing Is Not Defensible

If a required condition cannot be met, the available options include using a separate concurrent control group, revising the proposal through the appropriate design and approval processes, or declining or deferring the comparison. These alternatives are not equivalent. A separate control group may address incompatibility but changes operational demands. Deferral may allow clarification or approval, but should not create pressure to use data before authorization. A historical or nonconcurrent control should not be substituted merely because it is available; it requires explicit design support and cannot be made concurrent by agreement between sponsors.

Document and Communicate the Decision

A decision is not complete when the meeting ends. A useful record makes it possible to understand what was asked, which criteria controlled the decision, how evidence was assessed, and what actions follow. It should distinguish a final determination from a conditional one and identify any dissent or escalation. If participants were recused, the record should reflect that under the platform’s governance procedures without disclosing more conflict information than appropriate.

The communication should be concise enough to act on and complete enough to prevent different sponsors from receiving different accounts of the decision. If the decision is conditional, identify exactly what remains unresolved and whether the data may be used before that condition is met. If the matter is escalated, state who will decide it and what interim rule applies; absent authorization, do not treat disputed data as available for the comparison.

Common Pitfall and Alternatives

A common mistake is to treat shared controls as a bargaining benefit: “Both sponsors gain efficiency, so all existing control data should be shared.” The reasoning is wrong because mutual benefit does not establish that participants match the target population, that the data support the comparison, or that the proposed use is authorized. The correction is to apply the pre-agreed gates and then consider operational benefits among the options that remain permissible.

The main alternatives can be compared as follows:

The framework has limits. Governance criteria cannot establish statistical validity by themselves, and a scientifically plausible comparison cannot override consent, agreements, or applicable approvals. Nor does a written decision eliminate the need to carry out the conditions it specifies. The role of this process is to make the decision traceable, consistent, and appropriately bounded—not to guarantee a particular scientific result.