Why Shared-Control Policies Need Explicit Rules
In a platform trial, more than one treatment arm may use participants assigned to the same control group. This can reduce the number of control participants needed across comparisons, but it also creates a governance dependency: changes affecting the control group can affect several arms at once. A policy must therefore say not only that sharing is allowed, but which arms are eligible, for what period and population, and what happens when conditions change.
A shared-control policy is not a substitute for a statistical analysis plan or for review of the design’s operating characteristics. Governance establishes who may authorize shared use and the conditions that must be maintained. The statistical plan must separately address how comparisons using common control observations will be analyzed and how their dependence will be handled. Treating the same control participants as independent observations in multiple comparisons would be inappropriate.
The policy should be specific enough that the platform can determine, before enrolling participants under the arrangement, whether a proposed pair of arms qualifies. It should also identify a responsible authority, consistent with the platform’s governance charter and decision-rights map, to approve the arrangement and resolve uncertainty. A general statement such as “arms may share controls when appropriate” leaves too much discretion to be reliable.
Conditions for Sharing
Before approving shared use, the platform should assess whether the arms and their control assignments are sufficiently aligned for the planned comparisons. The assessment should address at least the following conditions:
- Overlapping enrollment and control availability. Control participants must be randomized during the period in which the relevant treatment arm is enrolling. A control group from an earlier, non-overlapping period is historical evidence, not a concurrent shared control.
- Compatible target populations. The eligibility criteria and intended participant population should be aligned. If an arm has a materially different eligibility rule, the shared control may not represent the population relevant to that arm.
- Comparable control intervention and care. The control treatment, background care, outcome assessment, and follow-up procedures should be sufficiently consistent for the comparisons. Changes in standard care or control treatment can undermine that consistency.
- Compatible allocation and implementation. The assignment process, participating sites, recruitment practices, and relevant calendar periods should be considered. A common control label alone does not make assignments comparable if operational conditions differ in important ways.
- Clear data and analysis boundaries. The protocol and analysis documentation should identify which control participants contribute to which comparisons, how changes are handled, and how dependence among comparisons is recognized.
These are governance screening criteria, not a guarantee that the statistical assumptions of a particular analysis are satisfied. The design team should assess those assumptions using the planned analysis and design assurance work. If the evidence does not support compatibility, the appropriate response may be separate control groups, a narrower sharing period or population, or no sharing.
Worked Example: Approving Sharing for Two Compatible Arms
Hypothetical platform. Arms A and B are proposed for a respiratory condition. Both use the same eligibility criteria, the same control treatment and background-care guidance, the same primary outcome and follow-up schedule, and the same 18 sites. Their enrollment periods overlap for six months. The platform randomizes participants among the active arms open at their site and time, including the common control.
Apply the policy. First, verify that control assignment is available during the enrollment of both arms: it is, for the six-month overlap. Next, compare the populations and procedures: eligibility, control care, outcome, and follow-up match. Then check implementation: the sites and assignment procedures are common. Finally, verify that the documentation maps the common control participants to both comparisons and that the analysis plan accounts for the shared data.
Decision. The governance authority may approve shared concurrent control use for Arms A and B during the specified six-month period, subject to continued monitoring of the stated conditions. The approval should name the arms, control version, eligible population, sites, start and end conditions, and accountable decision-maker. It should not be interpreted as permission to share control data for other arms or later periods without a new compatibility assessment.
When Sharing Must Stop or Be Reconsidered
A stopping rule should identify observable events, not rely only on a broad instruction to stop if sharing becomes “inappropriate.” The policy can distinguish an immediate stop from a prompt review. An immediate stop is warranted when an essential condition for valid concurrent use no longer holds—for example, when the control intervention changes materially or an arm’s eligibility criteria diverge. A review trigger may be appropriate for a potentially consequential change whose effect is not yet clear, such as a substantial shift in site participation or recruitment timing.
The rule must also define what “stop” means operationally. It may mean that new participants are no longer assigned to a common control for the affected arms from a specified effective time, while existing participants continue under their assigned treatment and follow-up. The policy should state how the effective time is recorded, who communicates the decision to sites, and how data accrued before and after the change are distinguished. Stopping prospective sharing does not automatically invalidate observations already collected; their use must follow the prespecified analysis approach and applicable oversight decisions.
Record the event, when it began, which arms and control assignments it affects, and whether it changes population, treatment, follow-up, sites, or timing.
If an essential compatibility condition is no longer met, suspend new shared assignments for the affected arms as the policy specifies, without making unapproved changes to participants already enrolled.
The designated authority determines whether sharing can continue, must be narrowed, or must end, consulting the relevant scientific, operational, and independent oversight roles as defined in the charter.
Document the rationale, affected comparisons, decision, effective time, responsible approvers, and implementation instructions; communicate them to sites and relevant platform teams.
Worked Example: Ending Sharing After a Control-Care Change
Hypothetical platform. Arms C and D have shared control participants at 12 sites. At the start of a new calendar period, the platform adopts a materially different control-care guideline. The change applies to new participants at all sites, while participants already enrolled continue their assigned care. The shared-control policy identifies a material control-care change as an immediate review trigger.
Apply the stopping rule. The control care received by new participants after the effective date differs from the care received by earlier participants. Therefore, treating all control observations as if they came from one unchanged concurrent-control condition is not justified by the existing approval. The responsible authority records the effective date and identifies Arms C and D as affected.
Decision and implementation. Unless a documented assessment supports continued sharing under a revised, approved arrangement, new shared assignments for C and D stop at the effective date. Existing participants remain under their assigned care. The data record distinguishes the earlier and later control-care periods, and the analysis team is informed that the comparison structure has changed. No conclusion about treatment efficacy follows from this governance decision alone.
Exceptions and Their Approval
Exceptions are controlled departures from the default policy, not informal permissions to ignore compatibility concerns. The policy should name who may request an exception, who evaluates it, who has authority to approve or reject it, and whether any independent review is required. Those roles should be consistent with the platform’s existing delegation and oversight arrangements. A person with a direct interest in an affected treatment arm should not be the sole approver where the governance framework requires independent review.
An exception request should specify the arms and participants affected; the proposed scope and duration; the reason the default rule cannot be followed; the potential consequences for comparability, participant protection, operations, and analysis; alternatives considered; and any conditions for ending the exception. Approval should be prospective whenever feasible. Urgent action needed to protect participants may require immediate action under safety procedures, but such action does not itself authorize ongoing shared-control use; the relevant authority must review and document the arrangement promptly.
Worked Example: Reviewing a Limited Exception Request
Hypothetical platform. Arm E is open at a small group of sites. Arm F is proposed at those sites, but its follow-up visit is scheduled at a different time from Arm E’s. The sponsor requests permission for Arm F to use the existing shared control group, arguing that this would reduce site burden. The platform’s default policy requires aligned outcome-assessment timing.
Assess the request. The request names the arms and sites, but the different follow-up timing is a substantive comparability concern, not merely an administrative inconvenience. The reviewer should ask whether the primary outcome is measured on the same schedule, whether control participants would be assessed in a way that supports both comparisons, and whether any proposed adjustment is prespecified and justified. The reviewer should also consider alternatives: align follow-up before activation, create a separate control arrangement, or decline the proposed sharing.
Decision. If the timing difference prevents comparable outcome assessment, the exception should be rejected; operational convenience is not sufficient justification. If a documented assessment establishes that the difference is immaterial for the planned comparisons and the statistical plan addresses the arrangement, an authorized body could approve a narrowly scoped, time-limited exception with monitoring and a stated end condition. The decision, rationale, and conditions must be recorded before the exception is implemented.
Justifying a Clear Rule
A useful policy can be understood as a sequence of necessary checks. Let a proposed shared-control arrangement be eligible only if it satisfies every essential compatibility condition, has an authorized approval, and remains within its approved scope. If any essential condition fails, the arrangement is not eligible for continued sharing under that approval. This rule follows from the purpose of the policy: shared control is authorized for a defined comparison context, so a material change to that context removes the basis for relying on the authorization.
The rule is deliberately conservative: it does not say that any operational difference automatically makes sharing invalid, nor does it claim that satisfying the checks proves statistical validity. Instead, it requires the responsible authority to evaluate a potentially consequential change before assuming the original approval still applies. A recorded decision provides an auditable link between the evidence considered and the scope of use authorized.
Diagnostics, Alternatives, and Limitations
Governance should include routine checks that the approved arrangement still describes actual practice. Useful indicators include whether arms continue to recruit during the approved period, whether sites follow the same control procedures, whether eligibility or follow-up has changed, and whether the data system can identify control participants by period and relevant protocol version. Unexpected gaps, protocol deviations, uneven site participation, or changes in background care should prompt assessment rather than being dismissed as administrative details.
The main alternative is to use separate concurrent control groups for different arms. This can reduce dependence on one common control condition and may be preferable when populations, treatment versions, sites, or timelines differ. It can require more control participants and may increase operational burden. Another alternative—using nonconcurrent or historical controls—is not equivalent to sharing a concurrent control: calendar-time changes can make earlier control outcomes less comparable. Each choice requires design and analysis justification, not a governance label alone.
Shared-control policies also have limits. They cannot eliminate confounding from changing care or recruitment conditions, guarantee adequate information for each comparison, or replace a prespecified plan for multiplicity and dependence. A governance approval is not a finding of efficacy, safety, or statistical validity. It authorizes a defined operational arrangement subject to the design assumptions and oversight obligations that apply to the platform.
Common mistake: “The arms are in the same platform, so they can always share the same control group.” Platform membership alone establishes none of the necessary compatibility conditions. The correction is to assess the proposed arms, population, control care, timing, sites, and analysis boundaries; approve only a defined scope; and reassess when a stopping or review trigger occurs.
Check Your Understanding
Answer each question, then compare your reasoning with the brief solutions.
- Two arms use the same control treatment, but one arm enrolls only after the other has stopped enrolling. Is their control use concurrent? Why?
- Name two changes that should trigger immediate suspension or prompt review under a well-specified policy.
- Does stopping prospective sharing mean that previously collected control data must be discarded? Explain.
- A proposed exception would ease site workload but creates different outcome-assessment schedules. What should the approver establish before deciding?
Solutions. (1) No. Without overlapping control assignment and enrollment, the earlier control observations are not concurrent controls for the later arm. (2) Examples include a material change in control care or a divergence in eligibility; the policy should distinguish immediate stops from review triggers. (3) No. The policy should stop or change new assignments as specified, while prior data are handled according to the prespecified analysis and applicable oversight decisions. (4) The approver should assess whether the timing difference affects comparability and the planned analysis, consider alternatives, and approve only a justified, bounded exception with documented conditions—or reject it if the concern cannot be resolved.