Shared controls in 2026 master protocols: why 25% more participants is not 25% more power

TL;DR

FDA's June 2026 draft guidance on master protocols includes a useful allocation example: with 600 participants, four investigational drugs and one shared control, allocating 200 to control and 100 to each drug gives 300 participants in each drug–control comparison. Equal allocation gives 120 per group and 240 per comparison. The 25% increase concerns the raw pairwise count; statistical power and effective sample size require different calculations. Under the example's equal-variance assumptions, the variance of a difference in means falls by 10%. [1]

The larger operational lesson is that a shared control is an analysis-eligibility decision, not simply a common database label. The draft generally recommends controls who were concurrently eligible for the drug being evaluated and could have been randomized to it. Eligibility for each drug-specific comparison must be established for each potential control. The document is draft guidance marked not for implementation, not a new binding mandate. [1]

For sponsors and CROs, the defensible design starts with the comparison: who can be randomized, under which allocation and consent conditions, during which period, and which controls belong in the planned analysis. The randomization and trial-data systems should preserve those decisions as they existed, rather than reconstruct them from the platform's current arm list after the fact.

What the square-root rule actually optimizes

Appendix A of the FDA draft considers a deliberately simple case: N participants, k drugs, a common control, equal allocation among drug groups, equal outcome variances and the same assumed treatment effect for each drug. If p is the fraction allocated to control, each drug receives N(1 − p)/k participants. The variance of a drug–control mean difference is proportional to:

k / [N(1 − p)] + 1 / (Np).

Minimizing that expression yields p = 1 / (1 + √k), equivalent to √k:1 for control relative to each drug. This is an allocation result under specified assumptions and a fixed total sample size. Its application to a multiarm trial depends on those assumptions and the chosen design. The draft itself notes that platforms with drugs entering and leaving have more complex optimal-allocation considerations, including how evaluation periods overlap. [1]

The reason for allocating more participants to the shared control than to each individual drug is that the control contributes to several comparisons. But “more to control” requires a comparator. In a four-drug design, a one-third probability of control is greater than the one-fifth probability under equal five-group allocation, yet less than the one-half probability in a conventional two-arm 1:1 trial. Keep the draft's two-arm comparator separate from equal allocation across all five groups. [1]

This distinction matters in participant-facing explanations and design discussions. Under the square-root example, the probability of receiving any investigational drug is two-thirds. Under equal allocation across four drugs and control, it is four-fifths. A statement that the square-root scheme “increases the chance of active treatment” would therefore be wrong if the intended comparison is equal allocation within the same five-group design.

Recalculate the 600-participant example

For k = 4, √k = 2, so the ratio is 2:1:1:1:1. The counts sum to 600: 200 control participants and four groups of 100. Under equal allocation, all five groups contain 120. The following calculations reproduce the draft's counts and make the statistical consequence explicit. [1]

Counts follow FDA's hypothetical example. Variance and probability calculations are our arithmetic under that example's assumptions, not empirical trial results. [1]

The variance ratio is 0.015 / 0.016667 = 0.90. The standard-error ratio is √0.90, approximately 0.9487, a reduction of about 5.13%. If information is expressed as inverse variance for this comparison, it increases by approximately 11.11%, because 1/0.90 − 1 = 0.1111. None of these quantities equals the 25% increase in raw pairwise headcount.

Power is a further quantity. Its numerical value depends on the assumed effect, outcome variability, testing threshold and analysis design. The example explains a potential power improvement under those assumptions, without specifying a universal percentage increase in power. A protocol's power claim should come from its actual calculation or simulation, including the relevant multiplicity strategy and design features, not from dividing 300 by 240.

This also explains why “effective sample size” is an unsafe label for the 300. The imbalance between 100 drug participants and 200 controls matters to the comparison's variance. Use the information calculation to account for that imbalance. Keeping the count and variance rows separate makes the argument auditable without overstating the efficiency gain.

A shared control must be eligible for the comparison

The draft generally recommends that a primary comparison use only concurrently eligible controls: participants who met the drug-specific criteria, could have been randomized to that drug, and were instead randomized concurrently to control. It distinguishes nonconcurrent controls from concurrent participants ineligible for randomization to the drug. [1]

Calendar overlap alone is therefore insufficient. A drug may have entered the platform before it became available at a particular site. A participant may have met the master protocol's general criteria but not that drug's kidney-function criterion. A participant may have consented only to a subset of drugs. Each case requires participant- and site-specific evidence of eligibility, beyond calendar overlap with a drug open somewhere in the platform.

The design needs a reconstructable eligible treatment set for the relevant randomization. That set can depend on protocol version, substudy criteria, site availability and consent, among other design-specific conditions. Our operational recommendation is to preserve the applicable inputs and approved derivation at assignment, with a traceable correction process. The record structure is an implementation proposal; FDA's recommendation concerns the underlying comparison.

A hypothetical example makes the distinction concrete. Drug B opens globally on a given date and at Site 7 only later. A Site 7 control participant enrolled between those dates overlaps the global program in calendar time but remains ineligible for assignment to B at that site. The draft's preferred concurrently eligible approach therefore excludes that record from B's primary shared-control comparison. The planned analysis might propose a justified alternative, but it must recognize the difference rather than hide it in a pooled-control flag.

The efficiency argument survives this restriction: the draft notes that shared controls can still yield considerable efficiency relative to separate trials even when comparisons use only concurrently eligible controls. Ask which controls preserve the intended randomized comparison. [1]

Nonconcurrent controls require an argument about bias, not only precision

The draft also leaves room for scientifically justified alternative control data. It discusses rare circumstances where a scientifically justified alternative may be appropriate, including settings with different bias–variance tradeoffs such as exploratory trials or rare diseases with feasibility constraints. For a platform intended to contribute to substantial evidence of effectiveness, it recommends early discussion with FDA and a decision agreed before the trial starts. [1]

A larger control dataset can improve precision while also introducing bias. The relevant threats include changes over time in participant characteristics, trial conduct and standard of care. Recovering an exchangeable randomized comparison requires more than model adjustment for observed variables, especially where changes remain unmeasured or model assumptions fail.

The draft asks for a rationale covering feasibility of concurrent-only data, likely temporal changes, the amount and timing of nonconcurrent information, and the statistical methods and assumptions. It also discusses evaluating operating characteristics under different scenarios and conducting sensitivity analyses, including analyses using only concurrent controls or giving them more weight. [1]

For an implementation team, this means the dataset should support the prespecified alternatives rather than collapse timing and eligibility history into a single final cohort label. If the analysis plan needs a concurrent-only sensitivity analysis, those records must be identifiable. If weighting depends on period or prognostic information, the required inputs must exist and be interpretable. Technology makes the justification for historical borrowing and its consequences available for review; the scientific decision belongs to the responsible team.

Selecting nonconcurrent controls after observing an unfavorable concurrent-control result is a different problem from a prospectively specified borrowing strategy. The draft explicitly warns against decisions made after seeing results. [1] A versioned analysis plan and preserved eligibility history support that assessment alongside statistical and regulatory judgment.

Arm changes are protocol events, not automatic recalculation triggers

The draft recognizes that a platform may use randomization ratios that change when drugs enter or leave, or may maintain a constant probability of control. It says comparisons should account for periods with different drug-to-control allocation ratios. [1] Implement those alternatives through prespecified design decisions; an operational status change alone provides no authority for an RTSM to recompute √k.

An arm can be unavailable for different reasons: planned closure, temporary supply interruption, a site not yet activated, participant-specific ineligibility or a safety restriction. Specify the intended randomization consequence for each situation. The approved specifications need to define which status affects assignment, which participants and sites it affects, and what should happen when the status changes.

The useful test is historical reproducibility. Given the information available at a participant's assignment, can the team explain which arms were possible and which allocation rule applied? Preserve the historical decision context separately from today's open-arm list. Our recommendation is to preserve effective periods and configuration versions, with controlled access to unblinded information and clear correction handling.

This is an editorial design-to-system handoff matrix, not a required FDA data schema.

Consent can change the comparability of the shared control

The draft recommends that consent cover all treatments to which a participant could be randomized and, in the shared-control setting, occur before randomization while avoiding drug-specific substudy consent after assignment. It explains that post-randomization drug-specific consent can select different participant populations across substudies and compromise the comparability of the pooled controls. It also discusses the need for strong justification and bias-mitigation steps if that approach is proposed. [1]

This is more than a document-timing check. If participants who would accept one substudy differ systematically from those who would accept another, pooling their controls can change the comparison. The consent process and eligible-treatment set therefore belong in the statistical design discussion, not only in site startup administration.

Where consent covers only a subset of drugs, the draft limits potential assignments to that consented subset. [1] A system test should include the restricted set and verify the approved allocation behavior. Record drug-specific consent coverage beyond a generic “consent complete” flag.

The software's role is to implement the approved process and preserve its evidence. Verify the applicable consent version and substudy explicitly, separately from screening attendance or earlier consent. The draft presents a recommended consent design; any proposed alternative must satisfy applicable informed-consent requirements. [1]

Distinguish the two 30-day concepts

The June 2026 draft recommends submitting amendments that substantively affect a master protocol's safety, quality or scope at least 30 days before initiating the changes. It separately recommends notifying the regulatory project manager at least 48 hours before submitting such an amendment. These are recommendations in the draft, including a footnote recognizing the apparent-immediate-hazard exception. [1]

The underlying IND rules must be described separately. Under 21 CFR 312.40(b), an IND goes into effect 30 days after FDA receives it unless a clinical hold applies, or earlier on FDA notification that the investigations may begin. That initial IND provision is not a blanket rule that every protocol amendment restarts a statutory 30-day waiting period. [2]

For covered protocol changes, 21 CFR 312.30(b) provides for implementation after submission to FDA and approval by the responsible IRB, in either order. Its scope distinguishes phase 1 changes significantly affecting safety from phase 2 or 3 changes significantly affecting safety, scope or scientific quality. It allows immediate implementation of a change intended to eliminate an apparent immediate hazard, with subsequent FDA and IRB notification as specified. [3]

A release plan should therefore identify which regulatory provision, draft recommendation and study-specific commitment controls each milestone. Label each milestone accurately as submission, a recommended review interval or an actual approval, as applicable. IRB approval, consent, site activation and other applicable conditions require separate verification from the software deployment date.

What to verify before choosing the operational platform

Ask for a demonstration using the proposed design: a participant with a restricted drug-eligibility set, a site whose arm availability differs from the global platform, an allocation change and a control record whose inclusion differs between two drug comparisons. Review what the system records and what the downstream analysis receives. A claim that a product “supports platform trials” is too broad to establish these details.

EClinCloud's RTSM, EDC and study services provide a relevant starting point for an implementation discussion. Scope and validate the required capabilities and interfaces against the actual protocol, including any proposed square-root allocation implementation; compliance remains a separate assessment.

Shared controls are useful when the design and operations preserve the intended comparison. In FDA's simple 600-participant example, 25% more people in a pairwise count corresponds to a 10% variance reduction under stated assumptions—not a universal power gain. In a changing platform, the harder question is which participants legitimately contribute to that comparison. Get that question right before treating a larger pooled dataset as stronger evidence.

Sources

1. FDA. Master Protocols for Drug and Biological Product Development. June 2026 revised draft, not for implementation. Section III.A–C, printed pages 6–11; section V.C, printed pages 22–23; Appendix A, printed pages 26–27. Allocation assumptions, concurrent eligibility, alternative controls, consent and amendment recommendations. Arithmetic in this article is an independent derivation from the appendix's hypothetical example.

2. Electronic Code of Federal Regulations. 21 CFR 312.40—General requirements for use of an investigational new drug in a clinical investigation. Paragraph (b), initial IND effectiveness; accessed October 11, 2026.

3. Electronic Code of Federal Regulations. 21 CFR 312.30—Protocol amendments. Paragraph (b), covered changes, FDA submission/IRB approval and apparent-immediate-hazard exception; accessed October 11, 2026.