Why Shared Controls Do Not Always Save Participants

When the Arithmetic Saving Is Not the Practical Saving

A common control group avoids recruiting a new control group for every treatment comparison. That advantage is real only when the same control participants can validly contribute to each comparison and when the resulting allocation meets the design requirements. A shared control can therefore save few participants, provide no usable saving, or fail to satisfy a planned allocation constraint.

The earlier material on shared-control sample sizes establishes the basic recruitment comparison. Here the focus is on the conditions behind that arithmetic: unequal requirements across comparisons, whether the control outcomes are genuinely reusable, and whether allocation rules permit one control group to serve all arms.

Definition: For comparison \(j\), let \(n_{Cj}\) be the required number of control participants and \(n_{Tj}\) the required number assigned to treatment \(j\). In separate two-arm trials, these are distinct groups for each comparison. In an ideal shared-control design, \(n_C\) denotes the number of unique control participants used in all comparisons. A participant is reusable across comparisons only when their control data are valid for each relevant comparison.

Assume the stated group sizes are the required numbers of evaluable participants for the planned analyses. The requirements must already reflect the chosen outcome, target effect, variability, error rate, power and other planning assumptions. If recruitment must allow for loss to follow-up, the number to recruit will be larger; the same retention assumptions should be applied consistently to both designs. For a continuous outcome, a usual planning model might assume independent participants within and between randomized groups and a common variance for the groups being compared. Other outcomes require their own suitable models. The sample-size comparison below does not establish those assumptions; it takes the required group sizes as inputs.

The ideal shared-control calculation also assumes that the control population, standard care, outcome definition and measurement schedule are compatible across comparisons, and that the same control observations can be used without violating the design. When those assumptions hold and sample sizes are simple minimum requirements, the smallest common control group must satisfy every comparison.

The Ideal Saving With Unequal Requirements

With separate trials, control participants are recruited anew for every comparison. With one valid common control group, that group must be at least as large as the largest individual control requirement. Treatment groups remain separate: sharing the control does not remove the need to recruit participants for each experimental arm.

N_{\mathrm{separate}}=\sum_{j=1}^{K}(n_{Cj}+n_{Tj})

N_{\mathrm{shared}}=\max_j(n_{Cj})+\sum_{j=1}^{K}n_{Tj}

These expressions apply only when one shared set of control participants is suitable for all \(K\) comparisons and the group-size requirements are minimums. Subtracting gives the ideal number of participants saved:

N_{\mathrm{separate}}-N_{\mathrm{shared}} =\sum_{j=1}^{K} n_{Cj}-\max_j(n_{Cj})

The calculation shows why unequal requirements matter. The common control must meet the largest requirement, not the average, and the treatment groups are still recruited in full. If one comparison accounts for nearly all required control participants, removing the smaller duplicate control groups yields only a modest reduction in the total.

Key result: Under the ideal assumptions, the control-count saving is the sum of the separate control requirements minus the largest of them. A small saving relative to total recruitment is not a contradiction: treatment-arm recruitment is unchanged, and the largest control requirement remains.

Worked Example: A Large Difference Between Control Requirements

Suppose two hypothetical comparisons require 36 control and 72 treatment participants for the first comparison, and 90 control and 48 treatment participants for the second. Assume both use the same eligible population, control condition, outcome and follow-up schedule, so that the same control observations are valid for both.

Separate trials. The first comparison requires \(36+72=108\) participants. The second requires \(90+48=138\). Thus,

N_{\mathrm{separate}}=108+138=246.

Shared control. The common control group must meet the larger control requirement, which is 90. Both treatment groups are still needed:

N_{\mathrm{shared}}=90+72+48=210.

The ideal saving is \(246-210=36\) participants, or \(36/246=0.1463\), about 14.6% of the separate-trial total. The control count falls from \(36+90=126\) to 90, but the 120 treatment participants are unchanged. The large difference between the two control requirements means the saving is only the smaller control requirement.

When Control Outcomes Cannot Be Reused

A shared control group is not merely a group labelled “control.” Its observations must represent the control condition for each treatment comparison. If the comparisons concern different eligible populations, different versions of standard care, incompatible outcome definitions or follow-up times, a control observation for one comparison may not be valid for another. In that case, counting a person once while using their data as if they supplied both required control observations can understate recruitment or misrepresent the estimands.

Different outcomes do not automatically rule out sharing. If the same control participants can be assessed on both outcomes at suitable times, and the protocol permits those data to contribute to both comparisons, they may still form a common control group. The relevant question is whether each participant supplies valid, usable data for every comparison—not simply whether the outcomes have different names. Extra visits, measurement burden, missingness or outcome-specific eligibility can make this infeasible or change the required sample sizes.

Worked Example: Treatment-Specific Outcomes and Separate Control Cohorts

Consider a hypothetical trial programme with two treatment comparisons. The first concerns a symptom score at 8 weeks and requires 50 control and 50 treatment participants. The second concerns a laboratory outcome at 24 weeks and requires 50 control and 70 treatment participants. Suppose the protocol requires separate control cohorts because the 8-week and 24-week control observations come from different eligible populations, and participants cannot contribute valid data to both comparisons.

Separate control cohorts require \(50+50=100\) control participants. The treatment groups require \(50+70=120\), so the total is:

N_{\mathrm{separate}}=50+50+50+70=220.

Although both comparisons list a control requirement of 50, a single group of 50 would not provide the two distinct, valid control cohorts specified by the protocol. On these assumptions, the required total remains 220; treating the control participants as reusable would give \(50+50+70=170\), an apparent saving of 50 that the design cannot deliver.

If instead the same eligible participants could provide both outcomes at the required times, and both observations were usable for the corresponding comparisons, 50 shared controls could meet both requirements. The total would then be \(50+50+70=170\). Whether this is appropriate depends on the population, measurements and protocol, not on the arithmetic alone.

Allocation Constraints Can Conflict

Sample-size requirements are sometimes paired with a specified allocation ratio. For comparison \(j\), the control-to-treatment ratio is \(n_{Cj}/n_{Tj}\). Separate trials can use a different ratio in each comparison. A single common control group, by contrast, has one size \(n_C\), so its pairwise ratios \(n_C/n_{Tj}\) are linked. If the design requires exact pairwise ratios that imply different control sizes, one common group cannot satisfy both constraints.

An exact ratio is a design constraint, not a universal statistical necessity. If the ratios were chosen as preferences or as a basis for the original calculations, a new shared-control allocation may be possible—but its precision and sample-size requirements need to be assessed under that allocation. Simply carrying over group sizes calculated for a different ratio does not demonstrate that the shared design achieves its planned operating characteristics.

Worked Example: Incompatible Exact Allocation Ratios

Suppose two hypothetical comparisons require 80 treatment participants each in the first arm’s design and 40 in the second arm’s design. The planned control-to-treatment ratios are 1:1 for the first and 3:1 for the second. Assume these ratios are exact protocol constraints.

For the first comparison, the control requirement is \(1\times80=80\). For the second, it is \(3\times40=120\). Separate trials therefore require \(80+120=200\) control participants and \(80+40=120\) treatment participants, for \(200+120=320\) participants overall.

A single shared control group would have to contain 80 participants to preserve the first exact ratio and 120 to preserve the second. Since \(80\ne120\), no one group size satisfies both. A shared group of 120 would give the first comparison a ratio of \(120/80=1.5:1\), not its required 1:1 ratio. That might be acceptable under a revised design, but it does not satisfy the stated exact constraints. The allocation and sample-size plan must be reconsidered before claiming a shared-control saving.

Diagnosing the Practical Advantage

A useful assessment distinguishes three questions: can the control observations be shared, does the proposed allocation meet each comparison’s requirements, and how large is the resulting reduction in unique participants? A positive answer to the first question alone does not settle the other two.

List requirements by comparison.
Record each control and treatment requirement, the outcome, population, control condition and follow-up schedule. Clarify whether counts refer to evaluable participants or recruits.

Test whether each control observation is reusable.
Check that the same person’s control data can validly contribute to every intended comparison. Identify any outcome, eligibility, timing or protocol restrictions that require distinct control cohorts.

Check allocation constraints.
Compare the proposed common control size with each comparison’s required allocation. If ratios or other operational rules are binding, verify that one common group can satisfy them.

Recalculate the total and the design requirements.
Count unique participants, include any extra recruitment needed for attrition, and assess the comparisons under the actual shared allocation. Report the saving against the appropriate separate-trial plan.

Warning signs include a control group sized by averaging the separate requirements, different control standards being treated as interchangeable, or a “shared” count that ignores a required outcome visit. Another warning sign is reporting the number of control observations used across analyses as though it were the number of unique people recruited. A participant may contribute to multiple comparisons, but should be counted only once in the recruitment total.

The main alternative is to retain separate control cohorts, either in separate trials or within a protocol that deliberately keeps cohorts distinct. This may be necessary for incompatible populations or outcomes, and it can preserve different allocation rules, but it gives up the recruitment reduction from reusing controls. A shared-control design is attractive when data are compatible and allocation works; it is not automatically preferable when the practical constraints make the nominal saving negligible or the planned design infeasible.

Common Mistake and Limitation

Wrong reasoning: “There are two treatment arms, so a common control must save one full control group.” This assumes that all control participants can be reused, that the larger group is sufficient for both comparisons, and that no allocation or measurement constraint intervenes.

Correction: Calculate the saving from the actual control requirements and verify that sharing is valid. Under ideal minimum-size requirements, the saving is the sum of control requirements minus their maximum—not necessarily one typical group. If the data are incompatible or the allocation constraints cannot be met, the ideal arithmetic does not describe a feasible design.

Even when sharing is valid, participant count is only one design consideration. A shared control may involve different randomization probabilities or operational demands than separate trials. The participant-count calculation is a screening comparison, not a substitute for checking that the planned analyses and design assumptions remain appropriate.

Check Your Understanding

Answer each question, then compare your reasoning with the solutions.

Two compatible comparisons require 30 and 75 control participants, respectively. How many control participants are needed separately and ideally with sharing? What is the control-count saving?
In Question 1, the treatment groups require 40 and 60 participants. Calculate both total recruitment counts and the percentage saving relative to separate trials.
Two comparisons require 45 controls each, but their control cohorts must come from distinct eligible populations. Can the ideal shared-control formula be used? Explain.
A common control group contains 100 participants. The treatment groups contain 50 and 25. What are the two control-to-treatment ratios? Why might these not meet a plan requiring exact ratios of 1:1 and 2:1, respectively?
Name two checks, beyond adding group sizes, that are needed before claiming a shared-control recruitment advantage.

Solutions. (1) Separate recruitment requires \(30+75=105\) controls. Ideal sharing requires \(\max(30,75)=75\), saving \(105-75=30\) controls. (2) Separate totals are \((30+40)+(75+60)=205\). Shared recruitment is \(75+40+60=175\). The reduction is \(30/205=0.1463\), or about 14.6%. (3) No. The control observations are not reusable across the distinct populations, so each comparison needs its own cohort: 90 control participants in total. (4) The ratios are \(100/50=2:1\) and \(100/25=4:1\), not the required 1:1 and 2:1. The common group size does not satisfy those exact constraints. (5) For example, check that the same control data are valid for each comparison and that the common allocation meets each comparison’s requirements; also check outcomes, eligibility, follow-up and recruitment losses.