Diminishing Recruitment Savings as Arms Are Added

Why the Number of Arms Matters

When several treatments are compared with the same control, each treatment can be compared with one common control group. The control participants then contribute information to every treatment-versus-control comparison, but each person is recruited only once. This can reduce total recruitment compared with running a separate two-arm trial for each treatment.

The amount saved depends on how many treatment arms are included and on the size of the control group each comparison would otherwise require. A useful distinction is between absolute savings, measured in participants, and percentage savings, measured relative to the recruitment required by separate trials. Absolute savings increase steadily as arms are added under the assumptions below; percentage savings increase more slowly, with each additional arm producing a smaller percentage-point gain.

Set Up the Comparison

Definition: Let \(k\) be the number of experimental treatment arms. Suppose each treatment-versus-control comparison requires \(n_T\) participants in its treatment group and \(n_C\) participants in its control group. If \(k\) separate two-arm trials are conducted, their total recruitment is \(k(n_T+n_C)\). If the \(k\) treatment arms share one control group of size \(n_C\), total recruitment is \(kn_T+n_C\). The absolute recruitment saving is the separate-trial total minus the shared-control total.

This comparison assumes the same target sample sizes for each treatment arm and for each control group that would appear in a separate trial. It also assumes that a single shared control group of size \(n_C\) is an adequate comparator for all \(k\) treatment arms. The calculation is about participant counts; it does not itself establish that the designs have the same statistical properties or that any other design requirements are satisfied.

In particular, the target sample sizes are inputs to this calculation. If a design requirement changes the sample size needed for each comparison, the totals must be recalculated using those revised targets. The arithmetic here does not determine what sample size is required for a particular study.

Deriving the Savings and Their Rate of Increase

Subtract the shared-control total from the separate-trial total. The \(kn_T\) treatment participants occur in both designs and cancel. What remains is the cost of recruiting \(k\) separate control groups rather than one:

\begin{aligned} S_k &= k(n_T+n_C)-(kn_T+n_C) \\ &= (k-1)n_C . \end{aligned}

Thus, the absolute saving is determined by the number of control participants avoided. Adding one treatment arm increases both designs' treatment recruitment by \(n_T\), but the separate-trial design also requires another control group of \(n_C\). The shared-control design does not. Consequently, the absolute saving increases by \(n_C\) participants per added arm.

To express the saving as a proportion of the separate-trial recruitment, divide \(S_k\) by \(k(n_T+n_C)\). Multiplying by 100 gives the percentage saved:

R_k = \frac{S_k}{k(n_T+n_C)} = \frac{(k-1)n_C}{k(n_T+n_C)}.

When allocation is equal within each comparison, so \(n_T=n_C=n\), this simplifies to:

S_k=(k-1)n, \qquad R_k=\frac{k-1}{2k} =\frac{1}{2}-\frac{1}{2k}.

This formula clarifies what “diminishing savings” means. The absolute saving grows by the same number of participants for every additional arm. But the percentage saved rises by progressively smaller amounts. For general \(n_T\) and \(n_C\), the increase in the saving proportion from \(k\) arms to \(k+1\) arms is:

R_{k+1}-R_k = \frac{n_C}{(n_T+n_C)k(k+1)}.

Because \(k(k+1)\) gets larger as \(k\) increases, this increment gets smaller. The percentage saved approaches \(n_C/(n_T+n_C)\) as the number of arms grows. With equal allocation, that limit is 50%. It is a limit on the percentage of separate-trial recruitment saved under this setup—not a claim that total recruitment stops increasing.

Worked Example: Equal Allocation With Increasing Arm Counts

Suppose each comparison requires 80 participants in the treatment group and 80 in the control group. Compare separate trials with one shared-control design for \(k=2\), \(3\), and \(5\) treatment arms.

For \(k=2\), separate trials recruit \(2(80+80)=320\) participants. A shared-control design recruits \(2(80)+80=240\). The saving is \(320-240=80\) participants, or \(80/320=0.25\), which is 25%.

For \(k=3\), separate trials recruit \(3(160)=480\), while the shared-control design recruits \(3(80)+80=320\). The saving is \(480-320=160\) participants, or \(160/480=0.3333\), which is about 33.3%.

For \(k=5\), separate trials recruit \(5(160)=800\), while the shared-control design recruits \(5(80)+80=480\). The saving is \(800-480=320\) participants, or \(320/800=0.4000\), which is 40%.

Each added arm saves another 80 participants in absolute terms. Yet the percentage rises from 25% to 33.3% to 40%: the gains are 8.33 and 6.67 percentage points, respectively. The percentage gains are diminishing even though the count saved continues to increase steadily.

Worked Example: Unequal Treatment and Control Targets

Suppose each treatment-versus-control comparison requires 90 treatment participants and 120 control participants. For four treatment arms, calculate the two recruitment totals and the saving.

Separate trials require \(4(90+120)=4(210)=840\) participants. With one shared control group, recruitment is \(4(90)+120=360+120=480\) participants.

The absolute saving is \(840-480=360\) participants. The percentage saved relative to separate trials is:

\frac{360}{840}\times 100\%=42.86\%.

Here each added arm saves 120 control participants relative to conducting that comparison as a separate trial. As \(k\) grows, the percentage saved approaches \(120/(90+120)=120/210=57.14\%\). It does not approach 50% because the treatment and control targets are not equal.

Worked Example: How Many Arms Reach a Target Saving?

With equal allocation, how many treatment arms are needed for shared control to save at least 40% of the recruitment required for separate trials?

For equal allocation, the saving proportion is \((k-1)/(2k)\). Set this greater than or equal to 0.40:

\frac{k-1}{2k}\geq 0.40 \quad\Longrightarrow\quad k-1\geq 0.80k \quad\Longrightarrow\quad 0.20k\geq 1 \quad\Longrightarrow\quad k\geq 5.

Since the number of arms must be a whole number, at least five treatment arms are needed. At \(k=5\), the saving proportion is \((5-1)/(2\cdot5)=4/10=40\%\). At \(k=4\), it is \(3/8=37.5\%\), which falls short. These percentages compare recruitment totals under the stated equal-allocation assumptions.

Check the Comparison Before Using It

The formulas are simple, but their interpretation depends on a fair comparison between designs. Use the following checks before drawing a conclusion from a calculated saving.

Use the same per-comparison targets. The calculation assumes every treatment arm needs \(n_T\) participants and each corresponding control group would need \(n_C\). If the targets differ across comparisons, add the separate requirements explicitly and count the shared control group only once.
Count people, not comparisons. A shared control participant may contribute to several comparisons, but is still one recruited participant. Do not count that person once for every treatment arm when calculating total recruitment.
Keep the percentage denominator clear. The saving percentage here is divided by recruitment for separate trials, \(k(n_T+n_C)\). A different denominator produces a different percentage and must be labelled accordingly.
Distinguish target sample size from actual recruitment. If the numbers used are required evaluable participants, extra recruitment to allow for loss to follow-up is not included unless it is added to the targets first.

A useful comparison with the main alternative—running separate two-arm trials—is shown below. It makes clear that sharing controls removes duplicated control recruitment, not treatment-arm recruitment.

Feature	Separate two-arm trials	One shared-control design
Treatment participants	\(kn_T\)	\(kn_T\)
Control participants	\(kn_C\)	\(n_C\)
Total recruitment	\(k(n_T+n_C)\)	\(kn_T+n_C\)
Control recruitment avoided	None relative to itself	\((k-1)n_C\)

These are recruitment-count comparisons under specified targets, not a universal guarantee that a shared-control design is preferable. The calculation does not account for every practical or statistical consideration. In particular, if the valid sample-size targets differ between designs, or if a shared control group is not appropriate for all treatment comparisons, the simple saving formula is not enough to decide between them.

Common Mistake: Expecting Percentage Savings to Be Proportional

A common error is to reason that doubling the number of treatment arms doubles the percentage saved. That confuses the count of duplicated controls avoided with the percentage of all participants saved. Under equal allocation, the absolute saving is \((k-1)n\), but the separate-trial total is \(2kn\); both change with \(k\). Their ratio is \((k-1)/(2k)\), which does not grow in direct proportion to \(k\).

For example, with equal allocation, two arms save 25%, while four arms save 37.5%, not 50%. The correction is to calculate the saving as a fraction of the full separate-trial total each time. The percentage approaches 50% under equal allocation, while the absolute number of participants saved continues to rise by \(n\) for each additional arm.

Key takeaway: With \(k\) treatment arms and a shared control group of size \(n_C\), the simple recruitment saving is \((k-1)n_C\). Its absolute increase per added arm is constant, but its percentage of separate-trial recruitment rises at a diminishing rate under fixed per-comparison sample-size targets.

Check Your Understanding

Try each question, then check the solutions below.

Each comparison requires 50 treatment participants and 50 control participants. With three treatment arms, calculate the separate-trial total, the shared-control total, and the percentage saved.
For four treatment arms, each requiring 70 treatment and 100 control participants, how many participants are saved by sharing one control group? What percentage of the separate-trial total is that?
With equal allocation, what is the saving proportion for six treatment arms? What is the increase in the saving proportion when moving from six to seven arms?
Why does a shared-control participant count once in total recruitment even if that participant contributes to several treatment comparisons?

Solutions. (1) Separate trials require \(3(50+50)=300\); the shared-control design requires \(3(50)+50=200\). The saving is 100 participants, or \(100/300=33.3\%\). (2) The saving is \((4-1)100=300\) participants. Separate trials require \(4(70+100)=680\), so the percentage saved is \(300/680=44.12\%\). (3) Six arms save \((6-1)/(2\cdot6)=5/12=41.67\%\). The increase from six to seven arms is \(1/[2\cdot6\cdot7]=1/84=0.01190\), or about 1.19 percentage points. (4) Recruitment counts distinct people enrolled, not the number of comparisons in which their data are used; reuse in several comparisons does not mean recruiting that person several times.