Choosing Between All-Comer And Enriched Recruitment

Why Recruitment Strategy Is a Design Choice

An all-comer trial enrolls eligible patients regardless of biomarker status. A marker-restricted, or enriched, trial enrolls only patients who meet a prespecified marker rule. The choice affects more than the number of patients who can enter: it determines which population is directly represented, how much of any treatment-effect heterogeneity the trial can observe, and how many people may need screening to recruit each participant.

The decision should therefore begin with the intended treatment claim. If the aim is to evaluate treatment for the broad eligible population, evidence from all-comer recruitment is usually the direct fit. If the intended claim is specifically for marker-positive patients, enrichment can focus recruitment and information on that group. A larger estimated effect in a selected subgroup alone does not establish that the treatment is ineffective in other patients, nor does it establish a broad-population benefit.

Definition: Let $M=1$ denote meeting a prespecified marker rule and $M=0$ denote not meeting it. Let $\pi=P(M=1)$ be the marker-positive prevalence in the eligible recruitment population. All-comer recruitment includes both marker groups; marker-restricted recruitment includes only $M=1$. The choice of recruitment population should be aligned with the population about which the trial is intended to make a treatment claim.

Effects, Prevalence, and the Population Question

For a clear comparison, first state the effect scale and direction. In the examples below, the outcome is measured in percentage points and higher values are better. Define the subgroup treatment effect as the mean outcome under treatment minus the mean outcome under control for that marker group. Write these effects as $\Delta_1$ and $\Delta_0$. A positive value favors treatment.

If the marker groups form an exhaustive partition of the eligible population, and $\pi$ describes that same population, the all-comer average treatment effect on this additive scale is the prevalence-weighted average of the subgroup effects. This is an identity for the population average under the stated partition; it does not imply that an enriched trial directly estimates the effect in marker-negative patients.

\Delta_{\mathrm{all}}=\pi\Delta_1+(1-\pi)\Delta_0

This equation makes the role of prevalence explicit. A large treatment effect in a rare marker-positive group may have a limited contribution to the average effect across all eligible patients. Conversely, if the treatment effect is similar in both groups, restricting recruitment may sacrifice broad representation without gaining much effect concentration. The arithmetic must be performed on the scale relevant to the scientific question: a weighted average of risk differences is not generally the same as an average of odds ratios or hazard ratios.

Worked Example: How Prevalence Changes the All-Comer Effect

Suppose a hypothetical trial-planning scenario has marker-positive prevalence $\pi=0.30$. On an outcome scale where higher is better, assume the treatment effects used for planning are $\Delta_1=8$ percentage points in marker-positive patients and $\Delta_0=1$ percentage point in marker-negative patients. Calculate the implied all-comer average effect.

Substitute the prevalence and subgroup effects into the weighted-average formula:

\Delta_{\mathrm{all}}=(0.30)(8)+(0.70)(1) =2.4+0.7=3.1\text{ percentage points}.

Under these assumptions, the all-comer average effect is 3.1 percentage points, while the marker-positive effect is 8 percentage points. Enrichment would concentrate the trial on the group with the larger assumed effect, but it would not directly answer whether marker-negative patients benefit. The calculation is only as relevant as its inputs: if prevalence or subgroup effects differ in the actual recruitment population, the implied all-comer average also changes.

Recruitment Cost and Yield

Enrichment often requires testing people before enrollment. When only marker-positive patients are eligible, low prevalence can mean that many people must be screened to recruit the required number. A simple planning calculation separates the screening cost from the cost of treating and following an enrolled participant.

Planning relationship: If prevalence is stable at $\pi$, the marker is measured for every person screened, and each marker-positive person can be enrolled, then recruiting $n_E$ marker-positive participants requires an expected $n_E/\pi$ screens. With screening cost $c_s$ per person and enrollment cost $c_e$ per participant, expected enrichment cost is approximately $n_Ec_e+(n_E/\pi)c_s$. This expectation ignores fixed costs, screen failures beyond marker status, and other operational complications.

The comparison must use sample sizes designed for the relevant trial objectives, not assume that the all-comer and enriched trials need the same number of participants. Their estimands, effect sizes, outcome variability, and analysis plans may differ. Sample sizes should be obtained using the prespecified design assumptions and operating characteristics; the cost formula then translates those requirements into a recruitment budget.

Worked Example: Expected Screening Cost Versus All-Comer Cost

Consider a hypothetical planning comparison. The marker-positive prevalence is estimated as $\pi=0.25$. Screening costs \$120 per person, and enrolling and following a participant costs \$480. The planning calculations for the distinct objectives call for 200 marker-positive participants in an enriched trial and 450 participants in an all-comer trial. Assume every screened marker-positive person can be enrolled, and the all-comer trial does not require marker screening.

For enrichment, the expected number screened is:

\frac{n_E}{\pi}=\frac{200}{0.25}=800\text{ people}.

The expected screening cost is $800\times \$120=\$96{,}000$. The enrollment and follow-up cost is $200\times \$480=\$96{,}000$, so total expected cost is $\$96{,}000+\$96{,}000=\$192{,}000$.

For all-comer recruitment, the stated planning sample size costs $450\times \$480=\$216{,}000$. Under these simplified assumptions, enrichment is expected to cost $\$216{,}000-\$192{,}000=\$24{,}000$ less. Relative to the all-comer cost, that is $\$24{,}000/\$216{,}000\approx 0.111$, or about 11.1%.

This is not a general rule that enrichment is cheaper. The conclusion depends on prevalence, assay and recruitment costs, and the sample sizes needed for each objective. If the prevalence is lower, more people must be screened; if screening is costly or the enriched sample size is large, enrichment may cost more. The expected number of screens is also not a guaranteed number: actual screening demand varies.

Matching Recruitment to the Intended Claim

Recruitment should be chosen in relation to the population for which the treatment is being evaluated. An all-comer design directly samples both marker groups and can estimate an average treatment effect across the enrolled eligible population, subject to its design and analysis assumptions. A marker-restricted design directly evaluates treatment in marker-positive participants. Its results do not, by themselves, establish an effect in marker-negative patients or an average effect for all eligible patients.

That distinction matters even when the marker-positive group has a much larger expected effect. A targeted claim may be appropriate if the clinical objective is to treat that group and the marker rule can be applied reliably. A broad claim requires evidence that supports the broader population; a large enriched-trial effect is not a substitute for observing the omitted group. Details of estimating effects for a target population are a separate question.

Worked Example: A Large Subgroup Effect Is Not a Broad Claim

Suppose a hypothetical eligible population has marker-positive prevalence $\pi=0.40$. The assumed treatment effect is 7 percentage points in marker-positive patients and 0 percentage points in marker-negative patients, on an additive outcome scale where higher is better.

The implied all-comer average is:

\Delta_{\mathrm{all}}=(0.40)(7)+(0.60)(0) =2.8\text{ percentage points}.

If the scientific objective is a treatment claim limited to marker-positive patients, enrolling only that group can directly address the target objective, assuming the marker rule and trial conduct are valid. If the objective is a claim for all eligible patients, the 7-point enriched effect is not the all-comer effect: the planning assumptions imply 2.8 percentage points across the full population. An all-comer design, or a design that otherwise obtains adequate evidence in both groups, is better aligned with a broad objective.

The calculation does not prove that the treatment effect is exactly zero in marker-negative patients; that value was an assumption in this hypothetical scenario. Trial conclusions must be based on observed data and the prespecified analysis, not on planning assumptions alone.

A Structured Comparison

A defensible choice makes the trade-offs visible rather than optimizing one feature in isolation. In particular, recruitment convenience and effect concentration do not override the intended population of the claim.

Design consideration	All-comer recruitment	Marker-restricted recruitment
Population directly enrolled	Marker-positive and marker-negative eligible patients	Patients meeting the marker rule
Effect concentration	May be diluted if effects differ substantially across groups	Can focus on a group with a larger expected effect
Screening burden	May not require marker testing for eligibility	Depends on marker prevalence, testing costs, and recruitment yield
Directly supported scope	Can address the broad eligible population, with suitable evidence and analysis	Directly addresses the marker-defined enrolled population
Main planning sensitivity	Subgroup prevalence, effect pattern, and all-comer sample size	Marker prevalence, test and screening costs, and enriched sample size

The comparison also depends on whether the marker rule is prespecified and operationally stable. The earlier material on treatment-effect heterogeneity and enrichment selection is relevant when considering how subgroup evidence is assessed; the earlier material on simulation operating characteristics is useful when a design must be evaluated across uncertain scenarios. Recruitment choice should not be based solely on the largest observed subgroup effect, especially when that group was identified after examining trial data.

Assumptions, Diagnostics, and Limitations

The weighted-average calculation requires a common additive effect scale, subgroup definitions that partition the relevant eligible population, and a prevalence appropriate to that same population. It also assumes that $\Delta_1$ and $\Delta_0$ refer to comparable outcomes, follow-up, and treatment contrasts. If subgroup effects are estimated imprecisely, their weighted average inherits that uncertainty; the arithmetic alone is not a confidence interval or a test.

The simple cost calculation assumes a stable prevalence during recruitment, a sufficiently accurate marker classification for eligibility, and no additional loss between a positive test and enrollment. In practice, track the observed screening yield and compare it with the planning prevalence. If fewer marker-positive patients are found than expected, projected costs and timelines can rise. Also inspect reasons for screen failure, assay turnaround, missing or indeterminate results, and whether recruitment is disproportionately coming from a small number of sites. These signals can show that the assumed recruitment process is not describing actual operations.

Prevalence and effects may vary across sites or over calendar time. A single planning value for $\pi$ can then be misleading, and an apparently favorable cost comparison can be fragile. Stress-testing plausible prevalence and subgroup-effect scenarios is preferable to treating one set of assumptions as known. A marker with imperfect classification can also make the group actually enrolled differ from the group intended by the marker rule; assessing the clinical utility of such classification is a distinct design problem.

Common Mistake and Correction

Wrong reasoning: “The estimated treatment effect is largest in marker-positive patients, so we should recruit only them; the result will support treatment for everyone.” This confuses a subgroup-specific effect with an effect in the full eligible population. It also ignores whether the subgroup difference is reliable and whether the intended claim is targeted or broad.

Correction: State the treatment claim first. Use marker prevalence and effects on a defined scale to understand how a subgroup effect relates to the all-comer average, and compare the recruitment and screening costs for sample sizes appropriate to each objective. An enriched trial directly supports conclusions for the enrolled marker-defined group; broader conclusions require evidence appropriate to the broader population.

Key takeaway: Choose all-comer or marker-restricted recruitment by jointly considering the intended claim, marker prevalence, plausible treatment-effect heterogeneity, and the full recruitment cost. Neither a larger subgroup effect nor a lower expected cost is sufficient on its own.

Check Your Understanding

Answer each question before reading the solution that follows it.

A marker-positive group has prevalence $0.20$, with assumed treatment effects of 6 percentage points in marker-positive patients and 1 percentage point in marker-negative patients. What is the implied all-comer effect on this additive scale?
If marker prevalence falls while the number of required marker-positive participants stays fixed, what happens to the expected number of screens under the simple cost model?
A trial enrolls only marker-positive patients. What population does it directly study, and why is its estimated effect not automatically an all-comer effect?
Name two assumptions behind the simple enrichment cost calculation that should be checked during recruitment.

Solutions: 1. $(0.20)(6)+(0.80)(1)=1.2+0.8=2.0$ percentage points. 2. Expected screens increase because $n_E/\pi$ increases as $\pi$ falls. 3. It directly studies patients meeting the marker rule; marker-negative patients were not enrolled, so the trial does not directly estimate their effect or the prevalence-weighted all-comer average. 4. Examples include stable prevalence, accurate marker classification, enrollment of eligible marker-positive people without further losses, and screening everyone at the stated per-person cost.