Changing the Target Population

Why Change Who Can Enter a Trial?

A trial’s eligibility criteria define which patients can participate. Sometimes evidence or practical experience suggests that the treatment may work differently across patient groups. A planned adaptation may then narrow eligibility to a group considered more likely to benefit, or otherwise change the characteristics required for entry. This can make the trial more focused, but it also changes who the trial can provide evidence about.

Changing eligibility is not simply a way to improve a result. It involves a trade-off: a narrower group may have a stronger treatment effect or a more feasible study, while the findings may be less applicable to patients who no longer qualify. The rationale, timing, decision rule, and consequences for recruitment and analysis therefore need to be considered before the change is made.

Enrichment may use a baseline characteristic measured before randomization, such as a laboratory marker, disease stage, or prior treatment history. The characteristic should be measured consistently, and the rule determining eligibility should be operationally clear. For example, “marker-positive” must correspond to a defined assay and threshold, rather than a judgment made differently across sites.

What Evidence Supports Enrichment?

A characteristic can be associated with a better outcome without identifying patients who benefit more from the experimental treatment. For instance, marker-positive patients might have better outcomes in both randomized groups. That pattern is prognostic: the marker is associated with outcome. A marker is predictive of differential treatment benefit when the treatment contrast differs across marker groups.

Randomized comparisons within marker groups can help assess whether treatment effects differ. A favorable outcome among marker-positive patients receiving the experimental treatment alone is not enough: their outcomes must be compared with those of similar marker-positive patients assigned to control. Nor does observing a promising result in one group automatically establish that the treatment effect truly differs between groups. Differences can arise from random variation, particularly when many subgroups are considered.

A population adaptation should therefore be supported by a prespecified rationale and a clear rule. The rule may rely on external evidence, blinded information about the distribution of a baseline characteristic, or interim treatment comparisons. These sources raise different statistical concerns. In particular, using unblinded outcomes to select a subgroup and then treating the selected subgroup’s result as if it had been chosen in advance can overstate evidence. If treatment outcomes influence the adaptation, the design and analysis must account for that selection and preserve appropriate error control.

Changing Eligibility During Recruitment

When eligibility changes after a trial begins, the adaptation must specify how the new rule will be applied. A basic plan states the criteria before and after the change, the date or information that triggers it, how existing participants are handled, and which analyses will address the resulting groups. These details help prevent an informal shift in eligibility based on individual cases or emerging results.

After the change, patients who were eligible earlier may differ from those eligible later. They may have different disease severity, access to care, background treatment, or likelihood of enrolling. Calendar time may also coincide with changes in clinical practice or recruitment sites. If treatment effects vary across these characteristics, combining all participants without accounting for the change can obscure what comparison is being made.

Randomization should continue to support a fair treatment comparison among patients eligible under the rule in force at their enrollment. Concurrent randomized controls are especially important: comparing experimental-treatment patients enrolled after a change with control patients enrolled only before it can confound treatment with calendar time. When the adaptation uses outcomes observed during the trial, the analysis must also consider how that data-dependent selection affects statistical error rates. Randomization alone does not make every post-selection analysis valid.

The calculation shows why “the treatment worked in the enriched trial” and “the treatment benefits the general patient population” are not interchangeable conclusions. The first concerns the enrolled, enriched group. The second needs evidence supporting extension beyond that group, including evidence about patients excluded by the adaptation.

Recruitment and Applicability

Narrower criteria commonly reduce the pool of eligible patients. The reduction can be substantial if the characteristic is uncommon, requires a specialized test, or is difficult to measure at participating sites. Screening may take longer, testing may add expense or delay, and some patients may become ineligible after initial assessment. Recruitment projections should be recalculated using the new criteria rather than carried forward from the original plan.

Applicability is also affected by how patients differ from those who were eligible. If the adaptation excludes groups that are common in clinical practice, the trial’s direct evidence will be limited for those groups. The direction and size of any resulting difference cannot be inferred merely from the fact that the criteria narrowed. It depends on whether treatment effects and relevant patient characteristics differ.

A Practical Assessment and Common Pitfalls

A useful assessment asks what the trial gains and what evidence it gives up. The following steps make that trade-off explicit.

A common mistake is to reason, “The subgroup had a better outcome, so the trial should enroll only that subgroup.” This fails if the subgroup’s outcome was better in both randomized groups, because that is not evidence of a larger treatment effect. It also fails if the subgroup was selected after looking at noisy treatment comparisons and the analysis ignores that selection. The correction is to compare randomized treatments within groups, assess whether their treatment contrasts differ, and use a prespecified, statistically appropriate adaptation and analysis plan.

Another mistake is to regard a more homogeneous or enriched sample as automatically superior. It may improve focus, but can slow recruitment, reduce representation, and leave uncertainty about patients outside the criteria. Enrichment is most defensible when there is a strong rationale for the group, a reliable identification method, a feasible recruitment plan, and a clear account of the limits on applying results more broadly.

Unlike changing a treatment arm, changing eligibility alters which patients enter the comparison. Both kinds of adaptation need a prespecified rule and careful interpretation, but population changes raise a particular question: whether the evidence remains relevant to the patients clinicians will later consider treating. No single enrichment strategy resolves that question for every trial.