Enrichment Design

An enrichment design is a clinical trial strategy that prospectively uses a patient characteristic, or a preliminary step such as a run in period, to select a study population in which a treatment effect, if one exists, is more likely to be detected than in an unselected population, then randomizes only those selected patients to receive the treatment or control. By concentrating the trial on patients who are less variable, at higher risk of the outcome, or more likely to respond, enrichment designs increase the probability of detecting a treatment effect, reduce the required sample size, and improve statistical efficiency. Enrichment strategies include predictive biomarker selection, run in periods, and therapeutic response enrichment.

Many treatments work well for a subset of patients but show modest or no average effect in an unselected population. A trial that enrolls all comers may fail to detect a real treatment effect because the benefit in responsive patients is diluted by the lack of effect in nonresponders. Enrichment designs address this problem by identifying the responsive subpopulation before randomization, enabling a more targeted and efficient evaluation. This approach is particularly important in oncology (where biomarker driven enrichment has become standard), psychiatry (where placebo response rates complicate efficacy evaluation), and precision medicine (where treatments are designed for genetically defined subgroups). In the US, the FDA's 2019 final guidance, Enrichment Strategies for Clinical Trials to Support Determination of Effectiveness of Human Drugs and Biological Products, describes enrichment as a way to make drug development more efficient, and names the generalizability and applicability of the results as the principal concerns.

The FDA's 2019 guidance groups enrichment strategies into three types: strategies that decrease variability (sometimes called practical enrichment), prognostic enrichment and predictive enrichment. Predictive enrichment selects patients based on a biomarker or clinical characteristic that predicts treatment response. For example, a trial of a targeted oncology drug may enroll only patients whose tumors express a specific molecular marker. This requires prior evidence that the marker predicts response and a validated assay for measuring it. Because the selected patients are more likely to respond, predictive enrichment can increase both the absolute and the relative size of the treatment effect.

Prognostic enrichment selects patients at higher risk for the outcome of interest, increasing the event rate and thereby the statistical power of the trial. For example, a cardiovascular prevention trial may enroll only patients with multiple risk factors rather than all adults, so that enough events are likely to occur within a feasible study duration. Prognostic enrichment increases the absolute difference between the groups but would not be expected to change the relative effect, such as the relative risk reduction.

Run in enrichment, one of the strategies for decreasing variability, uses a preliminary period before randomization to identify patients who tolerate the drug and adhere to the protocol; a placebo run in (placebo lead in) can also identify patients who improve spontaneously or respond strongly to placebo. Patients who do not tolerate the drug, adhere poorly or, in a placebo run in, respond to placebo are excluded before randomization. This can increase the trial's power by reducing dropout and nonadherence but limits generalizability to patients who can tolerate the treatment and adhere to the regimen. Other strategies for decreasing variability include defining entry criteria carefully, so that enrolled patients actually have the disease being studied, and enrolling only patients whose baseline measurements are consistent.

The design requires careful planning. The enrichment criterion must be specified before the trial begins, ideally supported by prior evidence from mechanistic studies, biomarker research, or exploratory trials. The statistical analysis plan must account for the enrichment, and the trial's conclusions must be explicitly limited to the enriched population. For US submissions, the FDA's guidance asks sponsors to consider whether the enrichment strategy could be used in practice to identify the patients who should receive the drug; when a marker selects the patients, the test used after approval would usually be an FDA cleared or approved companion diagnostic.

Enrichment designs are appropriate when there is credible prior evidence that a treatment is more effective in a definable subgroup, when the overall population effect is expected to be too small to detect with a feasible sample size, or when high placebo response rates in an unselected population would obscure the treatment effect. They are standard in oncology drug development (where molecular marker enrichment is routine), increasingly common in neurology and psychiatry, and relevant to any precision medicine application. They are not appropriate when the treatment is expected to benefit a broad population equally, when no reliable enrichment criterion has been identified, or when the enrichment criterion excludes a large proportion of patients who might benefit.

A common mistake is using enrichment without sufficient prior evidence that the enrichment criterion predicts response, which can produce misleading results if the criterion does not actually identify responders. Another error is failing to acknowledge the limited generalizability of enrichment trial results; a drug shown to be effective in biomarker positive patients may or may not be effective in biomarker negative patients, and the enriched trial cannot answer this question. Run in enrichment that excludes placebo responders or patients who cannot tolerate the drug raises the question of whether the resulting estimates overstate how well the drug works, or how well it is tolerated, in unselected patients. Post hoc enrichment, identifying a responsive subgroup after an unselected trial has failed and re analyzing only that subgroup, is a different and problematic practice that inflates false positive rates and should not be confused with a prospectively designed enrichment strategy. Finally, some enrichment designs exclude patients based on criteria that are not practical to assess in routine clinical settings, limiting the clinical applicability of positive results.

A Phase III trial of a targeted kinase inhibitor for non small cell lung cancer uses predictive enrichment, enrolling only patients whose tumors harbor a specific activating mutation (identified by a validated companion diagnostic test). Of 2,400 screened patients, 380 (16%) test positive for the mutation and are randomized to the kinase inhibitor or standard chemotherapy. After a median follow up of 18 months, the kinase inhibitor shows a median progression free survival of 12.4 months compared to 6.2 months for chemotherapy (hazard ratio 0.46, 95% CI: 0.34 to 0.62, p < 0.001). The drug receives regulatory approval specifically for patients with the target mutation, with the companion diagnostic test required for patient selection.