Study design: Population enrichment RCT

What is it?

Phase 2 or 3 randomised controlled trials (RCTs) typically exclude patients with certain baseline characteristics, which may be drivers of effectiveness (for a definition, see here) or treatment effect modifiers. The exclusion of these characteristics (which may occur in a large proportion of the real-life population) can hamper learning about a new medicine’s effectiveness in real-life. Additionally, excluding these patients can prevent the use of predictive modelling techniques to estimate the effectiveness of the medicine in these patients.

To address this problem, population enrichment combined with further predictive modelling techniques can be used within a traditional RCT by relaxing some eligibility criteria without increasing the sample size or compromising the success of the RCT.

This is done by:

  • Step 1: Identifying the exclusion criteria that might affect the medicine’s effect estimate in the RCT if applied strictly (i.e. the efficacy-effectiveness gap – see a definition here); consider including a subset of these patients (the enrichment subset) within the trial, taking into consideration feasibility and safety.
  • Step 2: Using modelling techniques to determine a suitable sample size of the enrichment subset. The sample size should be large enough to allow prediction of effectiveness, but not too large, to maximize the chances of success of the RCT.
  • Step 3: Performing the RCT exactly as usual; the whole RCT population is used to calculate the primary endpoint (i.e. no sub-group analyses are used).
  • Step 4: Predicting the medicine’s effectiveness in a real population using the information based on the ‘enrichment subset’; usual predictive modelling techniques may be used (for example, Bayesian or regression models).

Why is it useful?

Improving generalisability and prediction of real-life effects: enriching typical phase 3 trials with selected factors (those typically used as exclusion criteria in conventional RCTs) can improve their generalisability because the trial population will better reflect a real-world population. Consequently, the trials may improve the prediction of the real-life effects of the investigated medicine.

When is it useful?

This technique could, in theory, be useful in many disease areas where the exclusion of participants with particular characteristics reduces the generalisability of an RCT.

What are the limitations?

This type of study has not yet been conducted in practice. The acceptability of this technique to regulatory and health technology assessment agencies has not been fully explored. A modelling and simulation study on the use of this design in schizophrenia has been conducted as part of GetReal.

Key contributors

Helene Karcher and Clementine Nordon, LASER