Design and analysis of clinical trials when treatment benefit is suspected to be different in subpopulations

Ekkehard Glimm
Novartis Pharma, Basel, Switzerland

Recently, statisticians in the pharmaceutical industry are facing a need for more complex confirmatory trial designs. One of the drivers of this development is the improvement of diagnostic predictors (like genetic biomarkers). For example, laboratory experiments may hint at a larger treatment benefit in patients who express a certain gene. However, at the start of the clinical trial, such an increased subpopulation benefit is often still hypothetical. Hence, the confirmatory clinical trial begins with the multiple aim of (i) establishing the treatment effect in the full population, or (ii) in the subpopulation and (iii) finding out if the hypothesis about an enhanced effect in the subpopulation is true.
From the design perspective, this situation calls for designs where an interim analysis is used to decide about the primary comparison (subpopulation or full population) and potential changes in recruitment (e.g. an increase of the number of subpopulation patient in an “enrichment design”). With respect to data analysis, the multiplicity issue arising from the comparison of treatment effects in two (sub- and full population) or three (sub-, non-sub- and full population) needs to be addressed.
In this talk, designs and analyses for such clinical trials will be discussed. The discussion also adresses other situations where similar statistical challenges arise (e.g. multiregional confirmatory trials which have to be submitted to several health authorities who are primarily interested in “their” regional subpopulation; trials where treatments have different modes of application, each with a corresponding control treatment).