Subpopulations are often of key interest in clinical trials. For example, an investigational drug may be targeted to have efficacy on a particular subpopulation of patients who test positive for a specific biomarker. That drug may also be thought to have some degree of benefit, though lesser, on the opposite subpopulation—those who test negative for that biomarker.

This leads to consideration of at least three key hypotheses for the trial:

1.    efficacy in the biomarker-positive subpopulation;
2.    efficacy in the full population; and
3.    efficacy in the biomarker-negative population.

Hypothesis testing

These three efficacy hypotheses can be tested sequentially at the type-1 error level set for the trial, or the type-1 error can be divided among them. This leads to many potential approaches to set up the key hypotheses for the trial.

Sequential testing first in full population

Schematic 1: Sequential testing: first test full population, and only if statistically significant, then test biomarker positive subpopulation, and only if statistically significant, then test biomarker negative subpopulation.

Sequential testing first in biomarker-positive subpopulations

Schematic 2: Sequential testing: first test biomarker positive subpopulation, and only if statistically significant, then test full population, and only if statistically significant, then test biomarker negative subpopulation.

Testing with divided alpha

Schematic 3: Divide alpha=0.05 as follows and pass to other tests as indicated. Test Biomarker positive subpopulation and full population at alpha=0.01, and test biomarker negative subpopulation at alpha=0.005.  For each statistically significant test of those three, pass half that alpha to the other two tests if they were not first statistically significant at their respective original alpha level, then combine the original alpha and passed alpha and retest.

Selection of the best design for the trial and the best approach to addressing the hypotheses may involve maximizing statistical power and minimizing trial duration, cost (or sample size) for several potential true underlying scenarios of stratified response (for factors in addition to biomarker status), dropouts, and recruitment. These considerations counterbalance one another differently for different scenarios.

Simulation-guided design for subpopulation analysis

Cytel’s cloud-native trial design and simulation software can simulate thousands of design options for many true underlying scenarios to inform selection of the “best” clinical trial design.  East Horizon^TM offers the option to compare different approaches to subpopulation analysis as described above.

Interested in learning more? Valeria Mazzanti, Associate Director of Customer Success, and J. Kyle Wathen, Vice President, Scientific Strategy and Innovation, discuss the integration of East® and R. With this new capability, users have greater latitude in selecting input parameters, such as analysis types and test statistics, beyond those that are native to the software. Click to watch the on demand webinar: East + R Integration.

Share Via