Figure 3.

Biomarker-stratified design.

Note: At drug randomization (assuming a reasonable sample size), all variables other than biomarker status should be fairly evenly distributed in drug A and drug B groups. This explains the likely test (biomarker)–drug relationship but not the incremental benefit of the test, that is there may be uncertainty as to whether the biomarker +ve/-ve is responsible for the differential treatment effect or some other unmeasured variable.