A Simulation Study Comparing Different Statistical Approaches for the Identification of Predictive Biomarkers

Figure 1

Scenarios used in the simulation study for comparison of statistical methods. In scenarios 1 and 2 (a, b), data are generated under the null hypothesis of no biomarker-treatment interaction. In scenarios 3 to 6, which are illustrated in (c) to (f), the hazard ratio (illustrated on a log-scale by the red line) depends on the biomarker value, so biomarker-treatment interactions are present.

(a)

(b)

(c)

(d)

(e)

(f)