Fig. 1

Illustrative example of hypothesis testing in exploratory subgroup discovery using 1000 iterations of twofold cross-prediction. The example considers a primary RCT with two arms where a null hypothesis of ‘no improvement from the experimental treatment’ is not rejected; i.e. there is no significant evidence of the experimental treatment providing improvement over the standard of care. Each random division results in a corresponding p value against the null hypothesis of no benefitting subgroup. The p values are then aggregated for the overall test (Eq. 1)