Detecting critical treatment effect bias in small subgroups