make sure a "70% risk" means 70% across groups — you will necessarily flag *different proportions* of each group, because the base rates differ. And you may have different false positive rates across groups.