Quantifying treatment differences in confirmatory trials under non-proportional hazards