On the proportional hazards assumption