RMST or a hazard ratio?

Restricted mean survival time is the area under the survival curve up to a chosen time τ. It is the average time alive within the first τ years, and the difference in RMST between two arms is the average time gained — in months or years, not on a ratio scale.

It has three properties a hazard ratio does not. It requires no assumption about proportional hazards, because it summarises the curves rather than their ratio. It is collapsible, so a marginal RMST difference is the average of the stratum-specific ones and it can be compared across analyses adjusted differently. And it is in units anyone can interpret without training, which matters when the audience is a committee rather than a journal.

The cost is that you must choose τ, and the answer depends on it. A τ beyond the follow-up of the data cannot be estimated without extrapolating, and a τ chosen after seeing the curves is a choice a reviewer will ask about. Pick it in advance on clinical grounds — the horizon the decision actually cares about — and report the sensitivity of the conclusion to it.

In practice the two are complements more often than rivals. A hazard ratio is the conventional summary and will be asked for. RMST says how much time the effect is worth, survives non-proportionality, and does not change meaning when the follow-up does. Reporting both, with the horizon named, is harder to argue with than either alone.

MethodSurvival analysis