Restricted Mean Survival Time (RMST) calculator

Calculate Restricted Mean Survival Time (RMST) for survival data. Compare areas under KM curves up to a horizon tau.

What it does

The average event-free time up to a chosen horizon — the area under the survival curve out to that point. Expressed in time units, not as a ratio.

When to use it

When you want an interpretable difference ("3.2 months longer on average over two years"), or when the proportional-hazards test on your Cox model fails and a single hazard ratio is no longer defensible. RMST makes no proportionality assumption at all, which is exactly why it is the recommended fallback.

Cautions

• The horizon must be chosen in advance and reported. Different horizons give different answers, and picking one after seeing the curves is a form of p-hacking.
• The horizon cannot exceed your follow-up: beyond the data, the area is not estimable.

Alternatives

• Cox proportional hazards — hazards are proportional and you need covariate adjustment
• Kaplan-Meier median survival — the curves drop below 50% and a median is well defined

How to read the output

Restricted mean survival time (RMST)

The average event-free time accumulated up to your horizon τ — the area under the survival curve out to that point, in your time column's own units. 'Patients in this arm were event-free for an average of 8.4 of the first 12 months' is a sentence a reader can act on without knowing what a hazard is. It is restricted to τ and says nothing beyond it. Two arms with the same RMST at 12 months can diverge sharply afterwards.

Comparing the groups

The difference between two groups' restricted means is a difference in event-free time — valid whether or not hazards are proportional, which is what makes it the natural fallback when the Schoenfeld test fails or the curves cross. This tab reports one RMST per group and no difference. Subtract them yourself for the point estimate, and be explicit that it comes without a confidence interval — an unqualified difference reads as an estimate that has been tested when it has not.

Choosing τ

τ must lie within follow-up for every group; the usual choice is the smallest of the groups' largest observed times. Choosing τ after seeing which value separates the groups is the same error as choosing a cut point to maximise significance. Fix it in advance and report it.

How this calculator is validated · Which statistical test should I use?