Choose and interpret a restricted mean survival time

Define the horizon before comparing groups

Restricted mean survival time, or RMST, is the area under a survival curve up to a specified time horizon. It summarises expected event-free or survival time within that restricted period, depending on the event definition. The horizon is part of the estimand, not a plotting preference.

The survRM2 package authors' vignette describes RMST comparisons as alternatives to hazard-ratio summaries. RMST does not require a proportional-hazards interpretation, but valid estimation still depends on the study design, censoring assumptions and adequate follow-up.

Check support for the selected time

Choose a clinically meaningful horizon supported by follow-up in both groups. Inspect the numbers at risk and censoring near that point. Do not select the horizon after searching for the largest or most significant difference. If several horizons are scientifically important, define their roles and address multiplicity or exploratory interpretation.

Specify the event, time origin and analysis population. A composite event changes what “event-free time” means. Adjusted and unadjusted RMST comparisons also require clear descriptions of the target population and method. Keep the time units consistent in the dataset, plot and results paragraph.

Hypothetical example

Consider an invented step survival curve that equals 1 from zero to two months, 0.8 from two to five months and 0.5 from five to six months. Its RMST through six months is 2 × 1 + 3 × 0.8 + 1 × 0.5 = 4.9 months.

If another fictional group has RMST 4.2 months at the same horizon, the difference is 0.7 months of expected survival within the first six months. It is not a claim of 0.7 months added to unrestricted lifetime survival.

Report the quantity directly

Present the horizon, each group's RMST and the contrast with a confidence interval. Explain why the horizon was chosen and whether it was prespecified. Include the survival curves and relevant follow-up information so readers can see the data supporting the area calculation.

Use the survival risk-set planning worksheet to record support near the horizon. If follow-up is too sparse, shorten the justified horizon or explain the modelling assumptions needed for extrapolation. Do not conceal unsupported tail behaviour behind a precise-looking area estimate.

Source verification: 6 November 2026. Examples are hypothetical.