Restricted mean survival time averages survival over a window [0, τ] and compares groups by difference or ratio, an alternative to the hazard ratio when proportional hazards fails.
RMST is the area under the survival curve from 0 to a chosen truncation τ — literally the expected time alive within the window. Unlike HR, which is a multiplicative effect on an unobservable hazard, RMST is on the time scale, so a 'difference of 3.5 months' is directly interpretable to patients.
RMST does not require proportional hazards. When K-M curves cross or the hazards diverge over time — situations where HR misleads — RMST gives an honest summary of the cumulative survival experience over the chosen window.
Pick τ before looking at the data. Common choices: the time horizon of clinical interest, the last event time where both arms still have substantial follow-up, or τ where Ŝ in each arm exceeds some minimum (say 10% to keep estimates stable).
Time + event + arm (binary) + optional τ (defaults to the smaller of the two arms' last event times).
RMST per arm with CI, between-arm difference + ratio with p-values.
Difference in RMST in months is on the time scale and is interpretable as 'patients in arm A gained X months over τ years follow-up'.
Ratio of RMST (e.g. 1.15) is a multiplicative summary of the relative life-years; useful for relative-effectiveness framings.