The restricted mean survival time, commonly referred to as RMST, summarizes the event-free or survival time up to a pre-specified time horizon. It corresponds to the area under the Kaplan-Meier curve up to that point and is expressed in a directly interpretable time unit. The comparison of two groups can thus be reported as time gained or time lost within a clearly defined observation window.
Estimation up to the Time Horizon
From the observed event and censoring times, the survival curve for each group is first estimated. The RMST condenses the entire curve progression up to the chosen horizon into a single value. For overall survival, this represents the average time that participants survive within this window; for a composite time-to-event endpoint, it refers to the defined event-free period.
The comparison is typically the difference between the two RMST values. It does not indicate when an event occurs, but rather how many days or months on average are gained or lost within the defined window. The choice of horizon must be specified in the protocol, as a later time point addresses a different clinical question and may be based on substantially less information.
Requirements and Interpretation
The method requires a clearly defined time axis, event definition, and censoring rule. As with Kaplan-Meier estimates, it must be assessed whether censoring is plausibly independent of the subsequent event risk. A late time window should be supported by sufficient observation in both groups; otherwise, the estimate will be heavily influenced by the few individuals still at risk.
RMST can reveal differences that occur only in an early or late study phase. It can also be calculated when survival curves cross, as it does not require a constant relative difference in hazards over time. However, the summary time value may average opposing curve segments; the associated curves therefore remain important for clinical interpretation.
Distinction from Hazard Ratio
The RMST is an alternative to the hazard ratio. A hazard ratio describes a ratio of event rates under a model and is frequently interpreted under the assumption of proportional hazards, i.e., a constant relative effect size over time. The RMST does not require this assumption and presents the effect as a temporal difference up to a specific horizon.
The existing entry hazard-ratio addresses the relative measure, kaplan-meier-analyse the curve representation, survival-analyse and time-to-event the overarching methods. The restricted mean survival time is a distinct effect measure. It does not automatically replace the primary planned comparison, but can characterize the effect as a supplementary or pre-specified target parameter.
The RMST does not require a proportional hazards model, but is not a model-free shortcut for every survival analysis. The uncertainty of the estimated difference and the censoring assumption must be reported. With very different curve shapes, the time difference facilitates communication, while the graphical curve explains in which section of the horizon this difference arose.
Relevance for clinical trials
In oncology trials, endpoint definition, imaging timepoints, censoring rules, and the clinical horizon must align. The EMA identifies overall survival, progression-free survival, and disease-free survival as potential efficacy endpoints and requires prospective clarification of methodological questions for PFS and DFS studies. For an RMST analysis, it must therefore be documented which event time and which censoring enter the survival curve.
Full-service CROs such as Mediconomics support the specification of protocol-compliant time-to-event definitions, the creation of analysis datasets, and the validated evaluation of Kaplan-Meier curves and RMST differences. Clinical assessment and medical writing can explain the chosen horizon and the temporal significance of the difference without equating the measure with a constant hazard ratio.
When RMST is planned as the primary analysis, data cutoff, time origin, and management of subsequently identified events must be as clearly defined as for any other primary survival evaluation. The comparison will only be reproducible if the same administrative observation closure and the same horizon apply to both groups. Differences in recruitment duration must not inadvertently result in different evaluation windows.
Frequently Asked Questions (FAQ)
Can RMST only be used for overall survival?
No. It can be used for any appropriately defined time-to-event endpoint, such as progression-free survival. The clinical interpretation must then explicitly state which event terminates event-free status and up to which time point time is accumulated.
Why must the time horizon be determined in advance?
A time point selected after reviewing the curves could favor the observed effect. Pre-specification ties the measure to the study question and enables a comparable evaluation across all groups.
Can a positive RMST difference replace a non-significant hazard ratio?
The measures address differently formulated questions. Whether RMST can be used for a confirmatory conclusion depends on its role in the protocol, the multiplicity concept, and the intended decision rule.
Regulatory References
- ICH E9(R1), Addendum on Estimands and Sensitivity Analyses – assigns the population-level summary to the estimand.
- EMA/CHMP/205/95 Rev.6, Guideline on the Clinical Evaluation of Anticancer Medicinal Products – addresses the selection of oncology efficacy endpoints.
- EMA/CHMP/27994/2008/Rev.1, Appendix 1 on PFS and DFS – explains prospective endpoint and censoring questions.
