Multicenter Kaplan-Meier Analysis — Pooled Survival Estimation Across Sites
Multicenter Kaplan-Meier Survival Analysis · Also known as: pooled Kaplan-Meier, multi-site KM analysis, multicenter survival curve analysis, KM pooled analysis
Multicenter Kaplan-Meier analysis applies the Kaplan-Meier nonparametric estimator to time-to-event data collected from two or more clinical centers. By pooling or stratifying data across sites, it estimates survival functions and compares them between treatment groups while accounting for potential center effects, enabling conclusions with greater statistical power and broader generalizability than single-center studies.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use multicenter Kaplan-Meier analysis when you have time-to-event data from two or more clinical sites and need survival estimates with greater power or generalizability than any single site can provide — typical in Phase III trials, disease registries, and collaborative cohort studies. The method is appropriate when the outcome is a clearly defined event (death, relapse, readmission) and some observations are censored. Avoid it when centers are so heterogeneous in patient mix or follow-up protocols that a pooled curve would be misleading; in those cases, report site-stratified curves separately or switch to a meta-analytic approach. Also avoid when follow-up times are very short and few events have occurred — the curve will be unreliable.
Strengths & limitations
- Increases statistical power by pooling cases across sites, enabling detection of survival differences that individual centers cannot.
- Produces familiar, easily interpreted visual survival curves that clinicians and regulators understand well.
- Nonparametric — no assumptions about the underlying distribution of survival times.
- Stratification by center can control for site-level confounding without requiring a full Cox model.
- Widely accepted by regulatory agencies (FDA, EMA) as the primary display of time-to-event data in clinical trials.
- Assumes that censoring is non-informative (independent of the true event time); if patients at some centers are more likely to drop out because they are sicker, estimates will be biased.
- The basic estimator does not adjust for patient-level covariates; a Cox model is needed for covariate-adjusted comparisons.
- Heterogeneity across centers may make a single pooled curve misleading if not assessed and addressed through stratification.
- The log-rank test used alongside KM curves has low power when hazard ratios are not proportional over time.
Frequently asked
Should I stratify by center or just pool the data?
If centers differ in baseline patient prognosis — which is common — stratifying by center in the log-rank test is more conservative and protects against false-positive results. For the survival curve itself, a pooled curve is acceptable for display if the stratified log-rank p-value indicates no significant center-by-treatment interaction. When centers are highly heterogeneous, report center-specific curves or use a meta-analytic approach.
How do I handle centers with very small sample sizes?
Small centers contribute little information and can make stratum-specific curves unstable. Common practice is to pool small centers into a 'miscellaneous' stratum, use a fixed-effects model that downweights small sites, or collapse centers by geographic region or institutional type. Sensitivity analyses with and without small centers are advisable.
Can I adjust for patient-level covariates in a Kaplan-Meier analysis?
Not directly — Kaplan-Meier is a nonparametric, unadjusted estimator. For covariate adjustment, use a Cox proportional hazards model with center as a stratification factor or as a covariate. Inverse probability weighting can produce adjusted survival curves while preserving the Kaplan-Meier visual format.
When does the log-rank test fail alongside multicenter KM analysis?
The log-rank test assumes proportional hazards — that the treatment effect is constant over time. If one treatment is beneficial early but harmful later, the log-rank may show no significant difference even when a real effect exists. Weighted log-rank tests (Fleming-Harrington) or restricted mean survival time (RMST) comparisons are better alternatives in such cases.
What software can I use for multicenter Kaplan-Meier analysis?
R (survfit and survdiff in the survival package), SAS (PROC LIFETEST with STRATA statement), Stata (sts graph with by() and strata() options), and SPSS all support stratified Kaplan-Meier analysis. The R survival and survminer packages are most commonly used in recent publications and support flexible center stratification and publication-quality curve plotting.
Sources
- Kaplan, E. L., & Meier, P. (1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53(282), 457–481. DOI: 10.2307/2281868 ↗
- Therneau, T. M., & Grambsch, P. M. (2000). Modeling Survival Data: Extending the Cox Model. Springer. ISBN: 978-0387987842
How to cite this page
ScholarGate. (2026, June 3). Multicenter Kaplan-Meier Survival Analysis. ScholarGate. https://scholargate.app/en/epidemiology/multicenter-kaplan-meier-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Cox proportional hazardsEpidemiology↔ compare
- Kaplan-Meier AnalysisEpidemiology↔ compare
- Multicenter cohort studyEpidemiology↔ compare
- Multicenter Cox proportional hazardsEpidemiology↔ compare
- Survival AnalysisResearch Statistics↔ compare