Kaplan-Meier Estimator
Kaplan-Meier Survival Function Estimator · Also known as: KM estimator, product-limit estimator, Kaplan-Meier curve, survival curve estimator
The Kaplan-Meier estimator is a nonparametric method for estimating the survival function S(t) — the probability that an individual survives beyond time t — from data that include censored observations. Introduced by Edward L. Kaplan and Paul Meier in their landmark 1958 JASA paper, it is the standard first step in any survival analysis and is among the most-cited statistical methods in biomedical research.
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 the Kaplan-Meier estimator whenever the outcome of interest is time to a single event (death, relapse, failure, recovery) and some observations are right-censored — that is, the event had not yet occurred when the individual left the study. Key assumptions are: (1) censoring is non-informative (the reason a subject is censored is unrelated to their prognosis); (2) survival probabilities are the same for subjects recruited at different calendar times (often assessed by staggered enrollment plots); and (3) the event times are precisely measured. The estimator is purely descriptive and does not adjust for covariates; use the Cox proportional-hazards model for covariate-adjusted inference.
Strengths & limitations
- Requires no distributional assumption about the underlying survival time — valid even when the true hazard function is complex or irregular.
- Handles right-censored data correctly, making full use of partial information from censored subjects.
- Produces an immediately interpretable graphical output (the KM curve) that communicates survival patterns at a glance.
- Median survival time and its confidence interval are directly readable from the estimated curve.
- Universally available in all major statistical software packages and accepted as standard reporting in clinical journals.
- Purely descriptive — cannot adjust for confounding covariates or compare groups while controlling for other variables.
- Precision deteriorates at the tail of the curve where few subjects remain at risk, making late time-point estimates unreliable.
- Assumes non-informative censoring; if patients drop out because they are deteriorating, estimates will be biased.
- Cannot accommodate time-varying covariates or competing risks without extension to more specialised models.
Frequently asked
What is censoring and why does it matter?
A censored observation is one where the event has not yet occurred by the time the subject leaves the study — because the study ended, the subject was lost to follow-up, or they withdrew. Ignoring censored subjects (or treating them as events) biases the estimate. Kaplan-Meier handles censoring correctly by including each censored subject in the risk set up to the moment they are censored and then removing them, so they contribute information for the period they were observed.
How do I compare survival curves between two groups?
Plot the two KM curves separately and test for a difference using the log-rank test, which is the standard nonparametric companion to the KM estimator. The log-rank test weights each event time equally and is most powerful when the hazards are proportional. If proportional hazards do not hold (curves cross), the weighted Fleming-Harrington test may be more appropriate.
When should I use Cox regression instead of Kaplan-Meier?
Use Kaplan-Meier for unadjusted description and visualization of survival in one or more groups. Use Cox proportional-hazards regression when you need to adjust for covariates, estimate hazard ratios, or model continuous predictors. In practice, KM curves and the log-rank test are used first for descriptive and univariate analysis, followed by Cox regression for multivariable inference.
What is the median survival and how is it read from the curve?
The median survival time is the time t at which Ŝ(t) first drops to or below 0.5 — that is, the time by which half the subjects have experienced the event. It is read by drawing a horizontal line at S = 0.5 across the KM curve and finding where it intersects the step function. Its confidence interval is obtained from the intersection of that horizontal line with the upper and lower confidence bands of the curve.
Sources
- Kaplan, E. L., & Meier, P. (1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53(282), 457–481. DOI: 10.1080/01621459.1958.10501452 ↗
- Collett, D. (2015). Modelling Survival Data in Medical Research (3rd ed.). CRC Press. ISBN: 978-1439856789
- Kleinbaum, D. G., & Klein, M. (2012). Survival Analysis: A Self-Learning Text (3rd ed.). Springer. ISBN: 978-1441966452
How to cite this page
ScholarGate. (2026, June 3). Kaplan-Meier Survival Function Estimator. ScholarGate. https://scholargate.app/en/statistics/kaplan-meier-estimator
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
- Log-Rank TestSurvival↔ compare