Kaplan-Meier estimator
The Kaplan-Meier estimator describes the chance of staying free of a defined event over time, while using incomplete follow-up.
Fit
When to use it, and when not to
Use it when
- You know when follow-up starts and when each event or loss to follow-up occurs.
- You want an unadjusted survival curve or survival at a stated time.
Not when
- Death prevents a nonfatal outcome. Estimate cumulative incidence with competing events instead. Read: Choosing an analysis for competing risks
- You need to adjust for measured factors. Consider Cox regression when its assumptions fit. Read: Cox proportional hazards regression
Before you start
What you need
- A common startDefine eligibility and the start of follow-up before looking at outcomes.
- One event and its timeoutcomeTypeSay whether the event is death, relapse or a defined first composite event.
- Event or censoringCensoring means observation stopped before the event was seen. Keep its date and reason.
- Within each reported group, loss to follow-up is unrelated to future event times.How to check: Record why follow-up ended. Assess how results change if those lost have a worse outlook.
- Different people supply independent observations.How to check: For patients grouped within clinics or families, use inference that accounts for those groups.
- Event times are known with enough precision.How to check: If events are known only between visits, use an interval-censoring method.
Plain language
How it works
At each event time, count who is still at risk just before the event. Multiply the previous survival estimate by the share who survive that time. The curve steps down at events and stays flat between them. Censoring removes someone from later risk sets without making the curve drop. This product-limit construction is the estimator discussed by Satten and Datta.
Report survival at stated times, confidence intervals and numbers at risk. Use a survival-specific interval, with a stated variance method such as Greenwood and a suitable transformation. Borkowf studies variance estimation; sparse late follow-up needs care. A flat tail with few people is weak evidence. The separate log-rank test article explains a group comparison.
Worked example
A worked example
- Setting
- Ten patients are followed for death through month 6.
- Question
- What is estimated survival at month 6?
- Approach
- At month 2, two die among ten at risk. At month 4, three die among eight. Two survivors leave at month 5.
- Result
- Survival at month 6 is (8/10) × (5/8) = 0.50, or 50 in 100 (50%).
- What it means
- Censoring at month 5 does not lower the curve. It reduces the number still observed, so later estimates have less support.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
Censoring must be uninformative within each group. A Kaplan-Meier curve that censors a competing death or another event overstates cumulative incidence when read as one minus survival.
- Counting a censored person as known to survive to the end.Instead: Use their follow-up only until censoring.
- Calling the last observed event time the median.Instead: The median is the first time estimated survival is 0.50 or less. It may never be reached.
Inside Mallard
How Mallard uses this
Learn search lists Kaplan-Meier estimator. The method registry links this method to its published analysis article and code.
kaplan-meier-logrank).analysis.kaplan_meier_estimator, the id a plan cue records.Work it out
Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- The Kaplan-Meier Estimator as an Inverse-Probability-of-Censoring Weighted Average. (opens in a new tab) · Am Stat · doi 10.1198/000313001317098185
- A simple hybrid variance estimator for the Kaplan-Meier survival function. (opens in a new tab) · Stat Med · doi 10.1002/sim.1960