A practical guide to understanding Kaplan-Meier curves

Abstract

In 1958, Edward L. Kaplan and Paul Meier collaborated to publish a seminal paper on how to deal with incomplete observations. Subsequently, the Kaplan-Meier curves and estimates of survival data have become a familiar way of dealing with differing survival times (times-to-event), especially when not all the subjects continue in the study. "Survival" times need not relate to actual survival with death being the event; the "event" may be any event of interest. Kaplan-Meier analyses are also used in nonmedical disciplines. The purpose of this article is to explain how Kaplan-Meier curves are generated and analyzed. Throughout this article, we will discuss Kaplan-Meier estimates in the context of "survival" before the event of interest. Two small groups of hypothetical data are used as examples in order for the reader to clearly see how the process works. These examples also illustrate the crucially important point that comparative analysis depends upon the whole curve and not upon isolated points.

FIGURE 1

This figure illustrates the component parts of a clinical trial. Each trial may differ; however, all require a start up period, and end of accrual, and an end to the trial.

FIGURE 2

This figure illustrates subjects entering a trial and ending at different times. S, M, and L indicate a short, medium, and long serial times. The solid circle represents an event occurrence and the open circles represent censoring.

FIGURE 3

This is a Kaplan-Meier curve generated by SigmaPlot (SigmaPlot 11.0 for Windows, Systat Software Inc., San Jose, CA) from the data used as an example in Tables IA, B, C. Censoring is indicated by the black dot tic mark.