Cumulative Incidence rather than Kaplan Meier analysis should be used for competing non fatal event analysis

Supplementary files and Presentations
Ludbrook J, Royse AG.  Analysing longitudinal clinical studies: principles, practice and pitfalls of Kaplan-Meier plots. ANZ J Surgery 2008; 78: 204-210 .
DOI: 10.1111/j.1445-2197.2007.04405.x
Request a Reprint

There are two Supplementary files that should be downloaded:
K-M and N-A survival, incidence and hazard function for single adverse events and competing risks [Ludbrooksupp1.doc]
Melanoma Excel Spreadsheet [Ludbrooksupp2.xls]

Download the PowerPoint Presentation or see a Fash Video of the presentation 

Contacts
Prof John Ludbrook :    ludbrook@bigpond.net.au 
A/Prof Alistair Royse :  Alistair.Royse@unimelb.edu.au  

Abstract
Surgeons should use cumulative incidence (Actual) rather than kaplan-meier (Actuarial) statistical analysis for non fatal competing event analysis

John Ludbrook, Alistair Royse
Cardiovascular Therapeutics Unit, The University of Melbourne, Melbourne, Victoria

Introduction
Analysis of death as a single endpoint is usually performed using Kaplan-Meier actuarial analysis (KM). Where there are competing events such as death or relapse; or death or graft failure, then KM will falsely inflate the failure rate.

Methods
An example illustrates the two key failings of KM which are mostly addressed using Cumulative Incidence (CI or Actual) analysis. The first is that patients that are dead are censored, and KM assumes they will have the same rate of occurrence of the non fatal event as those where the outcome remains unknown – overestimating the probability of such events. Second, KM fits the data where there is an exponential decay in observed events. But when the curve of outcome is not approximating an exponential decay curve, then early analysis will falsely predict late outcome by considering the early part of the curve which may not accurately reflect the late part of the curve of actual events.

Discussion
Surgeons should not use Kaplan-Meier analysis for non fatal competing risk events; but rather should use the Cumulative Incidence (actual) analysis method.