Loading...
Show contents
Table of Contents

Kaplan-Meier (mixture model) plots

A Kaplan-Meier plot [Kaplan & Meier 1958 J. Amer. Statist. Assoc. 53:457] shows the survival of patient subgroups over time. The Gaussian mixture modelling procedure splits patients into a number of groups, or clusters. A Kaplan-Meier plot is then generated for each marker where more than one group of patients was found in the mixture modelling.

The difference in survival between the groups is tested for significance using the log-rank (Mantel-Cox) test [Mantel 1966 Cancer Chemother. Rep. 50:163].

Available for: continuous scoring

[Top]Viewing the results

See accessing and interpreting survival analysis results.

Proteins with a single cluster

The number of clusters used to group a protein's scores is determined using the Bayesian Information Criterion (BIC) - a statistical trade off between model likelihood (goodness-of-fit) with complexity (the number of parameters used to fit the model). More complex models will always deliver an equal or better fit to the data than simpler ones, however fitting more complex models can lead to overfitting.

The BIC selection process may yield protein models consisting of a single cluster, represented by a single Gaussian distribution. Proteins with a single cluster cannot be used for Kaplan-Meier analysis, as this requires multiple clusters with which to compare survival.

[Top]Example output

Kaplan-Meier example based on score grouping from mixture modelling

Figure 1: Example Kaplan-Meier plot using groups based on output from Gaussian mixture modelling.

Back to top