Survival prediction using gene expression data: A review and comparison

Abstract

Knowledge of transcription of the human genome might greatly enhance our understanding of cancer. In particular, gene expression may be used to predict the survival of cancer patients. Microarray data are characterized by their high-dimensionality: the number of covariates ( p ∼ 1000 ) greatly exceeds the number of samples ( n ∼ 100 ), which is a considerable challenge in the context of survival prediction. An inventory of methods that have been used to model survival using gene expression is given. These methods are critically reviewed and compared in a qualitative way. Next, these methods are applied to three real-life data sets for a quantitative comparison. The choice of the evaluation measure of predictive performance is crucial for the selection of the best method. Depending on the evaluation measure, either the L 2 -penalized Cox regression or the random forest ensemble method yields the best survival time prediction using the considered gene expression data sets. Consensus on the best evaluation measure of predictive performance is needed.