Ørnulf Borgan

Ørnulf Borgan, Matematisk Institutt, Universitetet i Oslo, skal snakke om

Assessment of evaluation criteria for survival prediction from genomic data

 

Sammendrag

When building a survival prediction model based on high-dimensional genomic data, one has to adopt a proper regularization method. Currently, however, there is no consensus in the literature on which criterion one should use to assess the performance of such survival prediction models. When comparing survival prediction models, it is therefore important to know whether the choice of evaluation criterion may affect the conclusions made as to which model performs best. In the talk I will present four commonly used evaluation criteria for assessing the performance of survival prediction models: the log-rank test for two groups, the area under the time-dependent ROC curve, an R^2-measure based on Cox's partial likelihood, and an R^2-measure based on the Brier score. The four criteria will be compared according to how they rank six widely used regularization methods for three publicly available microarray gene expression data sets.

The talk is based on joint work with Hege M. Bøvelstad

Published Mar. 29, 2011 8:05 AM - Last modified Aug. 22, 2011 12:37 PM