- Oral presentation
- Open Access
Prediction of treatment benefit in high-dimensional cox models via gene signatures in randomized clinical trials
© Ternès et al. 2015
- Published: 16 November 2015
- Lung Cancer
- Gene Signature
- Gene Expression Data
- Randomize Clinical Trial
- Alternative Scenario
Stratified medicine seeks to identify gene signatures predicting whether a patient will benefit from a treatment. We evaluated several approaches to identify such signatures using high-dimensional Cox models in randomized clinical trials (RCT).
We investigated four approaches: penalize biomarker main effects and biomarker-by-treatment interactions using a lasso penalty (full-lasso); control of main effects by principal components or ridge penalty, and lasso on interactions (sPCA+lasso or ridge+lasso); and ‘modified covariates’ in a penalized regression model (Tian et al. 2014). We performed simulations under null and alternative scenarios by varying the sample size n, number of biomarkers H, number of true main effects or treatment-modifiers, effect sizes and correlations. We proposed two novel measures of treatment effect prediction for gene signatures: a difference in C-indices and a Wald-based interaction statistic. We used gene expression data from a RCT of adjuvant chemotherapy in non-small cell lung cancer (n=133) for illustration.
When n=500 and H=20 or 100, methods performed similarly in null scenarios apart from the full-lasso that gives poor results in presence of main effects only. In alternative scenarios: the ridge+lasso and the full-lasso predicted well the treatment benefit for future patients; the modified covariates approach performed poorly when also main effects were present. More extensive simulation results will be presented. In the lung cancer trial, the full-lasso and the ridge+lasso selected a gene signature with four and seven treatment-modifiers.
Preliminary results suggest that ridge+lasso and full-lasso are promising approaches in high-dimensional Cox models to predict the treatment benefit.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.