Statistical methods applied to omics data: predicting response to neoadjuvant therapy in breast cancer

Abstract

Purpose of review: Omics technologies have become an essential part of clinical trials in oncology to provide a better understanding of molecular mechanisms and to unveil therapeutic targets. Standard statistical methods often fail in the high-dimensional setting. Therefore, an adequate modelling of the omics data is needed in order to identify 'target' genes of interest.

Recent findings: Several genes or gene signatures have been identified to predict the response to neoadjuvant therapies in breast cancer trials. We first reviewed statistical methods used to identify genes in 13 recent publications. Most of these studies had a small sample size (median: 42 patients) and were nonrandomized. We then focused on some popular methods - especially the so-called penalized methods used by three of the reviewed articles - and on the more recent methods proposed to predict causal estimates from observational data. We finally illustrated these methods in a nonrandomized neoadjuvant phase II trial of letrozole in estrogen receptor-positive breast cancer patients.

Summary: The review highlighted small sample sizes, few randomized trials and a large panel of statistical methods used in this setting. In our illustrated neoadjuvant example, causal inference methods did not outperform the penalized methods.