Abstract 2357: Gene expression profiling robustly predicts the outcome of patients diagnosed with early stage lung adenocarcinoma

• Published in: Cancer research (Chicago, Ill.) Ill.), 2014-10, Vol.74 (19_Supplement), p.2357-2357
• Main Authors: Gaston-Mathé, Yann, FERTE, CHARLES, gauthier, benoit, bateson, mathilde, planchard, david, besse, benjamin, Armand, Jean-Pierre, soria, jean-charles
• Format: Article
• Language: English
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2014-2357

Background: Although the management of metastatic lung adenocarcinoma has been profoundly modified by the identification of actionnable molecular traits, decision making for early-stage lung adenocarcinoma still relies on the tumor stage (TNM) only. Most of these patients recur within 5 years after tumor resection and are given platinum based adjuvant chemotherapy. The recent availability of multiple gene expression data and the concomitant developments of novel algorithms - such as the Hypercube algorithm - could dramatically improve such predictive challenges. Methods: Datasets were selected from public repositories (Director's challenge (DC) n=274, Rousseaux n=86, Hou n=35, JBR.21 n=31) upon the following criteria: lung adenocarcinoma, stage IA-3A, no adjuvant therapy, R0 tumor resection. To generate robust predictors of 3-year overall survival, an iterative feature selection framework (Hypercube algorithm) was applied successively on DC and Hou. Only highly predictive rules in both datasets (AUC >.7) were then included in a subsequent logistic regression model and finally assessed for external validation. The predictive performance of the models was evaluated by ROC-AUC and by Kaplan Meier curves. Results: Our best model combining gene expression (10 genes) and clinical data (TNM) was robust across three lung adenocarcinoma datasets (AUC= .85, .80 and .59). Particularly, it was able to better predict the 3 year of survival compared to the TNM-based clinical model only (100% vs. 77%). Kaplan Meier estimates of overall survival in the validation datasets were dramatically discriminated according to the high- and low-risk groups predicted by our model (all P values =.01). Interestingly, this predictive classifier was specific to lung adenocarcinoma and has no predictive value in lung squamous datasets. Conclusion: We developed a robust and highly performing gene expression predictive signature of 3-year overall survival, specific to the lung adenocarcinoma setting. These promising results have the potential to change the decision making at bedside for early stage lung adenocarcinoma patients. Citation Format: Yann Gaston-Mathé, CHARLES FERTE, benoit gauthier, mathilde bateson, david planchard, benjamin besse, Jean-Pierre Armand, jean-charles soria. Gene expression profiling robustly predicts the outcome of patients diagnosed with early stage lung adenocarcinoma. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer