A prognostic model integrating clinical data and gene signatures in phase III neoadjuvant trial CALGB 40601 (Alliance)

Abstract

Background: Clinical and pathological features, as well as genomic IgG signature, and correlation to HER2-enriched were independently prognostic in CALGB 40601 (Alliance), a neoadjuvant phase III trial of weekly paclitaxel (T) and trastuzumab (H) with or without lapatinib (L) for HER2-positive breast cancer. Our aim was to build a prognostic model for HER2+ patients receiving trastuzumab using clinical, pathological and genomic signatures. Methods: We performed a comprehensive analyses on event-free survival, integrating clinic-pathological information (stage, treatment [dual vs single HER2 drugs], pathological complete response [pCR], and immunohistochemical hormonal receptor status) as well as 509 gene expression signatures using mRNA-sequencing data from 265 pre-treatment tumor samples. Excluding 6 Normal-like tumor samples, the dataset consisted of 259 patients including 82 HER2-enriched, 14 Basal-like, 81 Luminal A, and 82 Luminal B. We first analyzed univariate Cox regression analysis, and then built multivariate Cox models using the Elastic net for predicting event-free survival (EFS). Given the limitations of this sample size, we performed 10-fold cross validation analysis and identified the most commonly selected features. Results: Using the Elastic Net regression, we selected the best alpha and lambda. and using these combinations, the Elastic models comprised 70-80 features. Among these, 20 features were always selected including treatment, pCR, signatures of IgG, Helper T cell, response to chemotherapy, CD44+cells as good prognostic factors, and a signature of lymph vessel as a poor prognostic factor. Those features were also significantly associated with EFS in univariate analysis. Conclusions: Our predictive model of EFS in HER2+ patients receiving trastuzumab regimen included key clinical and genomic features. These models need to be validated in independent datasets.