Development and validation of the new HER2DX assay for predicting pathological response and survival outcome in early-stage HER2-positive breast cancer

Abstract

Background: Both clinical and genomic data independently predict survival and treatment response in early-stage HER2-positive breast cancer. Here we present the development and validation of a new HER2DX risk score, and a new HER2DX pathological complete response (pCR) score, both based on a 27-gene expression plus clinical feature-based classifier.

Methods: HER2DX is a supervised learning algorithm incorporating tumour size, nodal staging, and 4 gene expression signatures tracking immune infiltration, tumour cell proliferation, luminal differentiation, and the expression of the HER2 amplicon, into a single score. 434 HER2-positive tumours from the Short-HER trial were used to train a prognostic risk model; 268 cases from an independent cohort were used to verify the accuracy of the HER2DX risk score. In addition, 116 cases treated with neoadjuvant anti-HER2-based chemotherapy were used to train a predictive model of pathological complete response (pCR); two independent cohorts of 91 and 67 cases were used to verify the accuracy of the HER2DX pCR likelihood score. Five publicly available independent datasets with >1,000 patients with early-stage HER2-positive disease were also analysed.

Findings: In Short-HER, HER2DX variables were associated with good risk outcomes (i.e., immune, and luminal) and poor risk outcomes (i.e., proliferation, and tumour and nodal staging). In an independent cohort, continuous HER2DX risk score was significantly associated with disease-free survival (DFS) (p=0·002); the 5-year DFS in the low-risk group was 97·4% (94·4-100·0%). For the neoadjuvant pCR predictor training cohort, HER2DX variables were associated with pCR (i.e., immune, proliferation and HER2 amplicon) and non-pCR (i.e., luminal, and tumour and nodal staging). In both independent test set cohorts, continuous HER2DX pCR likelihood score was significantly associated with pCR (p<0·0001). A weak negative correlation was found between the HER2DX risk score versus the pCR score (correlation coefficient -0·19).

Interpretation: The two HER2DX tests provide accurate estimates of the risk of recurrence, and the likelihood to achieve a pCR, in early-stage HER2-positive breast cancer.

Funding: This study received funding from Reveal Genomics, IDIBAPS and the University of Padova.

Keywords: De-escalation; Gene expression; HER2-positive breast cancer; HER2DX; Immune; Neoadjuvant; Pathological complete response; Prognosis; Risk of relapse.

Figure 1

Summary of the different cohorts of patients evaluated during HER2DX development and validation.

Figure 2

Survival outcomes of HER2DX low- and high-risk groups in early-stage HER2-positive breast cancer. (a) DRFS in Short-HER dataset (n=434); (b) DFS in Short-HER dataset (n=434); (c) OS in Short-HER dataset (n=434); (d) DFS in an independent combined validation dataset (n=268).

Figure 3

Summary of the variables included in the HER2DX assay and their association with each clinical endpoint. The type of association between a variable and each clinical endpoint is represented in different colours, where red means that a high score of that variable is associated with worse survival outcome or a lower likelihood of achieving a pCR, and blue means that a high score of that variable is associated with better survival outcome or a higher likelihood of achieving a pCR. Grey means no association of the variable with the clinical endpoint.