A multivariable prognostic score to guide systemic therapy in early-stage HER2-positive breast cancer: a retrospective study with an external evaluation

BACKGROUND: In early-stage HER2-positive breast cancer, escalation or de-escalation of systemic therapy is a controversial topic. As an aid to treatment decisions, we aimed to develop a prognostic assay that integrates multiple data types for predicting survival outcome in patients with newly diagnosed HER2-positive breast cancer.METHODS: We derived a combined prognostic model using retrospective clinical-pathological data on stromal tumour-infiltrating lymphocytes, PAM50 subtypes, and expression of 55 genes obtained from patients who participated in the Short-HER phase 3 trial. The trial enrolled patients with newly diagnosed, node-positive, HER2-positive breast cancer or, if node negative, with at least one risk factor (ie, tumour size >2 cm, histological grade 3, lymphovascular invasion, Ki67 >20%, age ≤35 years, or hormone receptor negativity), and randomly assigned them to adjuvant anthracycline plus taxane-based combinations with either 9 weeks or 1 year of trastuzumab. Trastuzumab was administered intravenously every 3 weeks (8 mg/kg loading dose at first cycle, and 6 mg/kg thereafter) for 18 doses or weekly (4 mg/kg loading dose in the first week, and 2 mg/kg thereafter) for 9 weeks, starting concomitantly with the first taxane dose. Median follow-up was 91·4 months (IQR 75·1-105·6). The primary objective of our study was to derive and evaluate a combined prognostic score associated with distant metastasis-free survival (the time between randomisation and distant recurrence or death before recurrence), an exploratory endpoint in Short-HER. Patient samples in the training dataset were split into a training set (n=290) and a testing set (n=145), balancing for event and treatment group. The training set was further stratified into 100 iterations of Monte-Carlo cross validation (MCCV). Cox proportional hazard models were fit to MCCV training samples using Elastic-Net. A maximum of 92 features were assessed. The final prognostic model was evaluated in an independent combined dataset of 267 patients with early-stage HER2-positive breast cancer treated with different neoadjuvant and adjuvant anti-HER2-based combinations and from four other studies (PAMELA, CHER-LOB, Hospital Clinic, and Padova) with disease-free survival outcome data.FINDINGS: From Short-HER, data from 435 (35%) of 1254 patients for tumour size (T1 vs rest), nodal status (N0 vs rest), number of tumour-infiltrating lymphocytes (continuous variable), subtype (HER2-enriched and basal-like vs rest), and 13 genes composed the final model (named HER2DX). HER2DX was significantly associated with distant metastasis-free survival as a continuous variable (p