RNA Sequencing-Based Single Sample Predictors of Molecular Subtype and Risk of Recurrence for Clinical Assessment of Early-Stage Breast Cancer

ABSTRACT Background Multigene expression assays for molecular subtypes and biomarkers can aid clinical management of early invasive breast cancer (IBC). Based on RNA-sequencing we aimed to develop robust single-sample predictor (SSP) models for conventional clinical markers as well as molecular intrinsic subtype and risk of recurrence (ROR) that provide clinically relevant prognostic stratification. Methods A uniformly accrued breast cancer cohort of 7743 patients with RNA-sequencing data from fresh tissue was divided into a training set (n=5250) and a reserved test set (n=2412). We trained SSPs for PAM50 molecular subtypes and ROR assigned by nearest-centroid (NC) methods and SSPs for conventional clinical markers from histopathology data. Additionally, SSP classifications were compared with Prosigna in two external cohorts (ABiM, n=100 and OSLO2-EMIT0, n=103). Prognostic value was assessed using distant recurrence-free interval (DRFi). Results In the test set, agreement between SSP and NC classifications for PAM50 (five subtypes) and Subtype (four subtypes) was high (85%, Kappa=0.78) and very high (90%, Kappa=0.84) respectively. Accuracy for ROR risk category was high (84%, Kappa=0.75, weighted Kappa=0.90). The prognostic value for SSP and NC classification was assessed as equivalent and added clinically relevant prognostic information. Agreement for SSP and histopathology was very high or high for receptor status, while moderate and poor for Ki67 status and Nottingham histological grade, respectively. SSP concordance with Prosigna was high for subtype (OSLO 83% and ABiM 80%, Kappa=0.73 and 0.72, respectively) and moderate and high for ROR risk category (68% and 84%, Kappa=0.50 and 0.70, weighted Kappa=0.70 and 0.78). In pooled analysis, concordance between SSP and Prosigna for emulated treatment recommendation dichotomized for chemotherapy (yes vs. no) was high (85%, Kappa=0.66). In postmenopausal ER+/HER2-/N0 patients SSP application suggested changed treatment recommendations for up to 17% of patients, with nearly balanced escalation and de-escalation of chemotherapy. Conclusions Robust SSP models, mimicking histopathological variables, PAM50, and ROR classifications can be derived from RNA-sequencing that closely matches clinical tests. Agreement and DRFi analyses suggest that NC and SSP models are interchangeable on a group-level and nearly so on a patient level. Retrospective evaluation in ER+/HER2-/N0 IBC suggested that molecular testing could lead to a changed therapy recommendation for almost one-fifth of patients.