RNA sequencing-based single sample predictors of molecular subtype and risk of recurrence for clinical assessment of early-stage breast cancer

Abstract

Multigene assays for molecular subtypes and biomarkers can aid management of early invasive breast cancer. Using RNA-sequencing we aimed to develop single-sample predictor (SSP) models for clinical markers, subtypes, and risk of recurrence (ROR). A cohort of 7743 patients was divided into training and test set. We trained SSPs for subtypes and ROR assigned by nearest-centroid (NC) methods and SSPs for biomarkers from histopathology. 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. Agreement between SSP and NC for PAM50 (five subtypes) was high (85%, Kappa = 0.78) for Subtype (four subtypes) very high (90%, Kappa = 0.84) and for ROR risk category high (84%, Kappa = 0.75, weighted Kappa = 0.90). Prognostic value was assessed as equivalent and clinically relevant. Agreement with histopathology was very high or high for receptor status, while moderate for Ki67 status and poor for Nottingham histological grade. SSP and Prosigna concordance was high for subtype (OSLO-EMIT0 83%, Kappa = 0.73 and ABiM 80%, Kappa = 0.72) and moderate and high for ROR risk category (68 and 84%, Kappa = 0.50 and 0.70, weighted Kappa = 0.70 and 0.78). Pooled concordance for emulated treatment recommendation dichotomized for chemotherapy was high (85%, Kappa = 0.66). Retrospective evaluation suggested that SSP application could change chemotherapy recommendations for up to 17% of postmenopausal ER+/HER2-/N0 patients with balanced escalation and de-escalation. Results suggest that NC and SSP models are interchangeable on a group-level and nearly so on a patient level and that SSP models can be derived to closely match clinical tests.

Conflict of interest statement

All authors declare no competing financial interests related to the study, and for all but A.E., H.L., and L.H.S. also no competing non-financial interests. A.E., H.L., and L.H.S. declare the following competing non-financial interests for this study: A.E. has received speakers’ honoraria from Novartis, Amgen, Roche, and advisory board fees from Roche. H.L. reports honoraria for lecturing from Astra-Zeneca, Novartis, Lilly, and Seagen, and working in advisory boards of Pfizer, Novartis, Daiichi, MSD, Amgen, and Pierre Fabre, and has received research support from Roche. L.H.S. has employment and ownership interest (including stock and patents) in SAGA Diagnostics AB.

© 2022. The Author(s).

Figures

Fig. 1. Outline of the study.

a Study cohorts: SCAN-B, ABiM, and OSLO2-EMIT0. Available data types and usage outlined. FT fresh-frozen tissue, FFPE formalin-fixed paraffin-embedded tissue. b Scheme for development and validation of SSP models for molecular subtypes and ROR versus NCN equivalents. Scheme outlines created SSP models and their usage in different cohorts. NCN and SSP 4-class subtype models include Basal-like, HER2-enriched, Luminal A and Luminal B subtypes. Binary SSP-ROR and NCN-ROR risk classes were created similar as described by Bartlett et al.. An emulated 3-group SSP-based treatment recommendation (SSP ROR ETR) was created based on published Norwegian guidelines for Prosigna usage and applied to relevant SCAN-B patients based on guidelines. c Scheme for development and validation of SSPs for clinical markers: ER, PR, HER2, Ki67, and NHG. Scheme outlines created SSP models and their usage in different cohorts.

Fig. 2. Validation of SSP classifications against NCN classifications in the independent test set of early breast cancer.

a Agreement chart and confusion matrix comparing SSP classifications (x-axis/columns) with NCN classifications (y-axis/rows) for PAM50 (five subtypes) and b Subtype (four subtypes). c Scatterplot of binned NCN-ROR values (y-axis) versus SSP-ROR (x-axis). d Boxplot of NCN-ROR values (y-axis) by SSP-ROR (x-axis). e Distributions of NCN-ROR values and f SSP-ROR values by SSP-PAM50 (five subtypes). g Distributions of NCN-ROR values or h SSP-ROR values, by SSP-Subtype (four subtypes) for ER+/HER2- breast cancer classified as Luminal A or Luminal B. i Agreement chart and confusion matrix comparing SSP classification (x-axis/columns) with NCN classification (y-axis/rows) for ROR risk classification and j emulated treatment recommendation. Boxplot elements: center line (median), bounds of box (upper and lower quartiles), whiskers (1.5x interquartile range).

Fig. 3. Validation of SSP models for clinical markers in the independent population-based test set of early breast cancer.

Agreement chart and confusion matrix comparing the SSP classifications (x-axis/columns) with clinical histopathology status (y-axis/rows) for a ER status, b PR status, c HER2 status using a general SSP model, d HER2 status using a SSP model specific for ER status, e Ki67-status, f NHG.

Fig. 4. Assessment of prognostic value of SSP stratification and NCN stratification.

Comparison of SSP and NCN classifications in the independent population-based test set by assessment of prognostic value. Kaplan-Meier plots for molecular subtype with five groups (PAM50) or four groups (Subtype) using DRFi as clinical endpoint: a PAM50 by NCN (left) and SSP (right), b Subtype by NCN (left) and SSP (right). c Cox regression analysis using DRFi as endpoint in the test set restricted to patients with ER+/HER2-/N0 disease diagnosed over 50 years of age that only received endocrine adjuvant treatment (n = 772). Test and reference group is specified on the left. Hazard ratios and 95% confidence interval ranges from univariable analysis (left forest plot) and multivariable analysis (right forest plot) with tumor size, age at diagnosis, and NHG as covariates. Kaplan–Meier plots for stratification of ER+/HER2-/N0 disease diagnosed over 50 that only received endocrine adjuvant treatment in the test set by NCN (left in each panel) and SSP (right in each panel) for: d PAM50 subtype, e Subtype, f ROR risk classification, and g the two-group ROR stratification according to Bartlett et al.. Error bars correspond to 95% confidence interval.

Fig. 5. SSP classifications for Subtype and ROR risk category in early-stage breast cancer and cross-comparison with administered systemic treatment.

The basis for comparisons is the 6660-patient follow-up cohort (Table 1). Summarized proportions are shown on the right side of bar graphs. The first and last year of enrollment (2010 and 2018) are not full calendar years and therefore include notably smaller numbers of enrolled patients. a Proportions of SSP-Subtype by year of diagnosis. b Proportions of SSP-ROR risk category by year of diagnosis. c Proportions for adjuvant treatment within ER+/HER2-/N0 patients diagnosed at age >50 years by different age at diagnosis. Endo: endocrine therapy only, ChemoEndo: adjuvant chemotherapy and endocrine therapy. None: no adjuvant systemic therapy. d Proportions for adjuvant treatment within ER+/HER2-/N0 patients diagnosed at age >50 ≤ 70 years by year of diagnosis. e Cross-comparison of the naive SSP ETR dichotomized for chemotherapy (yes/no) with records of administered systemic treatment within ER+/HER2-/N0 patients at age >50 ≤ 70. The groups for which SSP treatment recommendation is in agreement with the administered treatment are shown in black for regimen without chemotherapy (No CT) and in red for regimen including chemotherapy (CT). The discordant groups where SSP would lead to escalation of treatment (No CT to CT) are shown in orange and de-escalation of treatment (CT to No CT) in blue. f Kaplan-Meier plot for SSP stratification by SSP-ETR treatment recommendation within the N0 subgroup of ER + /HER2- patients diagnosed at age >50 ≤ 70 and no adjuvant treatment. g Kaplan-Meier plot for SSP stratification by SSP-ETR dichotomized for chemotherapy (chemotherapy vs. no chemotherapy) within ER+/HER2-/N0 patients diagnosed at age >50 ≤ 70 treated with adjuvant endocrine therapy only. h Forest plots of Hazard ratios and 95% confidence interval ranges from univariable and i multivariable Cox regression using DRFi as endpoint stratified using SSP treatment recommendation. Multivariable analysis is with tumor size, age at diagnosis, and NHG as covariates. Test (SSP stratification) and subgroup (administered treatment) is specified on the left of the univariable forest plot. Error bars correspond to 95% confidence interval.