Identifying Patients With Low Relapse Rate Despite High-Risk Estrogen Receptor-Positive/Human Epidermal Growth Factor Receptor 2-Negative Early Breast Cancer: Development and Validation of a Clinicopathologic Assay

Abstract

Purpose: Escalation of adjuvant systemic therapies (eg, with cyclin-dependent kinase 4 and 6 inhibitors) is now indicated for patients with clinically defined high-risk estrogen receptor-positive (ER+)/human epidermal growth factor receptor 2-negative (HER2-) early breast cancer, although it is unclear which will benefit from additional therapies. We developed and validated a prognostic clinicopathologic assay identifying a subpopulation of high-risk patients with good prognosis after standard adjuvant therapies, who may safely forgo treatment escalation.

Methods: We trained a Cox proportional-hazards model that integrates clinicopathologic variables with features derived from digitized hematoxylin-and-eosin-stained resection slides from a retrospective data set. The model assigns each patient to a low-risk or not low-risk group, reflecting their predicted risk of recurrence. Blind validation was successively performed on high-risk patients from the prospective trials CANTO (ClinicalTrials.gov identifier: NCT01993498) and UNIRAD (ClinicalTrials.gov identifier: NCT01805271).

Results: Built on data from 6,164 patients with ER+/HER2- early-stage breast cancer, this assay integrates four clinicopathologic variables, and 10 slide-derived features capturing tumor architecture, microenvironment, and proliferation. In the combined CANTO and UNIRAD trials (n = 633), 95.4% of the low-risk patients remained free of distant recurrence and death from breast cancer at 9 years, compared with 76.8% for the not low-risk group. Distant recurrence-free interval (subdistribution hazard ratio [HR], 0.21 [95% CI, 0.09 to 0.52]; P < .001), invasive disease-free survival (HR, 0.31 [95% CI, 0.16 to 0.60]; P < .001), and overall survival (HR, 0.35 [95% CI, 0.13 to 0.97]; P = .044) were all statistically significant. Multivariate analyses showed that the assay provided predictive information beyond clinicopathologic variables. Analytical validation showed robustness to data variability.

Conclusion: The assay demonstrated robust performance in identifying a core group of patients with high-risk ER+/HER2- breast cancer for whom additional adjuvant treatment may be futile.

FIG 1.

Study design. This figure illustrates the four-phase study design for developing and validating a prognostic assay for ER+/HER2– high-risk early breast cancer. (A) Clinical utility objective: the clinical positioning aimed to develop an assay identifying a subgroup of high-risk patients with minimal residual risk after adjuvant chemoendocrine therapy, who may safely forgo further treatment escalation. (B) Assay development: the prognostic model was trained using a data set of 6,164 patients, integrating clinicopathologic variables with computationally extracted features from H&E-stained histology digital slides (C). The assay was locked down at the end of the development phase, including after the choice of the cutoff converting continuous risk predictions to group classifications (low-risk or not low-risk). (D) Clinical validation: the assay was validated in two independent high-risk eBC cohorts, first from the CANTO clinical trial (n = 402) and then from the UNIRAD clinical trial (n = 231), assessing its prognostic performance for DRFI, IDFS, and OS. (E) Analytical validation: robustness was evaluated across tumor heterogeneity, different slide-scanning platforms, and clinicopathologic variability. CT, chemotherapy; DRFI, distant recurrence-free interval; ER+, estrogen receptor–positive; ET, endocrine therapy; H&E, hematoxylin-and-eosin; HER2–, human epidermal growth factor receptor 2–negative; IDFS, invasive disease-free survival; OS, overall survival.

FIG 2.

Flowchart of UNIRAD and CANTO. The flowchart illustrates the selection process for patients from the UNIRAD and CANTO cohorts included in the analysis. Patients were initially screened on the basis of the availability of primary tumor blocks, with exclusions made for those without samples. Patients who received neoadjuvant chemotherapy were excluded, as were those who did not meet the high-risk criteria, which included having node-positive tumor with at least one of the following: four or more positive lymph nodes, tumor size ≥50 mm, histologic grade 3, or Ki-67 ≥20%. In the UNIRAD cohort, 231 patients with high-risk ER+/HER2− eBC were included, while 402 patients were included from the CANTO cohort after excluding individuals with missing follow-up or event data. BC, breast cancer; eBC, early invasive breast cancer; ER+, estrogen receptor–positive; ER−, estrogen receptor–negative; HER2+, human epidermal growth factor receptor 2–positive; HER2–, human epidermal growth factor receptor 2–negative.

FIG 3.

Survival analyses by risk group (low-risk v not low-risk) assigned by the assay. Survival outcomes for DRFI, IDFS, and OS on the combined CANTO and UNIRAD cohorts, stratified by the risk group classification (low-risk v not low-risk) assigned by the assay. (A) DRFI estimated using the CIF to account for competing risks such as deaths unrelated to breast cancer. sHR and 95% CI were calculated using the Fine and Gray model, with P values obtained from Gray's test. (B) IDFS estimated using Kaplan-Meier survival analysis. HR and 95% CI were derived from Cox proportional-hazards models, with statistical significance assessed using the log-rank test. (C) OS also estimated using Kaplan-Meier survival analysis, with HR and 95% CI calculated using Cox proportional-hazards models and P values obtained from the log-rank test. In each figure, survival curves for the low-risk group (teal) and not low-risk group (red) are shown. Numbers at risk are displayed below the x-axis. Summary estimates of sHR (for DRFI) or HR (for IDFS and OS), with their corresponding P values, are reported within the plots. CIF, cumulative incidence function; DRFI, distant recurrence-free interval; IDFS, invasive disease-free survival; HR, hazard ratio; OS, overall survival; sHR, subdistribution hazard ratios.

FIG 4.

Subgroup analyses of DRFI and IDFS. Forest plots showing subgroup analyses of survival outcomes for (A) DRFI and (B) IDFS on the combined CANTO and UNIRAD cohorts. The x-axis represents sHR for (A) or HR for (B), with the vertical dashed line at 1.0 indicating no difference between the assigned low-risk and not low-risk groups. Squares represent the point estimates, with their size proportional to the number of patients in each subgroup. Horizontal lines denote 95% CI. In both plots, the overall hazard ratio and confidence interval are summarized in the row labeled Overall. Rows marked with * indicate subgroups where no events occurred in the low-risk group, making it impossible to compute HR or sHR values for those categories. (A; DRFI): sHR were estimated using Fine and Gray models to account for competing risks (deaths unrelated to breast cancer). Gray's test was used to compute P values for differences across subgroups, and p-interaction values evaluate heterogeneity between subgroups. (B; IDFS): HRs were calculated using Cox proportional-hazards models. The log-rank test was used to compute P values for differences across subgroups, and p-interaction values assess heterogeneity between subgroups. DRFI, distant recurrence-free interval; HR, hazard ratio; IDFS, invasive disease-free survival; sHR, subdistribution hazard ratio.