Machine Learning-Based Epigenetic Classifiers for Axillary Staging of Patients with ER-Positive Early-Stage Breast
Javier I J Orozco1, Julie Le2, Miquel Ensenyat-Mendez3
1Saint John's Cancer Institute, Providence Saint John's Health Center, Santa Monica, CA, USA.
Background:
In the era of molecular stratification and effective multimodality therapies, surgical staging of the axilla is becoming less relevant for patients with estrogen receptor (ER)-positive early-stage breast cancer (EBC). Therefore, a nonsurgical method for accurately predicting lymph node disease is the next step in the de-escalation of axillary surgery. This study sought to identify epigenetic signatures in the primary tumor that accurately predict lymph node status.
Patients And Methods:
We selected a cohort of patients in The Cancer Genome Atlas (TCGA) with ER-positive, HER2-negative invasive ductal carcinomas, and clinically-negative axillae (n = 127). Clinicopathological nomograms from the Memorial Sloan Kettering Cancer Center (MSKCC) and the MD Anderson Cancer Center (MDACC) were calculated. DNA methylation (DNAm) patterns from primary tumor specimens were compared between patients with pN0 and those with > pN0. The cohort was divided into training (n = 85) and validation (n = 42) sets. Random forest was employed to obtain the combinations of DNAm features with the highest accuracy for stratifying patients with > pN0. The most efficient combinations were selected according to the area under the curve (AUC).
Results:
Clinicopathological models displayed a modest predictive potential for identifying > pN0 disease (MSKCC AUC 0.76, MDACC AUC 0.69, p = 0.15). Differentially methylated sites (DMS) between patients with pN0 and those with > pN0 were identified (n = 1656). DMS showed a similar performance to the MSKCC model (AUC = 0.76, p = 0.83). Machine learning approaches generated five epigenetic classifiers, which showed higher discriminative potential than the clinicopathological variables tested (AUC > 0.88, p < 0.05).
Conclusions:
Epigenetic classifiers based on primary tumor characteristics can efficiently stratify patients with no lymph node involvement from those with axillary lymph node disease, thereby providing an accurate method of staging the axilla.
