Predicting the likelihood of <i>BRCA1/2</i> germline pathogenic variants in unselected patients with breast cancer: analysis of more than 10,000 individuals.

Abstract

Models for accurately predicting the likelihood of <i>BRCA1/2</i> germline pathogenic variants (PVs) based on a large cohort of unselected patients with breast cancer are limited. A logistic regression model to predict the <i>BRCA1/2</i> carrier probability, named PKCBRCA, was established and validated based on 10 167 unselected Chinese patients with breast cancer treated in Peking University Cancer Hospital between October 2003 and August 2020. All patients were tested for <i>BRCA1/2</i> germline variants. The discrimination and calibration of the model were assessed. A total of 601 (5.9%) patients carried <i>BRCA1/2</i> germline PVs in the entire cohort of 10 167 unselected patients with breast cancer. The cohort was separated into a training set (n=6331; 387 (6.1%) <i>BRCA1/2</i> carriers) and a validation set (n=3836; 214 (5.6%) <i>BRCA1/2</i> carriers). Five variables strongly associated with <i>BRCA1/2</i> carrier probability were incorporated in the establishment of PKCBRCA including age of diagnosis, bilateral breast cancer, family history of breast or ovarian cancer, hormone receptor and ERBB2. PKCBRCA showed a good ability to discriminate both in the training set (area under the receiver operating characteristic curve (AUC)=0.77) and in the validation set (AUC=0.77). Our model provides a useful tool for accurately assessing the <i>BRCA1/2</i> carrier probability for unselected patients with breast cancer.