We all want to think that when we receive a diagnosis from a reputable clinic or hospital that there is no need to doubt. However, when measuring breast cancer biopsy samples for ER, PR and HER2 sensitivity, it has been estimated that there is a discrepancy of up to 19%  between central laboratories and local pathology laboratories. Simply put, it is a tedious task and open to human err.

“The pathological review of tumor samples, even for common molecular biomarkers such as estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2), is time consuming. Moreover, there is not always concordance between pathologists on the interpretation of samples,” reports a new article in JAMA. (1)

Enter Artificial intelligence (AI). AI and machine learning technologies are new technologies being applied to address this variation as well as to improve reliability and add efficiency. Currently, such technology can differentiate between cancerous and noncancerous tissue as well as determine presence of metastases in lymph nodes and perform tumor grading.

Recently, investigators have conducted a retrospective, single-institution study to test the ability of a machine learning technique — referred to as morphological-based molecular profiling — to assess hormonal status of more than 20,000 digitized hematoxylin-eosin (H&E) pathology specimens from a microarray library of more than 5000 breast cancer patients.

Histology and biomarkers were found to be significantly correlated with all 19 assessed biomarkers, including most clinically relevant ER, PR, and HER2. For approximately half of the patients, the machine learning technique was able to predict biomarker expression with noninferiority to immunohistochemistry (IHC) in two validation cohorts with positive predictive values of 97% and 98%. Also, for patients with ER-negative/PR-positive tumors assessed by conventional IHC, machine learning techniques revealed resemblance to patients with ER-positive tumors, suggesting that the IHC result was falsely negative and that a fraction of ER-negative/PR-positive patients might benefit from endocrine therapy.

(1) Artificial Intelligence to Assess Hormonal Status of Breast Cancer Patients, William J. Gradishar, MD reviewing Shamai G et al. JAMA Netw Open 2019 Jul 26

shutterstock 341629238 1000x480