Extraction of breast cancer biomarker status using natural language processing

Abstract

We employed natural language processing (NLP) algorithms to extract estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor 2 (HER2) receptor status for females with breast cancer using unstructured (free text) EMR data, and to determine the prevalence of triple negative breast cancer in the Indiana network for patient care (INPC) population. We identified female patients in INPC with a history of breast cancer over a ten year period who had at least five oncology notes or one related pathology document. Based on manual chart review, our NLP algorithms for extracting ER, PR, and HER2 receptor status performed well with sensitivity 87.5% to 92.6%, specificity 88.6% to 95.8%, positive predictive values (PPV) 82.4% to 99.0%, and negative predictive values (NPV) 85.2% to 97.7%. This study confirmed our primary hypothesis that NLP algorithms are effective in identifying important breast cancer biomarkers in patients with breast cancer using unstructured data.

Machine Learning-Based Extraction of Breast Cancer Receptor Status From Bilingual Free-Text Pathology Reports

“…This study was designed to evaluate the possibility of automatically extracting the status of the 3 main breast cancer biomarkers (ER, PR, and HER2) from the contents of pathology reports written in two different languages, and coming from 82 different providers, using conventional machine learning models. After testing different classifiers, the best performing ones achieved macro-averaged F 1 scores ranging from 0.89 to 0.92 on the held-out test sets, which is on par with best efforts in the literature (6,7,11,12). The reported F 1 scores in the literature range between 0.87 and 1, but use only three possible labels for HER2, whereas five are used in the present work.…”

“…Within the context of these activities, having the breast cancer receptor status at its disposal would undoubtedly be of added value. In breast cancer, estrogen receptor (ER), progesterone receptor (PR), and Erb-b2 receptor tyrosine kinase 2 (ERBB2, previously named Human Epidermal Growth Factor 2 or HER2 or HER-2/neu 2 ) are biomarkers known to be related to tumor growth and prognosis, and assessing their expression is necessary to define therapeutic management (3)(4)(5)(6)(7). Currently this information is not available in a structured form at the BCR.…”

“…For instance, Dexter et al wrote a series of rules to identify and classify sentences corresponding information about the 1 https://kankerregister.org/media/docs/Exempleprotocoles-version2020.txt and https://kankerregister.org/media/docs/Voorbeeldprotocols%E2%80%93versie202 0.txt 2 The commonly used name HER2 will be used throughout this manuscript. biomarker status (7). Designing rule-based tools is very timeconsuming, because there are many ways to express the information of interest (6,11,12).…”

“…Several natural language processing (NLP) tools have previously been developed to automatically extract ER, PR, and HER2 status from free-text reports written in English ( 7 , 11 ), Chinese ( 12 ), and Bulgarian ( 6 ). In general, clinical NLP tools can be separated into three categories, according to the applied methodology: rule-based ( 7 , 13 ), conventional machine learning ( 11 , 12 ), and deep learning ( 6 ).…”

“…Rule-based clinical NLP tools rely on a set of rules written by experts describing how a computer should classify a report. For instance, Dexter et al wrote a series of rules to identify and classify sentences corresponding information about the biomarker status ( 7 ). Designing rule-based tools is very time-consuming, because there are many ways to express the information of interest ( 6 , 11 , 12 ).…”