Transforming unstructured breast cancer pathology reports into the Observational Medical Outcomes Partnership Common...

Although breast cancer pathology reports provide important clinical research data, for they are written as free text, previous studies extracted data from pathology reports through natural language processing (NLP). This study aimed to present the process of standardizing the data extracted from pathology reports to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) and demonstrate its potential utility.

Data were extracted from 11,374 immunohistochemistry, 5,904 molecular pathology, and 13,857 surgical pathology reports from 10,730 patients with breast cancer at a tertiary general hospital between May 2003 and December 2022. Using rule-based NLP, which includes text pre-processing, report segmentation, and pattern-based extraction, along with the mapping of the extracted data to OMOP standard vocabularies, we integrated breast cancer pathology data into the pre-existing OMOP-CDM.

NLP-derived data from the pathology reports were populated with the NOTE_NLP, CONDITION_OCCURRENCE, MEASUREMENT, SPECIMEN, FACT_RELATIONSHIP, EPISODE, and EPISODE_EVENT tables. In a feasibility study, patient-level prediction analyses were conducted to demonstrate the utility of the newly added breast cancer pathology data and open-source tools for OMOP-CDM, yielding an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.840 for predicting all-cause mortality within 5 years after the initial pathologic diagnosis.

By integrating breast cancer pathology information into the OMOP-CDM, this study facilitates the unified analysis of clinical and pathological data for retrospective oncology research. Moreover, the proposed NLP-based pathology report standardization framework for integrating into the OMOP-CDM can be readily adopted by other institutions as the use of OMOP-CDM continues to expand in cancer research.