Artificial intelligence for clinical data extraction from EHRs in breast cancer trials: the MIRROR study

Abstract

Manual data entry and curation is the most used method of data collection for real world evidence and clinical trials but is a time-consuming task associated with significant human effort and transcription mistakes. MIRROR is a retrospective study that compared manual data capture from electronic case report forms (eCRF) to automated data extraction from electronic health records (EHRs) using artificial intelligence (AI). Clinical information was extracted from EHRs of 113 breast cancer patients participating in 11 clinical trials, with the system providing references to the original text for each variable to enable efficient verification. All patients were enrolled at the Virgen del Rocío University Hospital (Sevilla, Spain). Analyses provided high rates of accuracy, precision, recall and F1-score for most structured and unstructured clinical data domains. Future efforts should focus on improving unstructured data extraction and implementing standardized eCRF among sites to ensure consistent extraction and evaluation of clinical information.