Information Extraction From Electronic Health Records Using Deep Learning

Abstract

Information extraction involves the automatic extraction of structured data from its unstructured or semi-structured form. Electronic Health Records contain useful information in an unstructured form, on which we can apply an information extraction strategy to extract structured information. Several traditional methods are available for information extraction, including both manual and rule-based approaches. The rule-based extraction is inadequate but straightforward and does not cover all aspects. The manual extraction, on the other hand, is effective but very slow, costly, and not scalable. Deep learning models have recently demonstrated excellent performance when applied to various Natural Language Processing tasks. In this study, we present a deep learning framework for multi-label medical entity extraction from Electronic Health Records, addressing the issue where a medical term may represent multiple clinical roles. Additionally, the proposed approach supports real-time entity prediction, enabling users to input terms and receive comprehensive insights instantly, including diseases, symptoms, diagnostic tests, and treatments, thereby providing a practical tool for clinical decision support. Firstly, it preprocesses the EHR data and then applies deep learning models and the Runtime Context Identifier to extract the information. The results have shown that the proposed model achieved an average accuracy of 89% across five diverse EHR datasets, demonstrating robust overall performance.