Natural language processing in electronic health records: A review

Abstract

The two fundamental tasks that a physician performs during every interaction with a patient are reading and updating electronic health records (EHRs). Reading the records is necessary to gain better knowledge of a patient’s health status while updating the records is essential for creating a database for future information extraction. If a patient’s history consists of only a few records, manual reading is the best approach. However, this method may lead to overlooking important aspects of the patient’s health, which could be detrimental. Therefore, automation is required to extract important information. Natural language processing (NLP) facilitates information extraction and operates on seven different levels. In our review, we aimed to understand how NLP levels assist in extracting information. We examined articles published in PubMed and, after critical evaluation, selected 65 out of 382 identified articles that met the inclusion criteria for the final review. Among these, 47 articles were included in the final review. We found a higher number of articles on the lexical (7), semantic (30), and morphological (4) levels, while fewer articles focused on the phonetic (1), syntactic (2), discourse (2), and pragmatic (1) levels. This distribution underscores the current emphasis within the literature on the specific aspects of NLP. In conclusion, our review underscores the critical role played by NLP in extracting information from EHR, shedding light on the varied levels at which this technology operates.