Ontology-aware Learning from Electronic Health Records

Abstract

Advances in natural language processing (NLP) in recent times has shown a great promise in improving the patient profiles with the help of their clinical notes. In medical practices, preparing clinical details for patients often happen through longer forms, which are really difficult to maintain and process. Therefore, people use abbreviations (writing a medical term in a shorter form) to record clinical details. In clinical notes, abbreviations are used recklessly without mentioning their definitions. These abbreviations can have different expansions based on their medical context. For example, the abbreviation “ivf” may denote either “intravenous fluid” or “in vitro fertilization” based on their contexts. It is thus a challenging task for NLP systems to correctly disambiguate abbreviations in their clinical notes. We have used the Naive Bayes approach for correctly disambiguating medical concepts and abbreviations by using NLP models. We have proposed a measure to find whether a given medical abbreviation is related to COVID or non- COVID.We have trained our model on the COVID ontologies and general medical concepts and tested it on the dataset whichwe have compiled at our own.We have tried to determine the correct senses for an abbreviation based on the associated context.