Where have you been-the usage of machine learning for capturing travel history

ABSTRACT

Background and objectives: Travel history has become an indispensable part of emergency department (ED) patient assessment due to the ongoing COVID-19 pandemic. The ability to highlight travel from free text notes may augment travel history that is not completely captured in structured data fields. We explore if named-entity recognition (NER), a natural language processing (NLP) technique, can be used to extract travel history from ED free text triage notes (FTN) using a widely available, off-the-shelf, open-source NLP tool. Methods: The FTN of 10,000 attendances at an ED were reviewed by a team of annotators, and the countries, regions, or cities of travel were extracted. The annotated notes were used to train the native, out-of-the-box NER model in SpaCy 3.0.5, which is based on a deep convolutional neural network. Predictions made by the trained model were evaluated on a previously unseen test set. Results: The NER model achieved F1 score of 97.64%, precision of 98.68%, and recall of 96.6% in capturing travel history. Conclusion: Machine learning can be used to accurately capture travel history from ED FTN.