A Natural Language Processing System for National COVID-19 Surveillance in the US Department of Veterans Affairs

Alec B Chapman, Kelly S Peterson, Augie Turano, Tamára L Box, Katherine S Wallace, Makoto Jones

Keywords: natural language processing, nlp, biosurveillance, covid-19, information extraction, public health, clinical nlp, clinical text, infectious disease, epidemiology

TL;DR: We developed and deployed an NLP system for identifying positive COVID-19 cases from text notes to accelerate chart review in a national surveillance system.

Abstract: Timely and accurate accounting of positive cases has been an important part of the response to the COVID-19 pandemic. While most positive cases within Veterans Affairs (VA) are identified through structured laboratory results, some patients are tested or diagnosed outside VA so their clinical status is documented only in free-text narratives. We developed a Natural Language Processing pipeline for identifying positively diagnosed COVID-19 patients and deployed this system to accelerate chart review. As part of the VA national response to COVID-19, this process identified 36.1% of total confirmed positive cases in VA to date. With available data, performance of the system is estimated as 82.4% precision and 94.2% recall. A public-facing implementation is released as open source and available to the community.