Tech & Health

April 15, 2021

Team analyzed clinical notes to identify COVID symptoms

At the start of the COVID-19 pandemic, public health authorities first had to identify and draw attention to common signs and symptoms of the disease.


by Paul Govern

At the start of the COVID-19 pandemic, public health authorities first had to identify and draw attention to common signs and symptoms of the disease.

On Feb. 25, 2020, with 14 cases having been diagnosed in the U.S., the Centers for Disease Control and Prevention (CDC) initially went with fever, cough and shortness of breath. This brief official list of coronavirus disease symptoms remained unchanged until mid-April.

Juan Zhao, PhD

Last spring at Vanderbilt University Medical Center, natural language processing of clinical notes from the electronic health records (EHRs) of people tested for COVID-19 had turned up loss of taste or smell as a symptom of the disease by April 5, nearly two weeks before the CDC added this symptom to the official list on April 17.

In setting out in the early days of the pandemic to identify clinical characteristics of COVID-19, Juan Zhao, PhD, Wei-Qi Wei, MD, PhD, and colleagues used a new method they call concept-wide association study, or conceptWAS, to analyze clinical notes of 19,692 adult patients who received COVID testing at VUMC between March 8 and May 27, 2020.

Using a tool developed at VUMC called KnowledgeMap Concept Indexer, they extracted concepts from notes written within days of a COVID test. In all, 1,483 of these patients, or 7.5%, tested positive for COVID. The team reports their analysis in the Journal of Biomedical Informatics.

Wei-Qi Wei, MD, PhD

“For any institution that uses computers to study its EHR, we developed conceptWAS as a flexible high-throughput computational method that can be readily adapted to study early symptoms of any emerging disease,” said Wei, assistant professor of Biomedical Informatics.

As test results accumulated over the 12-week study period, the team re-ran their analysis every two weeks. At completion they had turned up 68 clinical concepts significantly associated with COVID-19. To validate the associations, they conducted manual chart reviews.

Top concepts positively associated with COVID-19 included loss of taste, loss of smell, fever, and fever with cough. Depression and anxiety were each negatively correlated with COVID, markedly so, as were edema and the ruling out of fever. Consistent with other studies, the analysis also found a negative correlation between smoking and COVID-19.

“Compared to structured data such as diagnosis codes and lab results, clinical notes are seen as relatively unwieldy for research purposes, and they remain a comparatively untapped resource for discovery,” Wei said. “With this proof of concept, we’re presenting a method to tap this resource to quickly characterize an emerging disease and inform clinical decision-making without delay.”

Also on the study were Monika Grabowska, Eric Kerchberger, MD, Joshua Smith, PhD, Hatice Eken, QiPing Feng, PhD, Josh Peterson, MD, MPH, Trent Rosenbloom, MD, MPH, and Kevin Johnson, MD, MS. The study was supported in part by the National Institutes of Health (GM115305).