Smell and taste symptom-based predictive model for COVID-19 diagnosis

Abstract

Background: The presentation of coronavirus 2019 (COVID-19) overlaps with common influenza symptoms. There is limited data on whether a specific symptom or collection of symptoms may be useful to predict test positivity.

Methods: An anonymous electronic survey was publicized through social media to query participants with COVID-19 testing. Respondents were questioned regarding 10 presenting symptoms, demographic information, comorbidities, and COVID-19 test results. Stepwise logistic regression was used to identify predictors for COVID-19 positivity. Selected classifiers were assessed for prediction performance using receiver operating characteristic (ROC) curve analysis.

Results: A total of 145 participants with positive COVID-19 testing and 157 with negative results were included. Participants had a mean age of 39 years, and 214 (72%) were female. Smell or taste change, fever, and body ache were associated with COVID-19 positivity, and shortness of breath and sore throat were associated with a negative test result (p < 0.05). A model using all 5 diagnostic symptoms had the highest accuracy with a predictive ability of 82% in discriminating between COVID-19 results. To maximize sensitivity and maintain fair diagnostic accuracy, a combination of 2 symptoms, change in sense of smell or taste and fever was found to have a sensitivity of 70% and overall discrimination accuracy of 75%.

Conclusion: Smell or taste change is a strong predictor for a COVID-19-positive test result. Using the presence of smell or taste change with fever, this parsimonious classifier correctly predicts 75% of COVID-19 test results. A larger cohort of respondents will be necessary to refine classifier performance.

Keywords: COVID-19; predictors; receiver operating characteristic curve; smell; symptoms; taste.

FIGURE 1

ROC curve plots for symptom classifier models. The dashed diagonal line shows a nondiagnostic result. AUC for each symptom classifier group is displayed in Table 5. AUC = area under the ROC curve; ROC = receiver operating characteristic.