An improved multivariate model that distinguishes COVID-19 from seasonal flu and other respiratory diseases

Abstract

Abstract Background: In many countries, the COVID-19 pandemic is occurring in the middle of flu season. Since the responses to COVID-19 are dramatically different, it is critical to accurately discriminate COVID-19 from seasonal flu and pneumonia caused by other common respiratory pathogens. Methods: Fifty patients (eight patients with COVID-19, eight with influenza, and 34 with community-acquired pneumonia) were included in our study. Sixteen features, such as clinical symptoms, results of routine blood tests, first reverse transcription-polymerase chain reaction (RT-PCR), and chest CT, were collected. The importance of each feature in discriminating COVID-19 from others was ranked by the random forest algorithm. Models with single or multiple features were evaluated using receiver operating characteristic (ROC) curves, the F1 score, and Matthews correlation coefficient (MCC). Results: An integrated multi-feature model (RT-PCR, CT features and blood lymphocyte percentage) yielded an area under the ROC curve of 0.97 (95% CI: 0.86 – 1, P < 0.01), an F1 score of 0.81 and an MCC of 0.78 in the training cohort as well as an F1 score of 0.86 and an MCC of 0.85 in the validation set. Conclusion: The developed multivariate model showed better accuracy than the current nucleic acid-based method for the differentiation of COVID-19 from influenza and pneumonia caused by other common respiratory pathogens.