Triage Modeling for Differential Diagnosis between COVID-19 and Human Influenza A Pneumonia: Classification and Regression Tree Analysis

Background: The outbreak of COVID-19 overlaps with the usual flu season, and the main signs, symptoms and imaging manifestations in COVID-19 and flu infections share similarity. This study is aimed to constructed an accurate and robust model for initial screening and differential diagnosis of COVID-19 and influenza A, so as to provide an effective tool for global prevention and control of COVID-19 outbreak. Methods: The 151 COVID-19 infections patients and 155 influenza A patients in Fuyang No.2 People's Hospital were included and randomly assigned in a 4:1 ratio to train set and test set. Univariate logistic regression analysis was conducted between two diseases. The predictor variables were selected by variables importance assessed by random forest algorithms and were analyzed in train set to develop classification and regression tree models. The validity of the model was tested using data from test set. Findings: Recursive partitioning of the train set indicated that the optimal model for prediction was model A (signs and symptoms+ serum biochemistry), in which the best single predictor for COVID-19 patients was normal or high level of LDL-c, followed by low level of CK, then by frequency of respiratory symptom less than 3 times, and last by highest temperature on the first day of admission less than 38℃. The AUC, sensitivity, specificity was 93%, 100%, 87% respectively. The Suboptimal model was model B (signs and symptoms+ routine blood), in which the best single predictor for COVID-19 patients was low level of EO#, followed by normal level of MONO%, then by normal level of HCT, next by highest temperature on the first day of admission less than 37℃, and last by frequency of respiratory symptom less than 1 times. The AUC, sensitivity, specificity was 87%, 73%, 100% respectively. Interpretation: The two models that we created provides clinicians with a rapid triage tool and identification means for quickly identifying and screening pneumonia caused by COVID-19 and seasonal influenza. The optimal model can be applied to developed countries/regions and major hospitals, and the suboptimal one can be used in grass-roots areas, underdeveloped regions and small hospitals. Funding Statement: This work is supported by Shanghai Municipal Science and Technology Major Project (No.2018SHZDZX01) and ZJ Lab; the Shanghai Municipal Commission of Health and the Family Foundation for Young Talents (2017YQ023). Declaration of Interests: All authors declare no competing interests. Ethics Approval Statement: This study was approved by the Ethics Committee of the Fuyang No.2 People's Hospital, Anhui Province.