Establishment of a diagnostic model to distinguish coronavirus disease 2019 from influenza A based on laboratory findings

Background: Coronavirus disease 2019 (COVID-19) and Influenza A are common disease caused by viral infection. The clinical symptoms and transmission routes of the two diseases are similar. However, there are no relevant studies on laboratory diagnostic models to discriminate COVID-19 and influenza A. This study aims at establishing a signature of laboratory findings to tell patients with COVID-19 apart from those with influenza A perfectly. Materials: In this study, 56 COVID-19 patients and 54 influenza A patients were included. Laboratory findings, epidemiological characteristics and demographic data were obtained from electronic medical record databases. Elastic network models, followed by a stepwise logistic regression model were implemented to identify indicators capable of discriminating COVID-19 and influenza A. A nomogram is diagramed to show the resulting discriminative model. Results: The majority of hematological and biochemical parameters in COVID-19 patients were significantly different from those in influenza A patients. In the final model, albumin/globulin (A/G), total bilirubin (TBIL) and erythrocyte specific volume (HCT) were selected as predictors. Using an external dataset, the model was validated to perform well. Conclusion: A diagnostic model of laboratory findings was established, in which A/G, TBIL and HCT were included as highly relevant indicators for the segmentation of COVID-19 and influenza A, providing a complimentary means for the precise diagnosis of these two diseases.


Introduction
Since COVID-19 emerged in 2019, the disease has spread rapidly around the world and attracted global attention. The epidemic of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) was declared an international public health emergency by the World Health Organization (WHO). The early typical clinical symptoms of the disease are fever, respiratory symptoms and muscle pain (Mark R. Geier and Geier 2020). In severe cases, acute respiratory distress syndrome (ARDS) or even respiratory failure (RF) can develop.
During winter, viral infectious diseases gradually enter a high incidence period. As a common seasonal influenza, influenza A tends to be prevalent in winter (Lei et al. 2018). This virus is transmitted through three main routes (Sullivan et al. 2010) (contact transmission, droplet transmission and airborne transmission), which can lead to a wide range of human-to-human transmission. The early clinical symptoms of influenza A patients include fever, headache, muscle pain and dyspnea (Schoen et al. 2019), which are similar to the early onset of COVID-19. Hashemi et al. found the co-infection of SARS-CoV-2 with other respiratory viruses (Hashemi et al. 2020 http://www.sz.gov.cn/cn/xxgk/zfxxgj/zwdt/content/post_8307353.html Accessed November 27 2020.) has also reported that four patients were infected with influenza A and COVID-19 simultaneously, and thus, an accurate and efficient segmentation of influenza A and COVID-19 is of crucial importance.

Page 6
Some studies have reported that the count of white blood cell (WBC), lymphocyte (LY) and platelet (PLT) were decreased, and the levels of alanine aminotransferase (ALT) and aspartate aminotransferase (AST) were increased in H1N1 and H5N1 patients (Gao et al. 2013; Natalie L. Cobb et al. 2020). COVID-19 patients can also exhibit these changes Tian et al. 2020

Patient Selection and Data Sources
In the training set, the patients were recruited from three designated tertiary hospitals in

Identification of diagnostic indices
To identify the indices that have diagnostic value for distinguishing between COVID-19 and influenza A, we carried out the following machine learning procedure. First, the laboratory findings during the first five days of the hospital stay were averaged and then used as potential attributes for feature selection. Elastic net models were used for the first round of selection, in which the tuning parameter alpha was set at 0.6 and the optimal cutoff of lambda was chosen by performing 10-fold cross-validations. With different random seeds, the elastic net models were fit for 200 times. Then, the frequencies of the indices selected by these 200 models were calculated. When the indices with high frequencies were selected (> 90%), the pairwise correlation coefficients for each pair were calculated. To eliminate redundant indices Page 9 and to avoid collinearity issues among the highly correlated indices, those with the least biological meaning were excluded. For the second-round feature selection modeling, stepwise logistic regression models with AIC as the selection criterions were utilized. Finally, a ridge logistic regression model was fit with the selected indices as predictors. A nomogram was diagramed to graphically elucidate the final model. Using an independent dataset as an external validation set, the area under the receiver operating characteristic (ROC) curve (AUC) metric was calculated to evaluate the predictive performance of the final model.

Comparison at baseline
Chi-square test is applied to the processing of basic information data, and non-parametric method is applied to calculate the reference value of 2.5% -95% for the RIs of laboratory test index. The statistical processing by using IBM SPSS Statistics 26 software.

Demographics
In this study, 56 COVID-19 patients (aged 10 to 87 years) and 54 influenza A patients (aged 23 to 87 years) were included. The demographic characteristics of these two groups are presented in Table 1. Hospitalization time for COVID-19 patients was longer. There was no significant difference between two groups in males/females ratio, but the median age of COVID-19 patients was lower than that of influenza A patients. Fever, cough and fatigue were the common clinical symptoms of COVID-19 patients. The common clinical symptoms of influenza A patients were fever, cough and dyspnea.

Laboratory Findings at Hospital Admission
In this study, laboratory findings were performed on the hematological and biochemical parameters of 56 COVID-19 patients and 54 influenza A patients. The first laboratory findings from the hospital patients are shown in Table 2. Among the leukocyte parameters, the count of WBC, neutrophil (NE), and monocyte (MO) in the influenza A patients were significantly higher than those in the COVID-19 patients (P < 0.001); however, there was no significant difference in lymphocyte (LY), eosinophilic granulocyte (EO) and basophilic granulocyte (BA) (P > 0.05). In terms of erythrocyte parameters, the count of RBC, hemoglobin (HGB), HCT and mean erythrocyte hemoglobin concentration (MCHC) in the COVID-19 patients were all higher than those in the influenza A patients, while the RBC distribution width (RDW) was lower than that of influenza A patients (P < 0.001). Regarding PLT parameters, the mean platelet volume (MPV) of the COVID-19 patients was significantly reduced (P＜ 0.001). Regarding the biochemical parameters, there was a significant difference in the levels of other parameters except for aspartate aminotransferase (ALT), creatinine (Cr) and glucose (GLu). Among the electrolyte parameters, potassium (K+) and sodium (Na+) levels in the COVID-19 patients were higher than those in the influenza A patients, while the chlorine (Cl -) levels in the COVID-19 patients were lower. 90% over 200 elastic net models (the plot to select the optimal tuning parameter lambda for a single run of elastic net is shown in Figure 1A). Then, their correlation plot was diagrammed ( Figure 1B)  Page 12

Diagnostic Indices to Discriminate between COVID-19 and Influenza
Significant differences in most laboratory findings between COVID-19 and influenza A patients at the onset of diseases were found. For example, the count of WBC and NE decreased significantly in COVID-19 patients, which is consistent with the findings of Chen's study ) and Guan's study (Guan et al. 2020 stage of infection, H1N1 patients had a high frequency of fever and pneumonia (Weiss and Goodnough 2005), and thus the RDW of the influenza A patients was significantly increased.
In addition, the count of PLT in influenza A and COVID-19 patients were significantly reduced. Abelleira et al. found in control study of patients with influenza A that the count of PLT in the case group was lower than that in the control group (Abelleira et al. 2019). Chen Other studies have also confirmed (Gao et al. 2013;Zhang et al. 2014) that patients with H1N1 had clinical symptoms of hypoxemia and reduced partial pressure of carbon dioxide, which were associated with reduced CO2-CP. At the same time, the K + , Na + and Cl+ levels of patients in the two groups were also decreased. Gao (Gao et al. 2013) and Page 14 Chen

et al. mentioned that both COVID-19 and influenza A patients
suffered vomiting and diarrhea, and more than one-half of influenza A patients were reported to have hypokalemia and hyponatremia (Zhang et al. 2014). The specific reasons are not clear, but it is currently believed that electrolyte parameter changes may be related to the above clinical symptoms.
Although some laboratory findings were found to be different in the two diseases, their variation trends were similar. Therefore, to better classify the two diseases, this study aimed to establish a diagnostic model according to laboratory findings for COVID-19 and influenza A and then select the most representative indicators for the clinical identification of the two viral infections. Using machine learning methods, we showed that the three laboratory findings of A/G, TBIL and HCT possess predictive capacity to discriminate the two diseases (P<0.001, 0.014 and 0.037, respectively). Studies have found that influenza A patients exhibit hypoproteinemia and hypoalbuminemia (Zhang et al. 2014). Tang (Xiao Tang et al. 2020)et al. reported that the ALB level of influenza A patients was significantly lower than that of COVID-19 patients. In this study, the GLB level of influenza A patients was higher than that of COVID-19 patients. These conclusions indirectly proved that the A/G ratio in influenza A patients was decreased significantly, which was consistent with the results of this study. In addition, TBIL level in influenza A patients were significantly higher than those in COVID-19 patients. A cohort study by Zhang(YiMin Zhang et al. 2016) and Tang(Xiao Tang et al. 2020) et al. confirmed that the TBIL level in influenza A patients was increased significantly. This Page 15 may also be due to liver injury caused by clinical drugs, which somehow influence TBIL level.
The drug oseltamivir is commonly used for the treatment of influenza A virus, which is metabolized in vitro by liver esterase. The frequent use of oseltamivir reduces the level of liver esterase and leads to drug residues in the body of patients, thereby causing liver damage (Shengbo Fang et al. 2018). Hematocrit (HCT) refers to the volume ratio of sinking red blood cells to whole blood measured after centrifugal precipitation of a certain amount of whole blood treated with anticoagulant, which indirectly reflects the number and volume of red blood cells. In this study, the HCT level in the influenza A patients was lower than that in the COVID-19 patients, and the RBC count was significantly lower than that in the

Conclusion
Currently, there is a lack of effective drugs and vaccines to prevent and treat COVID-19 in the clinic. The winter time accelerates the spread of SARS-CoV-2 and influenza A virus.
Due to the two diseases are extremly similar in clinical symptoms and transmission routes Therefore, it was necessary to effectively diagnose and treat these two diseases, prevent the