Rapid and accurate identification of COVID-19 infection through machine learning based on clinical available blood test results

Jiangpeng Wu; Pengyi Zhang; Liting Zhang; Wenbo Meng; Junfeng Li; Chongxiang Tong; Yonghong Li; Jing Cai; Zengwei Yang; Jinhong Zhu; Meie Zhao; Huirong Huang; Xiaodong Xie; Shuyan Li

Except for glucose (GLU) and magnesium (Mg), almost all the parameters manifested the significant differences between COVID-19 and general pneumonia in Figure 3 . It also indicated the parameters were highly correlated to the infection of SARS-CoV-2 and the powerful ability of the tool to effectively distinguish patients with COVID-19 and patients with pneumonia. In fact, many abnormal changes of laboratory parameters had been widely reported for important clinical references, including total bilirubin (TBIL), GLU, creatinine (CREA), lactate dehydrogenase (LDH), creatine kinase isoenzyme (CK-MB), and kalium (K). 22, 25, 26 For example, the descriptive study of 99 cases with COVID-19 in Wuhan showed that 51 (52%) patients with the elevated level of GLU and 75 (76%) patients with the elevated level of LDH, which was consistent with our results. 27 The remaining parameters, including total protein (TP), calcium (Ca), magnesium (Mg), platelet distribution width (PDW), and basophil (BA#), were rarely received extensive attention due to the lack of information on the clinical characteristics of affected patient. However, they played an irreplaceable part in the random forest algorithm and had great potential for diagnostic markers with future clinical practice. Indeed, the existence of differences was not a decisive factor to select critical parameters for machine learning algorithms, it was helpful to promote the tool to quickly identify COVID-19 infection and provide a sufficient consciousness for dynamically monitoring the process of the disease.