A novel auxiliary diagnostic model for COVID-19 screening using enzymes based detection of serum biomarkers and...

The screening for the coronavirus disease 2019 (COVID-19) based on virus nuclear acid detection and radiology has encountered unprecedented difficulties due to the shortage of kits and facilities, and the lack of sensitivity and specificity, especially for developing countries. The study aimed to develop an auxiliary diagnostic score based on age, biomarkers, clinical characteristics (ABC) to rapidly and accurately screen COVID-19. Serum biomarkers were detected by enzymes catalyzed reaction method which is rapid and accurate. A retrospective case-control study among Chinese patients with laboratory-confirmed COVID-19 and those without COVID-19 was conducted. Data of age, sex, signs and symptoms, history of disease, complete blood counts, and serum biochemical items such as ALT and AST were used to establish the diagnostic model. ALT/AST was detected by enzymes based biochemical reaction method. Stepwise logistic regression and random forests with variable selection process were used to establish the model. Ten-fold cross-validation and out-of-bag error were used to assess the accuracy of the models. Decision curve analysis was used to compare different models. A total of 279 cases and 253 controls were recruited, with mean age of 60.7 ± 13.7 and 42.6 ± 20.2, respectively. The regression model selected nine variables with Kappa of 0.77, sensitivity of 0.90, and specificity of 0.87. The random forests retained eight variables with Kappa of 0.76, sensitivity of 0.87, and specificity of 0.89. Based on decision curve analysis, the ABC-based diagnostic score provided the largest net benefit compared to other models with less data input. Subgroup analysis by excluding severely ill cases showed a decreased but acceptable sensitivity (0.75) and unchanged overall accuracy. The ABC-based diagnostic score provides a rapid and valid method for COVID-19 screening, especially for regions with a shortage of facilities. The model performs less well in the diagnosis of mild-to-moderate COVID-19.