2.5 Logistic regression analysis
Potential risk factors for non-survived COVID-19 patients (n = 49), compared to survived severe patients (n = 78), were analyzed by a multivariate binary logistic model using Forward Stepwise (Wald) model method. Missing values of laboratory data for the logistic regression analysis, including affected lobe number(s), CRP, PCT and D-dimer, were replaced via multiple imputation. Cut-off value of neutrophil-to-lymphocyte ratio (NLR = 7.726) was calculated via ROC analysis, with an AUC of 0.6614, and NLR was analyzed as categorical variables for the logistic regression analysis. All variables were subject to univariate logistic regression, and odds ratios (ORs) were calculated between non-survived and survived severe groups, with 95% confidence intervals generated. Variables were included in binary logistic regression if corresponding p value was less than 0.05. The binary logistic regression analysis was employed to conclude a multivariate model to conclude the risk factors of death among critically ill patients.
Univariate analysis and multivariate regression analysis were performed by SPSS software (version 26.0, IBM), and R software (version 3.4.3, supported by R Foundation for Statistical Computing).