Statistical analyses were conducted with SPSS software (version 23.0; SPSS, Chicago, IL, USA). No formal sample size estimation was made because there has not been any published nationwide data on COVID-19. Nonetheless, our sample size was deemed sufficient to power the statistical analysis given its representativeness of the national patient population. Continuous variables were presented as mean±sd or median (interquartile ranges (IQR)) as appropriate, and the categorical variables were presented as counts and percentages. Since no random sampling was conducted, all statistical analyses were descriptive and no p-values were presented for the statistical comparisons except for the Cox proportional hazards regression model. Cox proportional hazards regression models were applied to determine the potential risk factors associated with the composite end-points, with the hazard ratio and 95% confidence interval being reported. Our findings indicated that the statistical assumption of proportional hazards analysis was not violated. Moreover, a Cox regression model was considered more appropriate than a logistic regression model because it took into account the potential impact of the various durations of follow-up from individual patients. Age and smoking status were adjusted for in the proportional hazards regression model because they had been recognised as the risk factors of comorbidities even in the general population. Smoking status was stratified as current smoker, ex-smoker and never-smoker in the regression models.