Meta-analysis results
Three high-quality studies, all of which were from China and clearly classified COVID-19 pneumonia as severe and non-severe were used for subsequent meta-analysis. When used to study fever, these studies included in total 262 cases in the severe group (245 cases with fever, 93.51%) and 1,095 cases in the non-severe group (967 cases with fever, 88.31%), and were considered to be homogeneous as they fit the fixed effect model nicely (Chi2=0.88, P=0.64, I2=0%). The pooled effect was 1.70 (95%CI, 1.01- 2.87; Fig. 2A ), indicating that the incidence of fever in the severe group was 1.70 times higher than that of non-severe group. On symptoms of dyspnea, there were 262 cases in the severe group (112 dyspnea, 42.75%) and 1095 cases in the non-severe group (179 dyspnea, 68.32%). Q-test and I2 statistic test showed that certain heterogeneity exists (Chi2=5.26, P=0.07, I2=62%), therefore the random effect model was adopted. The individual OR effect showed inconsistently thatArticle 1 and Article 2 show the severe patients have a higher incidence of dyspnea and Article 3 shows there is no statistically significant difference, but combined OR effect was 3.53 (95%CI, 1.95-6.38; Fig. 2B ), indicating that the incidence of dyspnea in the severe group was 3.53 times higher than that of non-severe group. On symptoms of diarrhea, we included 266 cases in the severe group (25 cases with diarrhea, 9.40%) and 1,110 cases in the non-severe group (48 cases with diarrhea, 4.32%). Model test showed homogeneity among these studies (Chi2=0.32, P=0.85, I2= 0%), and therefore the fixed effect model was used. The combined OR value was 1.80 (95%CI, 1.06-3.03; Fig. 2C ), indicating the incidence of diarrhea in the severe group was 1.80 times higher than that of non-severe group. Importantly, the funnel plots of fever, dyspnea and diarrhea are all largely symmetrical, suggesting no significant publication deviation exists (Fig. 3 ).