3.1. Study selection
A total of 364 relevant documents were retrieved by the search methods above, including 103 articles from PubMed, 96 articles from Web of Science, 66 articles from EMBASE, and 99 articles from MEDLINE. We hand-searched bibliographies of retrieved papers and 2 article was included. After removing 248 duplicate documents, 82 papers were deemed ineligible after title and abstract screening, and 22 papers were excluded after further screening through full-text reading. After the exclusion of all the 352 unqualified studies, a total of 14 retrospective case analyses were included in this meta-analysis (10-23). The process of study selection is illustrated in Figure 1.
3.2. Study characteristics
We extracted the features of the literature above. The study included 236 pregnant women with laboratory-confirmed COVID -19 from December 8, 2019 to April 4, 2020 of whom 160 were in China and 76 were in America. The characteristics of the included literature are presented in Table 1.
3.3. Assessment of quality
We evaluated the quality of the fourteen included documents according to IHE case series methodological quality evaluation tool. The result was that the thirteen articles ranged from 45% to 65% respectively, all of which were lower than 70% and were of low quality. Only one article met 70%, considered acceptable quality. These articles were all retrospective studies with few cases and without control group, intervention and blind method, so they were rated as low quality. However, there were only these documents at present, we had no choice but to include them. The literature quality assessment is shown in Table 2.
3.4. Quantitative data synthesis
Because of the high heterogeneity of this study, we used random effects model. Meta-analysis results showed: the incidence of severe case or death was 12%, 95%CI: 0.03~0.20, I2 =0%, P=0.006; the incidence of fever was 51%, 95%CI: 0.35~0.67, I2=89%, P<0.00001; the incidence of cough was 31%, 95%CI: 0.23~0.39, I2=38%, P<0.00001; the incidence of lymphopenia was 49%, 95%CI: 0.29~0.70, I2=83%, P<0.00001; the incidence of positive CT findings was 71%, 95%CI: 0.49~0.93, I2=90%, P<0.00001; the incidence of coexisting disorders was 33%, 95%CI: 0.21~0.44, I2=70%, P<0.00001; the incidence of preterm labor was 23%, 95%CI: 0.14~0.32, I2=21%, P<0.00001; the incidence of cesarean section was 65%, 95%CI: 0.42~0.87, I2=90%, P<0.00001; the incidence of fetal distress was 29%, 95%CI: 0.08~0.49, I2=68%, P=0.007; the incidence of neonatal asphyxia or neonatal death or stillbirth was 9%, 95%CI: -0.03~0.21, I2 =0%, P=0.14; the incidence of neonatal infection was 12%, 95%CI: -0.01~0.26, I2=0%, P=0.06; SARS-CoV-2 test of breast milk was only mentioned in Chen H 2020.2.12, and the incidence was 0, which cannot be calculated by meta-analysis.
To sum up, the P values of neonatal asphyxia or neonatal death or stillbirth and neonatal infection were both greater than 0.05, which have no statistical significance. We couldn’t calculate the incidence of positive SARS-CoV-2 test in breast milk as well. Otherwise the P values in remaining indicators were all less than 0.05, which was statistically significant. The most common clinical features were positive CT findings (71%), cesarean section (65%), and fever (51%), followed by lymphopenia (49%), cough (31%) and severe case or death (12%). Adverse pregnancy outcomes included coexisting disorders (33%), fetal distress (29%) and preterm labor (23%), which was in descending order. Among them, the I2 value of severe case or death was 0%, which was low heterogeneity. Despite the indicator mentioned above was referred to 10 literatures respectively, the incidences of eight documents were all 0 and there were only two none-zero data of the indicator. The I2 value of preterm labor was 21%, which was low heterogeneity. The I2 value of cough was 38%, which was medium heterogeneity, and the remaining I2 value of indicators ranged from 68% to 90%, which were all high heterogeneity.
Furthermore, we carried out the subgroup analyze based on the data from the fourteen retrospective analyses of COVID-19 infection in pregnant women above and one meta-analysis of epidemiology in all patients infected COVID-19(24). All patients were divided into two subgroups, namely pregnant women and nonpregnant patients. In the fifteen articles, only two indexes, fever and cough, were coincident, which were analyzed in subgroup. The results were as follows. The incidence of fever in pregnant women group was 51%, which were significant lower than 91% in nonpregnant patients group (P<0.00001). The incidence of cough in pregnant women group was also significant lower than nonpregnant patients (31% vs 67%, P<0.0001). The forest plot of subgroup analysis is illustrated in Figure 2 and Figure 3.
3.5. Risk of publication bias
The funnel plots of fever, cough, positive CT findings and coexisting disorders are symmetric, which meant that these indicators may not be published biased. On the contrary, the funnel plots of severe case or death, lymphopenia, preterm labor, cesarean section, fetal distress, neonatal asphyxia or neonatal death or stillbirth and neonatal infection are asymmetric, which meant that the indicators above may have publication bias. Since there was only 1 article about SARS-CoV-2 test of breast milk, it was meaningless to draw a funnel plot. Therefore, the publication bias of it was not evaluated. The funnel plots of fever and cough are shown in Figure 4 and Figure 5.