Covid-19 vaccination and pregnancy: a systematic review of maternal and neonatal outcomes

Inclusion and exclusion criteria

We searched for either Randomized Controlled Trials (RCTs) or Observational prospective and retrospective studies comparing the maternal and neonatal adverse events of Covid-19 vaccination (any vaccine, any dose) during pregnancy (any trimester). Our study group included vaccinated pregnant women and our control group unvaccinated pregnant women. Our primary outcomes were Thromboembolic events (both maternal and neonatal), Preterm delivery, stillbirth. Our secondary outcomes were SARS Cov-2 infection, Birthweight, Small-for-Gestational Age (SGA) neonates <5th and 10th percentile, Neonatal death, Maternal hemorrhage. We excluded studies that assessed serum antibodies’ levels after vaccination or SARS Cov-2 infection. Table 1 shows our study’s exact inclusion and exclusion criteria.

Search strategy and sources

The research strategy was designed based on the Peer Review of Electronic Search Strategies (PRESS) checklist [27] using free text and Medical Subject Heading (MeSH) terms and their synonyms. Search terms were “pregnancy”, “covid-19" and “vaccines” with synonyms and alternatives. No more terms were used since we decided to conduct a widely open search to gather all potentially appropriate studies. No filters, geographical, publication status, language, and year restrictions were applied.

The following databases were searched by two reviewers (C-GK, GK) independently: Medline-Pubmed, Scopus, Cochrane Library, Clinicaltrials.gov, openGrey, and EU Clinical Trials Register. PROSPERO (International Prospective Register of Systematic Reviews) database was likewise searched for ongoing SRMAs. Abstracts of conferences and meetings of Neonatal, Perinatal, and Obstetric Societies were also searched. The last searches were conducted on the 28th of May 2022.

Study selection and data extraction

Two reviewers (C-GK, GK) conducted study selection and data extraction separately. In case of discrepancies, a third reviewer (GM) was involved, settling the issue through discussion and consensus. Mendeley© (v.1.19.8) was used as a reference manager, and duplicates were removed. Predefined collection forms proposed by Cochrane collaboration for Intervention Reviews [28] were used for data extraction. In case of questions about study eligibility or data provided by the studies, the paper authors were contacted. The reviewers initially conducted pilot-calibration exercises for this process.

A list of the outcomes and variables for which data were sought was predefined (Supplementary Material, Table 1).

Definitions

Thromboembolic events (both arterial and venous) were a composite outcome, including myocardial infarction, stroke, deep venous thrombosis, and pulmonary embolism (obstetrical) [29]. Preterm delivery was defined as the delivery before 37 completed weeks of gestation or fewer than 259 days since the first day of the woman’s last menstrual period [30] [Late preterm: between 34 0/7 and 36 6/7 [31]; Moderate preterm: between 32 0/7 and 33 6/7 [31]; Very preterm: between 28 0/7 and 31 6/7 [30]; Extremely preterm: <28 completed weeks of gestation [30]. Stillbirth was defined as the death after the 20th gestational week [32, 33].

SARS Cov-2 infection was defined as a positive real-time polymerase chain reaction (PCR) test result acquired from nasopharyngeal swabs [12]. Birthweight was defined as the weight of the newborn in grams. Birthweight was defined as low if it was less than 2,500 g. It was defined as extremely low if it was less than 1,500 g [34]. SGA neonates were defined as birthweight below the 5th or 10th percentile for gestational localized anthropometric newborn curves (World Health Organization 1995) [34]. Likewise, defined as >2 standard deviations below the mean for weight and/or length for gestational age [34]. Neonatal death was defined as the death of the newborn during the first month of life (1–30 days). Early neonatal death was defined as the death of the newborn during the first week of life (1–7 days). Late neonatal death was defined as the death of the newborn between the 8th and 30th day of life [32, 33]. Hemorrhage (obstetrical) was defined as a hemorrhagic macroscopic event that led to a loss of 1000 mL of blood or more [35].

Risk of bias assessment

We used the ROBINS-I Cochrane Tool for assessing the risk of bias in non-randomized studies [36]. Studies with low/moderate risk of bias were included in the quantitative synthesis. A sensitivity analysis was conducted for studies with serious/critical risk of bias. Graphics visualizing the risk of bias were created using the Robvis tool [37]. Two reviewers (C-GK and GK) independently conducted the risk of bias assessment, and the experienced third reviewer (GM) settled any discrepancies.

Synthesis

The treatment effect of all outcomes (primary and secondary) was measured using mean/median, SD/IQR with 95% Confidence Interval (CI) for quantitative data, and Odds Ratio (OR) with 95% CI for dichotomous data. The above were calculated using RevMan (v.5.4.1) software.

First, a robust qualitative synthesis was conducted. Second, we conducted a quantitative synthesis with RevMan (v.5.4.1). Different forest plots were created. Statistical heterogeneity was evaluated using the Higgins I² test and Chi-Squared Cochran Q-test (α=0.1). When I² was over 75%, quantitative synthesis was prohibited. Mantel-Haenszel method was applied for dichotomous data and the Inverse Variance method for quantitative data. The fixed-effects model was adopted if I²<50%, while the random-effects model was applied if I²≥50% (or p>0.1). In case of persistent statistical heterogeneity, we removed the studies creating this heterogeneity by conducting sensitivity analysis (excluding studies of high risk of bias). Subgroup analyses were performed based on the type of vaccine. In case of missing data, we tried to contact the authors (maximum 3 times) by e-mail. No imputation method was necessary.

Publication bias was planned to be assessed if ≥10 studies were available per outcome. RevMan 5.4.1 was planned to be used to create funnel plots. R-Studio (v.1.2.50.42) was planned to be used to conduct the Egger’s and rank correlation test.

Quality of evidence

An assessment of the quality of evidence for each outcome was performed separately by two reviewers (C-GK and GK) using the GRADE reporting system (Grading of Recommendations Assessment Development and Evaluation System) [38]. Any discrepancies were resolved by the third reviewer (GM). The assessment was conducted using the online tool GRADEpro GDT [39].

Study selection

Figure 1 depicts the study selection process. 2947 studies were potentially appropriate after the duplicates’ removal. 2580 were excluded by screening the title and the abstract, and 367 were full text examined. Seven observational cohort studies (four retrospective and three prospective) were finally selected [40], [41], [42], [43], [44], [45], [46], enrolling 67,274 women, from which 19,871 vaccinated pregnant women were matched to 24,328 unvaccinated pregnant controls. Studies that appear to meet the inclusion criteria, but were finally excluded, are presented in Supplementary Material, Table 2 [47], [48], [49], [50], [51], [52], [53], [54].

Figure 1:

Study selection process.

Study characteristics

Table 2 summarizes the characteristics of the included studies.

Five studies were conducted in Israel [40], [41], [42], [43, 46], one in the USA [44], and the last one in the UK [45].

All the included studies administered the BNT162b2 (Pfizer-BioNTech) vaccine. However, five studies included exclusively women vaccinated with the BNT162b2 (Pfizer-BioNTech) [40], [41], [42], [43, 46], while only two included women vaccinated with two or more different vaccines. Blakeway et al. included women vaccinated with BNT162b2 (Pfizer BioNTech) and AZD1222 (Oxford AstraZeneca) [45], and Theiler et al. enrolled women vaccinated with BNT162b2 (Pfizer-BioNTech), mRNA-1273 (Moderna), and Ad26.COV2-S (Janssen Biotech) [44].

Regarding the populations, two studies included only women vaccinated at the 2nd and 3rd trimesters [45, 46], while three studies included women vaccinated at any trimester [40, 41,44]. Two of the included studies did not define the trimester of vaccination [42, 43]. Three studies did not specify the specific number of vaccinated participants who received one or two doses [40, 42, 43].

Concerning the primary outcomes, thromboembolic events were assessed by Theiler et al. and Goldshtein et al. (pulmonary embolism) [42, 44]. Theiler et al. also assessed the incidence of stroke [44]. Preterm delivery (<37 gestational weeks) was examined by four studies [40], [41], [42, 44], while Theiler et al. assessed late preterm deliveries (34 0/7–36 6/7), preterm deliveries between 24 0/7 to 31 6/7 gestational weeks, and deliveries below the 24th gestational week [44]. Stillbirth ≥ 24 gestational weeks was assessed by Theiler et al. and Blakeway et al. [44, 45], while Goldshtein et al. assessed stillbirths ≥20 gestational weeks [42].

Regarding the secondary outcomes, SARS Cov-2 infection rates were assessed by five studies [41], [42], [43], [44], [45], and hospitalization rates by only one [42]. Birthweight was examined by four studies [40, 42, 45, 46]. Theiler et al. divided the birthweight to “low birthweight” (<2,500 g) and to “extremely low” (<1,500 g) [44]. Wainstock et al. also studied the incidence of SGA neonates <5th percentile [46]. Likewise, Blakeway et al. examined SGA neonates’ <10th percentile rates [45]. Early neonatal death (within the first 7 days of life) was addressed by Theiler et al. [44], while no study assessed total neonatal death rates as an outcome. Hemorrhage as an outcome was investigated in different forms: Bleicher et al. examined antepartum hemorrhagic events [41], and Wainstock et al. the postpartum ones [46]. Theiler et al. assessed the transfusion rates, the postpartum bleedings which needed transfusion (as a composite outcome), and the incidence of any maternal hemorrhage with blood loss greater than 1L [44]. The latter was also examined by Blakeway et al. [45].

Risk of bias assessment

The ROBINS-I Cochrane Tool [36] for the risk of bias assessment of observational studies was used for the assessment. The results (traffic light plots and weighted summary plots) are presented in Table 3 and Figures 2 and 3.

Figure 2:

Traffic light plot for the risk of bias assessment of the included studies – all outcomes (ROBINS-I cochrane tool).

Figure 3:

Weighted summary plot for the risk of bias assessment of the included studies – all outcomes (ROBINS-I cochrane tool).

Most issues were met in the “confounding”, “measurement of outcomes” and “selection of the reported result” domains. Overall, as far as the outcomes of interest are concerned, Bleicher et al. is of “Serious” risk of bias (<25%) [41], and the other studies are of “Moderate” risk of bias (>75%) [40, 42], [43], [44], [45], [46]. Bleicher et al. recruitment method, which was based on social media accounts, negatively affects the “selection of participants” domain. Parallelly, all studies have issues regarding “confounding”, as they are observational, and the matching process was not always successful since population imbalances are present. Besides, all studies were open-label and not blinded, affecting the “measurement of outcomes” domain. However, the methods used are comparable across the intervention groups, which somehow “fixes” this parameter. Due to the nature of the intervention (vaccine), bias due to “deviations from intended interventions” is low.

Qualitative synthesis

SARS Cov-2 infection

Theiler et al. and Bleicher et al. noted a statistically significant decrease of SARS Cov-2 infections in the vaccinated group. Theiler et al. noted 2/140 in vaccinated vs. 210/1862 cases in unvaccinated (p=0.03). Bleicher et al. noted 3/202 in vaccinated vs. 8/124 cases in unvaccinated (p=0.024) [41, 44]. Blakeway, on the contrary, found no statistically significant difference [45].

SARS Cov-2 infection probability (118 cases in the vaccinated group, and 202 in the unvaccinated) in Goldshtein et al. differs significantly between vaccinated and unvaccinated women (Odds Ratio OR 0.58, 95% CI 0.46 to 0.73, p<0.001). Goldshtein et al. also noted a non-significant difference between the two groups during the first ten days after vaccination (adjusted Hazard Ratio aHR 0.96, 95% CI 0.69 to 1.33, p=0.79), and a significant 54% hazard reduction in the vaccinated group during 11–27 days after vaccination, and 78% hazard reduction during 28 days or more after vaccination (respectively: aHR 0.46, 95% CI 0.31 to 0.67, p<0.001, and aHR 0.22, 95% CI 0.11 to 0.43, p<0.001). Regarding infections, 105/118 in the vaccinated group and 179/202 in the unvaccinated group were symptomatic, while 13/118 infected women in the vaccinated group and 23/202 women in the unvaccinated group needed hospitalization. The latter results were not statistically significant [42]. Dagan et al.’s results are analogous; cumulative incidence curves in both groups are similar until the 14th day after vaccination. After the 14th day the infection rates in the vaccinated group started to decline sharply. Vaccine effectiveness was 67% (95% CI 40–84%) in 14–20 days after the first dose, 71% (95% CI 33–94%) in days 21–27 after the first dose, and 96% (89–100%) in days 7–56 after the second dose. Likewise, SARS Cov-2 infection probability differs significantly between vaccinated and unvaccinated women (Odds Ratio OR 0.55, 95% CI 0.45 to 0.68, p<0.001) [43].

Quantitative synthesis

For the quantitative synthesis, we created two subgroups: those studies assessing BNT162b2 (Pfizer-BioNTech) exclusively (“BNT162b2 Pfizer-BioNTech” subgroup), and those assessing multiple types of vaccines (“Multiple Vaccines” subgroup). Since Bleicher et al.’s study was of serious risk of bias, it was not included in the quantitative analysis. Publication bias was not conducted since less than ten studies were available for each outcome.

In the “BNT162b2 Pfizer-BioNTech” subgroup, we note a statistically significant decrease of 44% of the odds of infection among vaccinated women compared to unvaccinated women (n=36,782, pooled OR 0.56, 95% CI 0.48 to 0.66, p<0.001, I²=0%). In the “Multiple Vaccines” subgroup, we note a statistically significant decrease of 79% of the odds of infection among vaccinated women compared to unvaccinated women (n=40,112, pooled OR 0.21, 95% CI 0.08 to 0.57, p=0.002). However, we note a significant statistical heterogeneity in this subgroup (I²=81%) and overall (I²=72.6%). After sensitivity analysis, we removed the study, which created the statistical heterogeneity (in the above subgroup and overall). The rate of infections among the vaccinated women compared to unvaccinated is significantly reduced by 43% (n=38,110, pooled OR 0.57, 95% CI 0.49 to 0.66, p<0.001). The overall statistical heterogeneity is I²=0% (Figure 4).

Figure 4:

SARS Cov-2 infection odds. Before (A) and after (B) sensitivity analysis.

The pooled effect estimate for preterm deliveries, stillbirth, birthweight, SGA neonates, and maternal hemorrhage is not statistically significant. The forest plots are available in Supplementary Material (Supplementary Material, Figures 1–5).

Strength of evidence GRADE reporting system

We defined the number of studies for each outcome, the studies’ design, and we evaluated the inconsistency across studies measuring the same outcome. We also assessed if the outcome was examined directly across studies (indirectness), if the results were precise enough across studies (imprecision), the presence/absence of publication bias, of a large effect, of plausible confounding, and of a dose-response gradient. Results are depicted in Table 4.

The only outcome that was judged as of “High” strength of evidence was the rate of covid-19 infections after vaccination only with BNT162b2 (Pfizer-BioNTech). The following outcomes were judged as of “Moderate” strength of evidence: Covid-19 infection rates, and Birthweight (in grams), while Preterm delivery (<37 gestational weeks), Stillbirth, and SGA (<10th percentile) were judged as of “Low” strength for evidence. Postpartum hemorrhage and thromboembolic events were judged as of “Very low” strength of evidence”. We note that the assessment was conducted only for the “Moderate” and “Low” risk of bias studies.

Limitations and strengths

This systematic review has some limitations. First, no clinical trials addressing this issue were found. As a result, we only included observational studies (both retro- and prospective), which are de facto prone to biases. All the studies included were non-randomized and not blinded. Furthermore, the number of included studies was small. The interventions and populations were mixed; the use of different types of vaccines and administering a different number of doses is indeed problematic, while many studies marked population imbalances. The administration of the vaccines’ doses mainly concerned the 2nd and the 3rd trimester, failing thus to measure the impact of the 1st trimester’s administration. Moreover, studies that tried to assess the incidence of outcomes between populations who were administered one or two doses, failed to do so due to insufficient power to detect differences. Our systematic review failed to thoroughly investigate vaccination’s impact on the incidence of our primary outcome (thromboembolic events). Although concerns were reported globally concerning the association between vaccines against SARS Cov-2 and thromboembolic events, only two studies assessed the incidence of similar events [42, 44], limiting the possibility to make safe conclusions for pregnant populations. Besides, we failed to acquire preliminary data from an ongoing study [50]. We were informed that Pfizer currently conducts a phase 2/3 trial among pregnant women, but no results were likewise available [23].

Besides, in pregnant women, the assessment of the vaccine’s effects may be biased since, generally, pregnant women are more precautious, especially during the pandemic, as far as social distancing is concerned, independently from their vaccination status. Most studies were conducted during lockdown periods or periods with implemented social distancing. This fact may attenuate the vaccine’s beneficial effect. Unvaccinated women are more prone to undergo testing more often, increasing the incidence of SARS Cov-2 infections among them. Parallelly, herd immunity may protect both vaccinated and unvaccinated populations. It must be noted that socioeconomic differences are observed between vaccinated and unvaccinated pregnant women, suggesting that vaccination status is not entirely due to hesitancy [44]. It is important to underline that obstetric events and neonatal complications have a psychological impact on women, influencing their desire to participate in this type of study.

Despite the above limitations, this systematic review has some essential strengths. We strongly believe that the studies included are the best possible that a researcher could find. Our work is genuinely comprehensive; the research question was specific, and our search strategy was broad to secure the inclusion of every appropriate study that could potentially answer our research question. The methodology followed was proposed by the Cochrane Handbook, while no violations of our protocol were noted. The risk of bias assessment was conducted using the appropriate tools, while sensitivity and subgroup analyses were performed wherever a quantitative synthesis was conducted. Finally, to our best knowledge, this is the first systematic review addressing the issue of maternal and neonatal adverse events on vaccinated pregnant women compared to unvaccinated.

Implications for the future

The above results need to be confirmed by larger studies with more homogeneous populations, especially regarding the administration of SARS Cov-2 vaccines during the 1st trimester of pregnancy. The larger populations will permit the observation of more minor differences between potential subgroups of the pregnant population (two doses vs. one dose, three doses vs. two, administration during specific trimesters, same vaccine type, etc.), and the eradication of potential population imbalances. During the conduction of this systematic review the third booster dose was proposed for those pregnant women who are immunocompromised [16,25]. Thus, the need to evaluate the safety and efficacy of the third dose in this specific subgroup is essential. In addition, a newer vaccine NVX-CoV2373 (Nuvaxovid by Novavax), which contains the spike protein of SARS Cov-2, was approved by the European Medicines Agency (EMA) [56]. This vaccine may be used among pregnant women, and new studies should address this issue.

Apart from the above, pregnant women’s vaccination may protect the fetus and the newborn. Beharier et al. reported that protective antibodies were present in the fetal circulation 15 days after the administration of the first dose. At the same time, SARS Cov-2 infection likewise led to fetal humoral immunity at delivery [40]. Humoral immunity was significantly lower in vaccinated pregnant women compared to vaccinated non-pregnant women [47]. Binding and neutralizing antibodies were found in breast milk and the cord blood of infants born to mothers vaccinated against Covid-19 with mRNA vaccines [57]. Lactating women who were vaccinated showed an increase in SARS Cov-2 antibodies in their breast milk [58], [59], [60]. More studies need to define further the immunity duration and the significance of these results.