Article Text
PDF
Abstract
Objective We conducted a systematic review to evaluate associations between influenza vaccination during pregnancy and adverse birth outcomes and maternal non-obstetric serious adverse events (SAEs), taking into consideration confounding and temporal biases.
Methods Electronic databases (Ovid MEDLINE ALL, Embase Classic+Embase and the Cochrane Central Register of Controlled Trials) were searched to June 2021 for observational studies assessing associations between influenza vaccination during pregnancy and maternal non-obstetric SAEs and adverse birth outcomes, including preterm birth, spontaneous abortion, stillbirth, small-for-gestational-age birth and congenital anomalies. Studies of live attenuated vaccines, single-arm cohort studies and abstract-only publications were excluded. Records were screened using a liberal accelerated approach initially, followed by a dual independent approach for full-text screening, data extraction and risk of bias assessment. Pairwise meta-analyses were conducted, where two or more studies met methodological criteria for inclusion. The Grading of Recommendations, Assessment, Development and Evaluation approach was used to assess evidence certainty.
Results Of 9443 records screened, 63 studies were included. Twenty-nine studies (24 cohort and 5 case–control) evaluated seasonal influenza vaccination (trivalent and/or quadrivalent) versus no vaccination and were the focus of our prioritised syntheses; 34 studies of pandemic vaccines (2009 A/H1N1 and others), combinations of pandemic and seasonal vaccines, and seasonal versus seasonal vaccines were also reviewed. Control for confounding and temporal biases was inconsistent across studies, limiting pooling of data. Meta-analyses for preterm birth, spontaneous abortion and small-for-gestational-age birth demonstrated no significant associations with seasonal influenza vaccination. Immortal time bias was observed in a sensitivity analysis of meta-analysing risk-based preterm birth data. In descriptive summaries for stillbirth, congenital anomalies and maternal non-obstetric SAEs, no significant association with increased risk was found in any studies. All evidence was of very low certainty.
Conclusions Evidence of very low certainty suggests that seasonal influenza vaccination during pregnancy is not associated with adverse birth outcomes or maternal non-obstetric SAEs. Appropriate control of confounding and temporal biases in future studies would improve the evidence base.
- PUBLIC HEALTH
- Epidemiology
- Infection control
- IMMUNOLOGY
Data availability statement
All data relevant to the study are included in the article or uploaded as online supplemental information. All relevant data are provided within the manuscript and its supporting files. Our supplemental files include additional detail regarding the methods and findings of our review, and we also have provided spreadsheets containing all of our gathered data.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
STRENGTHS AND LIMITATIONS OF THIS STUDY
We used robust systematic review methods and restricted meta-analyses to studies that adjusted for confounding and accounted for temporal biases while focusing syntheses on outcome definitions that matched accepted criteria.
The available evidence was associated with limitations related to between-study inconsistencies in methods for adjustments for confounding and immortal time bias and variability in outcome definitions.
Outcomes assessed using the Grading of Recommendations, Assessment, Development and Evaluation framework were judged to have very low certainty of evidence.
Introduction
Maternal influenza vaccination during pregnancy can reduce influenza and associated complications among pregnant women and newborn children throughout the first months of life.1 The WHO has recommended vaccination of pregnant women since 2005,2 and many countries have adopted guidance advising vaccination of all pregnant women with a licensed, recommended, age appropriate, inactivated influenza vaccine, regardless of trimester.3–8 The safety profile of influenza vaccines during pregnancy is considered favourable9; however, the evidence base is continually evolving, warranting ongoing synthesis of new data.
Most studies on this topic have used observational designs. Such studies are susceptible to bias, including confounding bias, if appropriate methods are not used to control for critical confounders (eg, maternal age, sociodemographic characteristics), as well as temporal biases, such as immortal time bias, if appropriate study design or statistical methods are not used.10–13 Confounding and immortal time bias can substantially impact both the magnitude and direction of results if not adequately addressed.13 14 Additionally, other complex temporal issues may further distort results, such as the seasonality of the influenza disease in many countries, the restricted availability of influenza vaccinations to specific times of the year, annual changes to the viral strains circulating and present in vaccines, and reduced opportunity for vaccination during pregnancy in women with shorter gestations (ie, preterm birth (PTB)).
Previous systematic reviews of influenza vaccination during pregnancy have generally used all-or-nothing approaches to synthesis and have elected to either (A) pool all data, often without considering study design, interventions (eg, multivalent seasonal influenza vaccines and monovalent 2009–2010 A/H1N1 influenza pandemic vaccines), confounding bias, immortal time bias or reporting format15–18 or (B) summarise all data descriptively due to concerns about heterogeneity of study designs, interventions and reporting formats reducing the validity of pooled results.19 20 Neither approach is ideal—robust strategies are needed to restrict meta-analysis to studies meeting important methodological and statistical considerations on an outcome-by-outcome basis and to address pooling of different measures of effect. We conducted a systematic review and meta-analyses to address the following review question, taking into consideration the potential impacts of confounding and temporal biases, and stratifying by seasonal and 2009 pandemic A/H1N1 influenza vaccines: Are influenza vaccines received at any time during pregnancy safe for women and their unborn children, compared with no influenza vaccination?
Methods
Protocol, registration and reporting
A review protocol was developed a priori, guided by established systematic review methodology,21 and registered on the Open Science Framework (OSF; doi 10.17605/OSF.IO/XEY2K) and with PROSPERO (CRD42020159030). The original protocol included a second review question regarding the efficacy/effectiveness of influenza vaccination during pregnancy against influenza outcomes; the focus of this manuscript is solely on vaccine safety. A full technical report describing our findings for both review questions is available on the OSF (https://osf.io/xey2k/). A summary of minor protocol amendments has been provided in online supplemental file 1, section 1.
Supplemental material
This manuscript has been written in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-analysis 2020 statement.22
Patient and public involvement
No patients or the public were involved in the performance of this systematic review.
Study eligibility criteria
Table 1 summarises the study eligibility criteria. We sought observational studies that assessed the safety of any seasonal or pandemic influenza vaccine not contraindicated for pregnant women, given at any time during pregnancy. Randomised controlled trials (RCTs) were excluded for this review of safety outcomes, given the rarity of most birth outcomes of interest (ie, RCTs were underpowered for these outcomes23) and given the different settings of RCTs (lower-to-middle-income countries) on this topic compared with observational studies (high-income countries)14 that would increase clinical and statistical heterogeneity. All birth outcome data from RCTs have previously been pooled in other recent publications23 24 and are summarised in the Discussion section.
Study eligibility criteria
Description of methods
A brief description of the most pertinent review methods is provided in table 2, while a complete description of our methods is reported in online supplemental file 1, section 2. Briefly, we searched multiple literature databases, using a comprehensive search strategy. A liberal accelerated approach was used for title/abstract screening, while dual independent methods were used for full-text screening, data extraction and risk of bias (ROB) assessment. Pairwise meta-analyses were conducted, where two or more studies met strict inclusion criteria for meta-analysis, including accounting for confounding and/or temporal biases. ORs and risk ratios (RRs) were considered to approximate each other, given the rarity of events and small anticipated effects of vaccination. Assessment of the certainty of evidence followed the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) framework.25 Outcomes reported in the main manuscript were prioritised according to GRADE methods as detailed in table 2, and syntheses for these outcomes have been reported in the main text, along with secondary syntheses that provide context to the prioritised findings.
Overview of review methods
Supplemental material
Supplemental material
Supplemental material
Results
Availability of the literature
Of 9443 unique references identified by our searches, 504 full texts were assessed and 93 were included in the original review. Eight RCTs reported in 16 references,1 26–40 8 cohort studies41–48 and 6 case–control studies49–54 were removed because they focused solely on vaccine efficacy/effectiveness outcomes, leaving 63 studies in the current review of vaccine safety (figure 1; 55 cohort studies13 55–108 and 8 case–control studies109–116). Citations of the 411 studies excluded at full-text screening have been provided in online supplemental file 1, section 12, along with reasons for exclusion.
Flow of evidence in a systematic review of the safety of influenza vaccination during pregnancy.
Study characteristics
Twenty-nine studies compared a seasonal influenza vaccine to no vaccine and are the focus of the main text (trivalent inactivated influenza vaccine (TIIV), n=2455 57 58 61 68 77 83–94 97 102 103 108 110 113; TIIV or quadrivalent inactivated influenza vaccine (QIIV), n=564 104 111 115 116). Thirty-four studies that evaluated other vaccines (eg, 2009 monovalent A/H1N1 pandemic influenza vaccine) or vaccine combinations, or had influenza vaccine comparators have been summarised in online supplemental file 1, sections 8–11.13 56 59 60 62 63 65–67 69–76 78–82 95 96 98–101 105–107 109 112 114
Characteristics of the 29 TIIV/QIIV studies are presented in table 3, and an evidence map of reported outcomes has been provided in the online supplemental file 1, section 13. Five were case–control studies110 111 113 115 116 and 24 used a cohort design.55 57 58 61 64 68 77 83–94 97 102–104 108 Most studies were conducted in high-income countries (ie, Australia, Canada, Japan, the USA; n=25),55 57 61 64 68 77 83–85 87–92 94 97 102 104 108 110 111 113 115 116 while three were conducted in lower-middle-income countries (ie, India, Laos, Nicaragua),58 93 103 and one in South Africa, an upper-middle-income country.86 Sample sizes ranged from 346 to 407 745 women (nmedian=11 225). All but two studies61 88 were published after 2010 (range: 2004–2020).
Characteristics of the 24 cohort studies and 5 case–control studies that compared seasonal influenza vaccination (TIIV and or QIIV) to no vaccination
Eight studies (28%) reported no adjusted analyses.61 83–85 91 92 102 103 In the remaining studies, confounding was most commonly addressed with multivariable regression models that included either (A) potential confounders as covariates,64 68 77 83 87–89 93 94 108 110 111 113 116 (B) propensity score as a single covariate55 57 64 or (C) both individual and propensity score covariates.97 Substantial heterogeneity was observed in the sets of covariates used to adjust for confounding. All 15 studies included in the prioritised syntheses reported below adjusted for critical confounders (ie, maternal age, smoking during pregnancy and socioeconomic status (SES)), except 3 cohort studies evaluating small-for-gestational-age (SGA) that did not adjust for smoking during pregnancy58 90 93 and 1 case–control study evaluating stillbirths that did not adjust for SES.116 Additional details can be found in online supplemental files 2–4.
ROB assessment
Across all studies, the most prevalent ROB concerns were related to lack of appropriate adjustment for confounding (‘comparability bias’ in 62% of studies; figure 2). Twenty-one studies (72%) were judged to have an overall low ROB, while eight were judged to have a moderate ROB. A domain-level summary of ROB by study and the original outcome-level ROB data files have been provided in online supplemental file 1, section 6 and online supplemental files 2–4.
Summary of risk of bias assessments by domain of the Newcastle Ottawa Scale for observational studies comparing seasonal influenza vaccine (TIIV and/or QIIV) to no vaccine, when safety outcomes with the lowest risk of bias were included. QIIV, quadrivalent inactivated influenza vaccine; TIIV, trivalent inactivated influenza vaccine
Syntheses of prioritised outcomes
Table 4 summarises all synthesis findings for the prioritised outcomes. Numbers of mothers or infants included in individual study analyses and overall in each meta-analysis are provided in the associated forest plots (see figures 3 and 4).
Summary of findings table of GRADE-prioritised outcomes synthesised from 29 included observational studies that compared seasonal influenza vaccination (TIIV and or QIIV) to no vaccination
Forest plots of studies reporting adjusted analyses for preterm birth, by timing of vaccine administration.*Denotes outcomes for which meta-analysis could not be conducted. Forest plots for these outcomes illustrate effects reported in studies that were descriptively summarized. Risk of bias was assessed using the 9-point Newcastle-Ottawa Scale. Higher values are associated with lower risk of bias. Numbers followed by plus signs (+) are lower than the actual total numbers included in the meta-analysis due to unreported raw data in one or more studies. aOR, adjusted odds ratio; aRR, adjusted risk ratio; NR, not reported; ROB, risk of bias; TIIV, trivalent inactivated influenza vaccine.
Forest plots of studies reporting adjusted analyses for the prioritised adverse birth outcomes. ITB, immortal time bias; NR, not reported; ROB, risk of bias; TIIV, trivalent inactivated influenza vaccine. Numbers followed by plus signs (+) are lower than the actual total numbers included in the meta-analysis due to unreported raw data in one or more studies.
Preterm birth
Fifteen cohort studies assessed TIIV at any time during pregnancy on the occurrence of PTB (<37 completed gestational weeks; table 4).55 57 58 63 64 68 77 83 86–88 90 93 94 108 Five studies57 63 64 87 108 accounted for immortal time bias and confounding and were included in the prioritised meta-analysis, while nine studies55 58 68 77 83 88 90 93 94 were pooled in a secondary synthesis to evaluate differences in pooled results when temporal biases were not considered. A final study was descriptively summarised in online supplemental file 1, section 5 due to insufficient details about its analytic methods.86
Prioritised synthesis
The estimated adjusted HR (aHR) of PTB for women who received TIIV during pregnancy was 1.08 (95% CI 0.89 to 1.30, I2=28.2; 5 studies, >152 476 women; figure 3). The evidence was assessed to be of very low certainty due to concerns regarding to ROB and inconsistency (table 4). One large high-quality study108 that reported an association close to the null value of 1 was weighted heavily (55.6%; n=145 867). There was a trend of decreasing effect size over time, based on date of publication.
Secondary syntheses
In the meta-analysis of nine studies (n>151 032) that did not consider temporal biases, the estimated adjusted RR (aRR) of PTB for women who received TIIV during pregnancy was 0.86 (95% CI 0.76 to 0.97, I2=71.1; figure 4, Any trimester, lower plot). Additionally, statistical heterogeneity was high, potentially due to differences in the background risk of PTB (range: 6%–16%) in different study populations and variation in the outcome definitions (eg, lower limits of gestational age at birth).
Six studies (n>47 211) reported adjusted findings for first trimester vaccination and were included in a meta-analysis.57 58 64 87 90 104 The lower gestational age limit of the PTB definitions ranged from >20 weeks to ≥28 weeks. All studies were at a reduced risk of temporal bias because first-trimester vaccination, by definition, occurred before the at-risk period for PTB. Four studies further reduced temporal bias through statistical methods57 64 87 104 and two did not.58 90 The estimated aHR of PTB for women who received TIIV during their first trimester was 1.05 (95% CI 0.92 to 1.20, I2=30.5; figure 4, First trimester).
Spontaneous abortion
Five studies evaluated the effect of TIIV during pregnancy on the occurrence of spontaneous abortion (SAB) under 2064 87 110 111 and 2291 gestational weeks. However, because one study did not adjust for confounding91 and different exposure windows were assessed,110 111 only two studies64 87 were pooled.
Prioritised synthesis
Both studies64 87 (n=1900) assessed an exposure window of up to 20 gestational weeks and accounted for immortal time bias. The estimated aHR of SAB for women who received TIIV during pregnancy was 0.77 (95% CI 0.31 to 1.89; I2=37.5; figure 4). This evidence was of very low certainty due to serious concerns regarding ROB, inconsistency and imprecision (table 4). Neither study restricted events to the influenza season, therefore, some of the included women may not have had the opportunity for exposure to either influenza vaccine or virus.
Secondary syntheses
Two case–control studies conducted by the same investigators used an exposure window of 1–28 days before the index date (ie, date of SAB for cases or the equivalent time postconception for controls).110 111 The initial study (n=782) was conducted during the first two influenza seasons following the 2009 A/H1N1 pandemic and found a significant association between SAB and TIIV received within 1–28 days before the index date during the first postpandemic season (2010–11 adjusted OR (aOR) 3.70, 95% CI 1.40 to 9.40), but not during the second season (2011–2012 aOR=1.40, 95% CI 0.60 to 3.30). When women were stratified by receipt of monovalent A/H1N1 2009 pandemic vaccine in the previous year, the association remained significant only for those who had received pandemic vaccine. A second case–control study (n=1934) spanning the three influenza seasons between 2012 and 2015 was conducted to evaluate the interaction between previous year vaccination and TIIV or QIIV exposure during pregnancy. No interaction was found and, overall, the estimated aOR of SAB for women who received TIIV or QIIV during pregnancy was 0.80 (95% CI 0.60 to 1.10).111 Very high statistical heterogeneity (I2=86.0%) precluded meta-analysis of the two studies.
Stillbirth
Seven studies reported data for stillbirth occurring at ≥18–22 gestational weeks or with birth weight ≥500 g.64 86 91 97 102 104 116 Two studies that evaluated vaccination at any time during pregnancy and reported results of adjusted analyses that accounted for immortal time bias were pooled97 116; however, very high statistical heterogeneity was found (I2=82.9), potentially due to differences in study designs, covariates considered to be confounders, and methods used to control confounding. The high heterogeneity suggested that pooling was inappropriate due to differences between the two studies; thus, the pooled results have not been reported and the studies have been descriptively summarised.
Prioritised synthesis
In a cohort study,97 the estimated aHR of stillbirth for women who received TIIV during pregnancy was 0.49 (95% CI 0.29 to 0.83; n=50 008), while in a case–control study,116 the estimated aOR of stillbirth for TIIV or QIIV received between 14 days after the last menstrual period and 7 days before the index date was 0.98 (95% CI 0.81 to 1.17; n=3975; did not adjust for SES; figure 4). A very low certainty of evidence was assessed due to concerns regarding inconsistency between the two studies (table 4).
Secondary syntheses
The aforementioned cohort study97 also reported stratified analyses for gestational age at birth (preterm and full term) and for timing of events with respect to the influenza season (before, during, after). Sample sizes were not reported. This study was conducted year-round in Australia and spanned two influenza seasons of differing durations. Among PTB, the estimated aHR of stillbirth for women who received TIIV during pregnancy was 0.45 (95% CI 0.26 to 0.81), but among term births, the estimated aHR of stillbirth was 1.13 (95% CI 0.27 to 4.71; figure 4). As well, in the 2.5–3 months after the influenza season, the estimated aHR of stillbirth for women who received TIIV during pregnancy was 0.33 (95% CI 0.12 to 0.88), but the effect was not significant in the 3–4 months during (aHR 0.57, 95% CI 0.25 to 1.31) or the 2–6.5 months before the influenza season (aHR 0.60, 95% CI 0.22 to 1.61; figure 4).
Another cohort study104 that accounted for immortal time bias through design and assessed only first trimester vaccination found the estimated aRR of stillbirth to be 1.18 (95% CI 0.64 to 2.18; n=11 955).
Small-for-gestational-age birth
Thirteen studies evaluated the association between TIIV received during any trimester and SGA birth.55 57 58 64 68 77 85–87 90 93 94 108 One reported findings from unadjusted analyses85 and 1 did not report statistical methods sufficiently,86 leaving 11 studies for meta-analysis.55 57 58 64 68 77 87 90 93 94 108
Prioritised synthesis
Two90 108 of 11 studies included in the meta-analysis comprised almost 75% of the pooled estimate by weight and reported aRRs of 1.00; the estimated pooled aRR of SGA for women who received TIIV during any trimester was 0.99 (95% CI 0.95 to 1.04, I2=15.8; n>297 424; figure 4). Three studies did not adjust for smoking during pregnancy.58 90 93 A very low certainty of evidence was assessed due to concerns regarding inconsistency (table 4).
Congenital anomalies
Four studies87 91 102 115 evaluated the association between TIIV or QIIV during the first trimester and the occurrence of congenital anomalies identified either at birth87 91 102 or up to 6 months of age.115 Two studies (one cohort87 and one case–control115) adjusted for confounding but used differing periods to ascertain anomalies in infants (ie, at birth and up to 6 months of age), precluding meta-analysis.
Prioritised synthesis
TIIV or QIIV during the first trimester was not significantly associated with congenital anomalies in either of the studies, whether ascertainment occurred at birth87 (aRR 0.33, 95% CI 0.04 to 2.73; n=1207) or up to 6 months of age115 (aOR 1.01, 95% CI 0.85 to 1.21; n=4277; figure 4). Both studies included both live and stillborn infants from multiple seasons.87 115 The certainty of the evidence was assessed to be very low due to ROB at both time points (table 4).
Maternal non-obstetric serious adverse events
Three cohort studies reported non-obstetric serious adverse events (SAEs).88 89 103 One study89 adjusted for confounders and found no significant association with Guillain-Barre syndrome within 42 days postvaccination (p=0.34; vaccinated: 0 events/75 906 women, unvaccinated: 1 event/147 992 women). The other two studies reported raw data only: no significant associations with non-obstetric hospitalisations within 42 days postvaccination88 (vaccinated: 2 events/225 women, unvaccinated: 3 events/826 women; RR 2.45, 95% CI 0.41 to 14.56) or with SAEs over an unknown follow-up period103 (vaccinated: 0 events/288 women, unvaccinated: 0 events/58 women). Events were rare for all outcomes and sample sizes potentially too small to detect significant differences between groups. All maternal SAE outcomes were assessed with a very low certainty of evidence due to concerns regarding imprecision (table 4).
Discussion
Our systematic review of the safety of seasonal influenza vaccination during pregnancy was designed to ensure our findings were relevant to currently used multivalent products that may have a potentially different safety profile than monovalent influenza vaccines and to ensure reporting of least biased estimates. We identified no significant harmful or beneficial effects of TIIV or QIIV on birth outcomes or maternal non-obstetric SAEs among 29 studies. However, clinical, methodological and statistical heterogeneity and lack of adjustment for confounding and/or immortal time bias precluded meta-analysis for stillbirth, congenital anomalies and maternal SAEs, and reduced the number of studies available in meta-analyses of PTB, SAB and SGA. Meta-analyses for PTB and SGA were driven by one or two heavily weighted studies each. Syntheses of the effects of 2009 A/H1N1 pandemic vaccination on adverse birth outcomes demonstrated similar results (see online supplemental file 1, section 8).
Research regarding the safety of seasonal influenza vaccination during pregnancy is ongoing, with five new studies published in 2020 alone,86 87 91 104 116 warranting continued review of the literature. Recent systematic reviews have opted to either pool all available studies, regardless of vaccine type (ie, seasonal and 2009 A/H1N1 pandemic) or the presence of confounding,15 16 or not to pool any studies due to differences in vaccine types and effect estimates and the presence of confounding bias.19 To the best of our knowledge, no previous review has attempted to reduce temporal and confounding biases. Pharmacoepidemiological studies can be biased by immortal time117 if each participant’s times of enrolment, treatment start, and event occurrence are not measured, and the duration of unexposed and exposed time not accounted for through either analytical methods10 or study design.10 11 Observational studies of vaccination during pregnancy that report time-dependent birth outcomes (eg, PTB and stillbirth10 13) are susceptible to immortal time bias if the time at risk prior to vaccination is misclassified as being exposed. Approaches, such as use of Cox proportional hazards models with a time-varying exposure variable for vaccination status correctly account for each woman’s exposed and unexposed time, reducing immortal time bias.10 13 118 Our meta-analyses for PTB demonstrated that immortal time bias can have a substantial impact on study results. When immortal time bias was not accounted for, a significant protective effect of seasonal influenza vaccination during pregnancy on PTB was observed. However, when appropriate analytical methods were used, seasonal influenza vaccination was not associated with PTB. Similar impacts of temporal biases were found in meta-analyses of the effects of 2009 A/H1N1 pandemic vaccine during pregnancy on stillbirth (see online supplemental file 1, section 8). These differences may have important implications for decision-makers and knowledge users.
Four RCTs have been conducted in pregnant women, with objectives to evaluate seasonal influenza vaccine efficacy to prevent infant lab-confirmed influenza1 36 38 40 and maternal influenza-like illness36 as primary outcomes. Data on adverse birth outcomes were also collected as secondary outcomes; however, due to the inherent rarity of these events, the individual studies were underpowered to detect differences between groups for these outcomes.23 Omer et al23 pooled individual patient data from three of these RCTs1 36 38 (one excluded due to small sample size40) and found that within the combined dataset of 10 000 women no significant associations with adverse birth outcomes were found.23 They concluded that the pooled dataset was still substantially underpowered to detect a meaningful difference between groups for the stillbirth outcome, and moderately underpowered for all other adverse birth outcomes.23 Regan and Munoz24 pooled fetal death data from the same three RCTs1 36 38 and included data from the fourth RCT37 40 in meta-analyses of PTB, LBW and SGA birth. No increased risks of any adverse birth outcomes were found. We reported two meta-analyses that were in common with the above publications,23 24 and our findings were similar: no significant associations with PTB (aHRcurrent=1.09 (95% CI 0.89 to 1.33) vs RROmer=0.97 (95% CI 0.87 to 1.08) vs RRRegan=0.95 (95% CI 0.85 to 1.06)) or SGA (aRRcurrent=0.99 (95% CI 0.95 to 1.04) vs RROmer=0.99 (95% CI 0.93 to 1.06) vs RRRegan=0.96 (95% CI 0.90 to 1.02)). Furthermore, Regan and Munoz descriptively summarised maternal adverse events postinfluenza-vaccination reported in RCTs,31 119 prospective cohort studies,120 121 and postmarketing surveillance studies122 123 and found no evidence of increased safety signals beyond those found in the general public or compared with pertussis vaccine.24 Vaccination for seasonal influenza during pregnancy is recommended by the WHO, therefore, ethical approval of a future large-scale (>10 000 women) RCT powered for birth outcomes is unlikely. It is encouraging that data pooled from methodologically sound observational studies can approximate findings from pooled individual patient data.
Our review has several strengths, including a broad search and use of robust systematic review methods. To reduce clinical heterogeneity and to align with currently available vaccines, we did not pool seasonal and monovalent 2009 A/H1N1 pandemic influenza vaccines. Entry into meta-analyses was restricted to least-biased estimates that were adjusted for confounding and accounted for temporal biases, where appropriate. As well, we prioritised synthesis of very focused outcome definitions that matched published accepted criteria.124 As a limitation, we made assumptions regarding equivalence of effect estimates (ie, RR=OR=hour); however, only the meta-analysis for SGA included heterogeneous effect estimates, and for this outcome, the assumptions behind the approximation of effect estimates were met. The available evidence was limited by inconsistent adjustment for confounding and immortal time bias and by high variability of outcome definitions and time points, which precluded pooling of some of the identified data. As well, poor reporting of detailed characteristics of the seasonal influenza vaccines in use in the study location during the study periods prevented the development of inferences regarding the effects of vaccine adjuvant or production technology (eg, egg based, cell based, recombinant). Finally, there were limitations in the use of GRADE in that only observational designs were considered for inclusion. GRADE assessment of syntheses of observational studies begins with a rating of low certainty, and evidence is only rated upward to moderate certainty if a large statistically significant effect is found. When the review objective is to assess safety (ie, no significant difference between exposure groups), large significant effects are not expected and, thus, evidence would be considered at low certainty, at best, even when well-conducted studies are included.
In conclusion, our review suggests that vaccination against seasonal influenza during pregnancy is not associated with significant safety issues, with respect to adverse birth outcomes and maternal non-obstetric SAEs. A small but non-significant association with an increased risk of PTB was found in a meta-analysis of studies that accounted for immortal time bias. This was in contrast to a significant decrease in the risk of PTB, when studies that did not account for immortal time bias were pooled. Improvements in study design and analytical methods in future studies would advance the evidence base. Additionally, although no safety signals have been identified through routine pharmacovigilance, there is limited published peer-reviewed evidence regarding the safety of administration during pregnancy of more recently licensed inactivated influenza vaccines products that are based on different vaccine technologies, including quadrivalent mammalian cell-culture-based vaccines (eg, Flucelvax Quad) and recombinant influenza vaccines (eg, Flublok, Supemtek). Similarly, there is limited evidence regarding the safety of administration of influenza vaccine during the first trimester, specifically. Ongoing primary research in these areas is necessary, warranting future systematic review to inform guidance for healthcare providers counselling pregnant women about the benefits and safety of influenza vaccination.
Data availability statement
All data relevant to the study are included in the article or uploaded as online supplemental information. All relevant data are provided within the manuscript and its supporting files. Our supplemental files include additional detail regarding the methods and findings of our review, and we also have provided spreadsheets containing all of our gathered data.
Ethics statements
Patient consent for publication
Acknowledgments
We wish to thank Raymond Daniel for his consistently excellent support toward article retrieval and citation management throughout the review. We wish to acknowledge Matthew Tunis and Kelsey Young from the Public Health Agency of Canada for their input throughout the review.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Twitter @cgarritty, @dmoher, @bh_epistat
Contributors CG and BH conceptualised the design and initial protocol for this study. DF, CH, ACT andDM provided feedback on the study protocol. BS designed and executed the literature search. DW, CG, CH, CB, MH, NA, DBR, LE, AM, CS, MG and PAK were responsible for study selection and data collection. DW and BH performed and summarised findings from data analysis. DW and BH prepared the initial draft of the manuscript. All coauthors (DW, DF, CG, CH, CB, MH, NA, DBR, LE, AM, CS, MG, PAK, AS, BS, ACT, DM and BH) reviewed and approval the final draft of the manuscript. BH is guarantor of the review.
Funding This study was funded by the Canadian Institutes of Health Research and the Drug Safety and Effectiveness Network. ACT is funded by a Tier 2 Canada Research Chair in Knowledge Synthesis.
Disclaimer The funders had no role in the design, analysis, interpretations or reporting of this research.
Competing interests BH has previously received honoraria from Eversana for methodologic advice related to the conduct of systematic reviews and meta-analysis. All remaining authors have no competing interests to declare.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.