Review stage | Details |
Literature search | As per protocol for review of both efficacy and safety outcomes, separate search strategies were developed for observational studies (MEDLINE and Embase), RCTs (Ovid MEDLINE ALL, Embase Classic+Embase and the Cochrane Central Register of Controlled Trials), and systematic reviews (OVID platform: Ovid MEDLINE ALL, Embase Classic+Embase, Cochrane Database of Systematic Reviews, Database of Abstracts of Reviews of Effects and Health Technology Assessment). Study design filters were applied for RCTs and systematic reviews. No language or date restrictions were used. Search strategies are provided in online supplemental file 1, section 3. Searches were peer reviewed using Peer Review of Electronic Search Strategies (PRESS) Checklist.125 Searches were run 11 December 2019 and updated to 3 June 2021. Bibliographies of included systematic reviews and primary studies were screened. Grey literature searching (June 2020) included clinical trial registries and other platforms, and was guided by the Canadian Agency of Drugs and Technologies in Health (CADTH) Grey Matters Checklist.126 |
Study selection | Search results were deduplicated in Reference Manager* (original search) or EndNote† (update) prior to being uploaded to online review management software DistillerSR.‡ Two levels of screening were performed (title/abstract and full text), with pilots at each level. A liberal accelerated approach was used to screen titles/abstracts (one reviewer to include, two to exclude), while dual independent screening was used for full texts. Conflict resolution was through discussion. |
Data extraction | Dual independent data extraction was performed in DistillerSR, using standardised piloted forms. Conflict resolution was reached through discussion. Data extraction files have been made available online (see online supplemental files 2–4). Study characteristics collected: author, year of publication, language, design, objective, funding, whether conflicts of interest were declared, country of conduct, study period and influenza seasons, longest follow-up, sample size, inclusion criteria, method of measurement of gestational age. Demographic data collected: maternal age, race, gestational age at vaccination, maternal comorbidities, whether specific maternal risk factors were reported, proportion vaccinated. Intervention data collected: vaccine type (seasonal or pandemic), valency, name, presence of adjuvant, strains, setting of vaccine delivery. Outcome data (all outcome definitions and time points were extracted): numbers of events and participants per group, and, if reported, point estimate with 95% CI, whether confounding was controlled and the primary approach (eg, matching, inclusion as covariates, propensity score), and the confounders that were controlled, including three critical confounders: maternal age, smoking during pregnancy, and socioeconomic status. Country income level was determined for each study as per The World Bank World Development Indicators.127 |
Risk of bias | Dual independent assessment using adaptations of the Newcastle-Ottawa Scale (NOS)128 for cohort and case–control studies, with conflict resolution through discussion. NOS scores can range from 0 to 9, with lower scores representing higher ROB. Suggested score cut-offs are as follows: low ROB (7–9), some concerns (4–6), high ROB (0–3). Details regarding the different ROB domains are provided in online supplemental file1, section 4. |
Data synthesis/ Statistical methods | Data were cleaned and collated in Microsoft Excel, and tables summarising study characteristics were developed separately for cohort and case–control studies. Results data were tabulated to assess feasibility of meta-analyses. ORs and risk ratios (RRs) were considered equivalent, given the rarity of the outcomes of interest and anticipated small effects of vaccination. HRs were considered approximate to RRs, ORs and incidence RRs, given short follow-up periods (eg, duration of gestation), rarity of events and small effects. Pairwise meta-analyses using random effects models were considered when two or more studies reported findings that were adjusted for confounding (and immortal time bias, where appropriate; see online supplemental file 1, section 2 for a brief discussion of immortal time bias) for a given outcome definition, time point and comparison, and when the available evidence was not clinically heterogeneous. Approaches to handle confounding that were of interest included matching, multivariable adjustment and propensity score methods. Where study effect estimates for meta-analysis were heterogeneous (eg, both RRs and ORs), meta-analyses were conducted and reported using RRs. Effect estimates <1 indicated reduced risk of event for vaccinated subjects (ie, greater safety). Statistical heterogeneity was assessed with the I2 statistic, where values ≥50% suggested important heterogeneity. Statistical heterogeneity was explored using stratified analyses, and studies were pooled if no important differences in study characteristics were found. All meta-analyses were conducted in Comprehensive Meta-analysis Software (V.3.3.070, Biostat, Englewood, New Jersey, USA). Descriptive summaries were developed for prioritised definitions, when meta-analysis was not feasible, guided by the Synthesis without Meta-analysis statement.129 |
Assessment of level of evidence | Strength and certainty of evidence were assessed following the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) framework,25 using GRADEpro GDT online software (https://gradepro.org/). Syntheses were prioritised for assessment based on outcome definition (as per the GAIA Network and Brighton Collaboration Network124; see online supplemental file1, section 2, intervention/comparison (trivalent inactivated influenza vaccine or quadrivalent inactivated influenza vaccine vs no vaccine), content expert prioritisation exercise, and analysis type (full-sample analyses only). Dual independent assessment was conducted, with discussion of conflicts. |
Reporting | Only findings from GRADE-prioritised syntheses and secondary syntheses that provide additional context have been reported in the main text; all other seasonal influenza vaccine results are reported in online supplemental file 1, section 5 or as raw data in online supplemental files 2–4. RoB has been summarised across studies, using each study’s primary outcome with the lowest RoB. Study-level RoB data have been provided in online supplemental file 1,section 6. GRADE lay statements129 and Summary of Findings tables have been reported in the main text, with an Evidence Profile table provided in online supplemental file 1, section 7. Syntheses and/or data for vaccines other than seasonal influenza (eg, pandemic vaccines, combinations of seasonal and pandemic vaccines) have been provided in online supplemental file 1, sections 8–11 and online supplemental files 2–4. |
*Reference Manager, V.12.0.3, Thomson Reuters.
†EndNote, V.X9.3.3, Clarivate.
‡DistillerSR, V.2.35. Evidence Partners; 2021. Accessed December 2019–November 2021. https://www.evidencepartners.com.
RoB, risk of bias.