We conducted a systematic review and meta-analysis of studies which explored the association between COVID-19 lockdowns and ABPOs in HICs. The review was registered on PROSPERO (CRD42022327448), where amendments and progress have been updated accordingly, and followed the relevant PRISMA guidance.
Eligibility criteria:
The intervention of interest was COVID-19 lockdowns, defined as those non-pharmaceutical, home-confinement, and non-essential service closure interventions imposed by governments in 2020 with the aim of reducing the spread of COVID-19 infection.
To be included in the review, studies had to report on at least one of the following ABPOs or outcome sub-classifications (definitions in Appendix A).
Adverse Birth and Pregnancy Outcomes (ABPOs)
Perinatal/Neonatal outcomes:
- Preterm birth (PTB)
- Stillbirth
- Low birth weight (LBW)
- Neonatal mortality
- Neonatal (neonatal intensive care unit) admissions
- Hypoxic Ischemic Encephalopathy (HIE)
- Prolonged stay in hospital
Maternal outcomes:
- Caesarean section
- Obstetric anal sphincter injuries (OASI)
- Peripartum hysterectomy
- Sepsis (in puerperium)
- Prolonged stay in hospital
- Readmission to hospital
- Maternal mental health (depression or anxiety antenatally or postpartum)
The outcomes of this review were decided upon based on current literature, expert advice, importance to patients, and pragmatic considerations. Patient’s perspectives and priorities were obtained from a women’s reference group, the group was formed to consult on the COVID Maternity Equality Project (CMEP) study, 15 where SI was the principal investigator. Their insights and priorities informed which outcomes were included and explored in this review and research. Pragmatic considerations included data access and availability.
The population was restricted to those residing in HICs, as defined by the World Bank Classifications. 16 To be included studies had to present a comparator cohort from before lockdowns or the pandemic in 2020. We included the following study types: observational studies, case-control studies, cohort studies, brief reports comparing outcomes before and during lockdowns. Studies had to be published between Jan 1, 2019-June 22, 2023. Language requirements entailed that title and abstracts had to be written in English to be included.
We excluded non-HIC populations or countries that did not enforce lockdowns, for example, Sweden. We excluded studies which explored only COVID-19 infection in pregnancy and neonates. We also excluded qualitative studies, systematic reviews, narrative reviews, discussion articles, viewpoints/opinion articles, and editorials.
Search strategy & selection: Search terms were developed to include all key perinatal, neonatal, and maternal health outcomes, and word variants of COVID-19 lockdowns (Appendix B). We searched EMBASE, MEDLINE/PubMed, and Web of Science. We also searched for pre-prints (MedRxiv) and unpublished reports, these were identified by co-authors through research networks. References of eligible studies were reviewed to identify additional eligible studies. Search results were exported to Endnote, where duplicates were removed, both automatically and manually. The results were then independently reviewed by two reviewers (HNS & IH), who applied the eligibility criteria first to titles, then abstracts and full texts. Reviewers then compared selections and any disagreements were resolved by discussion between reviewers and a third author arbitrated (SI).
Data extraction: Key characteristics of studies and data on outcomes and total births were extracted. Aggregated data stratified according to deprivation and ethnicity were also extracted, multiple indicators of deprivation were used (Appendix C). Where data was missing, illegible, or presented in the wrong format, we emailed authors to ask them to provide relevant aggregated data. If the authors did not respond, then the raw values were calculated based on available information, however, in cases where information was insufficient for calculations, the study was excluded from meta-analysis. Risk of bias in each study was assessed and scored using the Newcastle-Ottawa risk of bias assessment tool for cohort studies. 17 Following extraction, studies that had a cohort overlap were removed, where the smaller study was removed in favour of the larger study.
Statistical analysis: Meta-analysis was conducted on outcomes which had more than three studies presenting relevant data, to calculate accurate and meaningful pooled estimates. For each outcome, we ran random effects estimate pooled risk ratio to account for heterogeneity across studies and populations, and 95% confidence intervals were calculated, as well as an I2 value summarizing inter-study heterogeneity. 18 Where heterogeneity was considerable (i.e. >75%) and sufficient data was available, subgroup and stratified meta-analyses were conducted to explore potential sources of heterogeneity. 18
Due to the differing measures and implementation of lockdowns between countries, a high degree of spatial heterogeneity was expected and subgroup analyses were conducted according to continent. 8 Studies which reported data on a second defined lockdown period and/or a post lockdown period were also meta-analysed as subgroup analysis. The reference period for both was the pre-pandemic epoch. This analysis attempted to explore whether the effect of second lockdowns varied greatly from the first. To improve insight into lockdowns’ impact on distinct pathological processes, sub-classifications of clinical outcomes were explored where data was available in supplementary analyses, for instance, spontaneous compared to iatrogenic (medically induced) PTB. 8 There are various sub-classifications of outcomes included, these are listed with outcome definitions (Appendix A).
To explore inequalities within HICs, stratified meta-analysis according to deprivation and ethnicity was conducted where more than three studies presented sufficient data. Studies with stratified data were included and subgroup pooled estimates were calculated and compared.
We additionally controlled for underlying temporal trends in outcomes when data were available, failing to account for such trends can lead to spurious results. 8,19 Therefore, we conducted a meta-analysis of time-adjusted studies; we excluded studies without adjustment for underlying temporal trends using a quasi-experimental design. 19,20 The same pathway as the main analysis was followed for selection and data extraction. Two reviewers (BG & IH) independently screened studies previously identified as eligible for inclusion. Additional data was extracted from the studies, including total sample sizes, time-adjusted risk and odds ratios, and corresponding upper and lower confidence intervals. In papers where odds ratios were presented, they were converted to risk ratios using the formula described by Faber and colleagues. 21 Once data was extracted, an inverse-variance meta-analysis was conducted using the random-effects pooled estimate. 18
Funnel plots were constructed, and Egger’s tests run on outcomes which had data aggregated from more than ten studies to explore asymmetry and publication bias, for the Egger’s test a significance threshold of p<0.1 was adopted. All analyses were conducted in STATA 17.0.