Protocol and search strategy
The study protocol was registered on PROSPERO (CRD42020185906). A search strategy was conducted for papers published before the date of the search (May 2020) on PsycINFO, MEDLINE, Web of Science Core Collection, CINAHL and EMBASE databases. The search strategy included key-terms related to the prenatal period, maternal anxiety, maternal stress, childhood and cognitive development (see supplement materials). The literature search was limited to humans and studies published in English. There were no criteria for years of publication. Additional articles were identified using reference list from previous meta-analyses and review articles.
Inclusion and exclusion criteria
Prospective and retrospective observational studies investigating the association between symptoms of prenatal maternal anxiety and/or stress and children’s cognitive outcomes were considered for inclusion. Measures of prenatal maternal anxiety and/or stress included validated physiological measures and self-report questionnaires. The outcome measures were validated age-standardised measures of general intelligence, learning, memory (long-term free recall, long-term recognition), language (receptive, expressive), speed of information processing, attention (span, divided, switching, sustained, selective), executive functions (working memory, inhibition, planning, organisation, self-monitoring, decision making, cognitive flexibility, word generation) and academic skills (reading, maths, writing). Both direct assessment and parent and teacher-rated questionnaires were included if relevant.
Studies were excluded for the following reasons: (i) the study was a treatment or clinical trial, case-control study, review, comment, letter, thesis or book, (ii) individual cognitive skills were not assessed separately, (iii) the offspring were older than 18 years old or specifically selected based on a previous diagnosis (e.g. autism and ADHD), (iv) anxiety and/or stress was examined in the postnatal period, (v) a measure of anxiety and/or stress was not included, (vi) measures of anxiety and stress were combined in the analyses, (vii) children’s cognitive function was assessed using experimental measures only, and (iix) the study involved non-human participants.
When the same cognitive domain was measured at multiple time points in the same cohort or when different measures of a same cognitive domain were examined, preference was given to the largest sample size. When two timepoints were of similar sample size, the timepoint where children were the eldest was given preference. If the above two criteria were the same, the cognitive outcome evaluated by the most valid or widely used measure in the field was used 32,33. For the overall analysis, when a study reported multiple stress and/or anxiety exposures, preference was given to the to the exposure that was the most comparable to those included in the analysis.
All the references retrieved using the strategy search outlined above were downloaded and deduplicated into Endnote before being transferred, stored and managed into Covidence. Titles and abstracts were screened independently by two reviewers, GD and SV. Full texts of potentially relevant articles were obtained and independently assessed for inclusion by the two reviewers using the previously described criteria. In case of disagreement, the reviewers discussed eligibility. When disagreement could not be resolved, PJA made the final decision. Ninety-nine percent agreement was reached after the first screen of titles / abstracts, and 91% after the first screen of full texts. After discussion, 100% agreement was reached by the three reviewers.
A customised extraction excel sheet adapted from the Cochrane Review handbook was used to guide extraction. Reviewers (GD and SV) extracted relevant information independently and discussion was organised in case of discrepancy. When available, extracted information included:
Independent variable: Measures used to assess maternal anxiety and/or stress during pregnancy; anxiety levels during pregnancy; stress levels during pregnancy; trimester(s) of exposure.
Outcome variables: Tasks; Performance on cognitive tasks; unadjusted and adjusted effect sizes
Confounding factors: Covariates considered and adjusted for in analysis.
Methodology: Participants’ demographic information; Participant attrition rate; study design; times of measurement, sample size.
Bias and quality assessment
The quality of the studies selected was assessed by GD using a modified version of the Scottish Intercollegiate Guideline Network appraisal tool (SIGN) criteria for cohort studies (https://www.sign.ac.uk/sign-50). SIGN is a widely used tool which appraises studies by examining six domains: study design, withdrawals and drop-outs, potential for selection bias, measurement of outcomes and exposure factors, confounders and report of statistical analyses, and blinding. Using a scoring algorithm adapted from the Quality Assessment Tool for Quantitative Studies (http://www.ephpp.ca/tools.html), each of these domains were rated as being of low, moderate or strong quality, and an overall study quality rating was made based on the quality of each domain. Studies encompassing four or more strong and no weak domains were rated as strong; those with less than four strong and no more than one weak domain were rated as moderate; and those with more than two weak domains were rated as weak.
Using Comprehensive Meta-Analysis Version 3, a series of standard meta-analyses of correlations estimated the pooled associations between prenatal maternal anxiety and/or stress and children’s cognitive function. Due to the variability in characteristics and sample sizes across studies, the more conservative random-approach was used 34. Separate meta-analyses were performed when at least two studies provided an effect for an outcome 35. Results were stratified by type of exposure (i.e., stress and anxiety) and cognitive domain (i.e., general intellectual function, language, attention, learning and memory and working memory). While pregnancy-related anxiety was found to constitute a valid and reliable construct 36,37, the scales used to measure this psychological construct assess the nature of their anxiety or worries rather than the type and severity of symptoms of anxiety 30,36,37,38. As such, studies examining this construct were not included in the main analyses examining the relationship between all types of anxiety and/or stress combined and children’s later cognitive outcomes to maximise consistency. When enough data was available, results were stratified by type of anxiety and stress exposure (i.e., perceived stress, stress response, trait anxiety and state anxiety) and timing of exposure (trimester 1, 2 or 3). Although the original intent was to assess associations for different developmental periods, this was not possible due to the small number of eligible studies which examined stress and/or anxiety at different timepoints during pregnancy. Consequently, analyses were performed combining children from all age groups to increase power and were also performed for the infancy period only.
A measure of effect size (coefficient of correlation r) was calculated for each study. A correlational effect size r of .05 was considered as very small, .10 as small, .30 as medium, .30 as large, and .50 or greater as large 39. The statistical significant level used was α = 0.05 . Both I2 statistic and Tau squared were computed to assess for heterogeneity of effect sizes 34. I2 reflects the percentage of variability in the correlational effect size that is attributed to heterogeneity rather than sampling error. A value of 25% or below can be interpreted as a small amount of heterogeneity, while values of 50% and 75% or above represent moderate and high heterogeneity, respectively 40. Tau squared represents the variance of the true correlational effect size across studies.