Effects of Nutritional Interventions during Pregnancy on Infant and Child Cognitive Outcomes: A Systematic Review and Meta-Analysis

Background: Epidemiological studies have demonstrated that folate, iodine and iron intake during pregnancy impacts on foetal brain development and cognitive function. However, in human studies, the relationship with other dietary nutrients is less clear. Objective: This systematic review aims to critically appraise the current literature and meta-analyses results from nutritional interventions during pregnancy that aimed to optimise infant and child cognitive outcomes. Design: Ten electronic databases were searched for articles published up to August 2017. The search was limited to articles published in English. Randomised controlled trials (RCTs) testing the impact of any nutritional intervention (dietary counselling, education, nutrient supplementation, fortified foods and/or foods) during pregnancy on cognitive outcomes of children (<10 years old). Two independent reviewers assessed study eligibility and quality using the American Dietetic Association quality criteria checklist for primary research. Standardised mean differences were used for nine cognitive domains to measure effects for meta-analyses. Results: A total of 34 RCTs were included (21 studies included children aged less than 35 months, 10 studies included children aged 36–60 months and 3 studies included children aged 61–119 months). The types of nutritional interventions included nutrient supplements, whole foods, fortified foods and nutrition education. The following nine cognition outcomes: attention, behaviour, crystallised intelligence, fluid intelligence, global cognition, memory, motor skills, visual processing, and problem solving were not significantly impacted by nutritional interventions, although 65% of studies conducted post-hoc data analyses and were likely to be underpowered. Although, long chain polyunsaturated fatty acids (LCPUFA) supplementation was associated with a marginal increase in crystallised intelligence (Effect size (ES): 0.25; 95% confidence interval (95% CI): −0.04, 0.53), the effect was not statistically significant (p = 0.09), with significant study heterogeneity (p = 0.00). Conclusions: LCPUFA supplementation may be associated with an improvement in child crystallised intelligence, however further research is warranted. The remaining eight cognition domains were not significantly impacted by maternal nutritional interventions.


Study Selection
Eligibility of the retrieved publications was assessed by two independent reviewers (RMT and SMF) as recommended by PRISMA [29]. Publication title and abstracts were evaluated against the inclusion and exclusion criteria (Table 1) Publications were excluded by a hierarchical approach based on the study design, population, intervention and outcomes. Publications meeting the initial eligibility screening were retrieved in full and were then subjected to a second round of screening by the same two reviewers to determine the final inclusion or exclusion. Assessment discrepancies between reviewers were resolved by discussion or assessment by a third party independent reviewer (AJH). The reference lists of the included publications were searched separately to identify any relevant articles that were not detected by the electronic search strategy. The abstract and then the full texts were retrieved for consideration. After the retrieval of the full text a final decision was made about the eligibility of the publication by the first reviewer (RMT).  Table 1 outlines the inclusion and exclusion criteria for this systematic review. Studies that delivered nutritional interventions to the mother during pregnancy (exclusively) or during pregnancy plus the post-natal period (<10 years) were included. This review included studies that measured cognition in children up to nine years of age because the effect of nutritional interventions may not be apparent until later life [18]. In addition, cognition assessment tests in older children (>7 years) are a more reliable predictor of adult cognitive function [31]. Nutritional interventions that commenced after pregnancy were excluded. This review included studies that measured cognition (as a primary or secondary outcome) in infants and young children using one or more cognition assessment tests or subtests. Cognition domains were based on the Cattell−Horn−Carroll (CHC) Theory of Intelligence [32]. The domains fluid and crystallised intelligence, specified in the CHC model [32], were combined to form the global cognition domain for cognition assessment tests that reported an overall global composite score. Verbal composite scores were grouped in the crystallised intelligence domain and non-verbal composite scores were grouped in the fluid intelligence domain. Short-term memory and long-term retrieval domains were combined to form an overall memory domain. The remaining six cognition domains of the CHC model [32] (quantitative reasoning, reading and writing ability, visual processing, auditory processing, processing speed and correct/decision speed) were included. Attention, motor skills and behaviour were also analysed because these domains are commonly analysed in cognition assessment tests. For example, the Intergrowth-21st Neurodevelopment Assessment reports child cognitive scores for the domains attention, motor skills and behaviour [33]. In addition, Bayley Scales of Infant development (third edition) includes a behaviour rating scale, which measures behavioural factors including orientation/engagement and emotional regulation [34]. This review included the outcome visual processing that focuses on both the response and comprehension of visual stimuli. Visual acuity which primarily focuses on vision clarity, was not reported as an outcome in this review. Similar methods have been described previously by Eilander et al. [35].

Quality Assessment
Articles considered eligible were assessed for methodological quality in their reporting using the American Dietetic Association (ADA) Quality Criteria Checklist for Primary Research [36] which is applicable to evaluating the validity of nutrition studies. The quality of the articles was assessed by two independent reviewers (RMT and SMF) and assessment discrepancies between reviewers, were managed through discussion or resolved by the third independent reviewer (AJH). The Quality Criteria Checklist [36] rates the study design and execution, as well as the risk of bias using ten validity questions, including four priority questions which must be satisfactory to gain a positive rating. Based on the responses to these questions, the checklist assigns a quality rating: positive (answered "yes" to six or more validity questions, including all four priority questions), negative (answered "no" to six or more validity questions) or neutral (answered "no" to one or more of the four priority criteria questions) [36]. No studies were given a negative rating in this review, therefore no study exclusions were made.

Data Extraction
The first reviewer (RMT) extracted relevant data from the included studies using an Excel spreadsheet. The Organisation for Economic Co-operation and Development (OECD) criteria [37] were used to classify country income into four categories: low, lower-middle, higher-middle and high. Nutritional interventions were divided into six categories; single nutrient supplement, multiple micronutrient supplement, fortified foods, foods, supplements with foods, supplements with foods and nutrition education or no nutrition intervention (control group). For the purpose of this review a multiple micronutrient supplement was defined as two or more nutrients which is consistent with other authors [38][39][40]. The cognition outcomes data included the mean, standard deviation, confidence intervals, minimum and maximum range and odds ratio. Corresponding authors were emailed if additional data or clarification was required.

Data Synthesis
The nutritional interventions provided across the included studies were grouped together according to the nutrient supplied and the type of intervention (i.e., supplement, dietary changes). The cognition outcomes reported were organised into related cognition domains and the type of scoring (i.e., subtest or composite score). Thirteen cognition domain categories were derived: (i) fluid intelligence; (ii) crystallised intelligence; (iii) quantitative knowledge; (iv) memory; (v) visual processing; (vi) auditory processing; (vii) processing speed; (viii) correct decision speed; (ix) reading and writing; (x) global cognition; (xi) attention; (xii) motor skills; (xiii) behaviour.

Statistical Analysis
The aim of meta-analysis was to present a single effect size for the 13 specified cognition scores. The standardised mean difference (SMD) was used to measure the effect considering that all outcomes were measured using different cognition assessment tests and reporting units. The SMD, also known as Cohen's d, is defined as the difference between the mean of the intervention and control group divided by the standard deviation of the data. Methods to convert other units (e.g., Odds ratio) to SMD are described in Sanchez-Meca et al. [41]. Studies that presented multiple cognition outcomes within the same cognition domain, multiple time-points, or multiple interventions for multiple micronutrient supplementation with only a single control group, the respective method described by Borenstein et al. [42] was used to calculate a summary SMD and standard error which accounts for non-independence within each study. As the correlations between outcomes were not reported by the studies, the correlation was assumed to be 0.5, and sensitivity analyses were performed using a higher (0.8) or lower (0.2) correlation to assess the changes in the standard error.
Supplementary Table S2 shows an example of how the composite effect size was created. A random effects meta-analytic model (using the method of DerSimonian and Laird [43]) was conducted for each cognitive outcome, where the observed effect sizes are regarded to vary due to sampling variance. A fixed effects model using inverse variance weights was also conducted for secondary analysis. The heterogeneity and total variation of studies were analysed using the I 2 statistic [44]. Funnel plots were used to assess publication bias. Subgroup analysis was performed for each outcome by the nutrient of interest. A sensitivity analysis with a random effects model was conducted in which the studies were stratified by country income. The statistical analysis was conducted using the metan command in the statistical software package STATA 13 (College Station, TX, USA: StataCorp LP 2013).

Search Results
A total of 1195 publications (excluding duplicates, n = 439) were assessed against the inclusion and exclusion criteria ( Figure 1). Seven included publications were identified from hand searching the reference lists of included publications detected from the initial search strategy. These articles all used the keyword mental development, which was not included separately. This systematic review resulted in 34 included publications after inclusion and exclusion criteria were applied to full texts. the reference lists of included publications detected from the initial search strategy. These articles all used the keyword mental development, which was not included separately. This systematic review resulted in 34 included publications after inclusion and exclusion criteria were applied to full texts.

Participants
The study populations were mostly healthy pregnant women, except one publication [61] which included pregnant women with human immunodeficiency virus type 1 (HIV-1), although pregnant women were excluded if they were diagnosed with acquired immunodeficiency syndrome (AIDS) according to the World Health Organisation (WHO) definition [77].

Intervention
A summary of the nutritional interventions of the studies is provided in Supplementary Table S4. In summary, 76% of studies provided a nutrient supplement exclusively for the study intervention, no studies provided nutrition counselling or education as a stand-alone intervention. Table 2 summarises the cognition assessment tools used in each study. Further details about the cognition assessment tests used is provided in Supplementary Table S5. The age of child cognitive testing ranged from less than two weeks [76] to nine years old [71], however 62% of studies (n = 13 trials) reported cognition outcomes in children less than 35 months, and 46% of these studies (n = 6 trials) provided LCPUFA interventions in high-income countries. Meta-analysis was completed for nine cognition outcomes measured in the 34 included studies. The analyses for each cognition domain is described below, however in summary all cognition domains were not significantly affected by nutritional interventions.  There were no significant differences between the intervention and control groups for child attention, working memory and inhibitory control development test scores (p > 0.05) Positive effects were demonstrated from the fish oil intervention group on the problem-solving task (p = 0.017), total intentional score (p = 0.011) and the number of intentional solution on both cloth (p = 0.008) and cover steps (p = 0.004) at 9 months  UNIT [110], Go/no-go [111], the Stroop test [112], backward digit span [113], MABC [114], finger-tapping test [115] The difference across all outcomes (UNIT, executive function and motor functioning) was significant in children whose mother received iron and folic acid supplementation (p = 0.001) compared to the control group but not for the other intervention groups (p > 0.05)

Cognitive Outcomes
Christian et al. [    There were no significant differences between the intervention and control groups for child cognitive, social or behavioural development test scores

Outcome 8: Attention
Four RCTs [47,66,74,76] measured the effect of a nutritional intervention on attention. There was no significant effect of multiple or LCPUFA supplementation (n = 2 trials) [47,74] on attention. Vuori et al. [76] reported that multiple micronutrient supplementation during pregnancy did positively affect attention in infants at 15 days postnatally.

Outcome 9: Problem Solving
Two RCTs [59,64] measured the effect of a nutritional intervention on the problem solving composite score. There was no significant effect of multiple micronutrient intervention (n = 1 trial) [64] or LCPUFA intervention (n = 1 trial) [59] on problem composite scores.

Discussion
This systematic review and meta-analysis indicates that nutritional interventions during pregnancy do not significantly impact on child cognitive outcomes. There is some evidence to suggest that LCPUFA supplementation may marginally improve child crystallised intelligence, however the effect was not statistically significant and study heterogeneity among the trials was significant. Although a wide scope of literature was assessed and a rigorous selection methodology was applied, the data included in this systematic review were affected by significant methodological limitations. From the included studies, 79% were exclusively nutrient supplementation trials, the effect of food-based interventions or nutrition counselling alone during pregnancy on child cognition was not well explored. Many studies were small, and attrition rates were high due to the length of follow-up required. Child cognition was a secondary outcome of interest in many studies, even with meta-analysis our findings are likely to be underpowered. This systematic was unable to perform a meta-analysis in specific age ranges due to the lack of data available for children older than 35 months, especially in low and middle income countries. Therefore, it is important that future trials include a longer child follow-up, to explore any possible delayed effects of maternal nutritional interventions on child cognition.

Quality of Included Studies
Many included studies were considered to be of high quality. While, the Quality Checklist Criteria [36] is designed to assess the methodological quality and reporting within an article, it does not assess the appropriateness of the methodology in addressing the research question. This is a concern considering that many of included RCTs are likely to be at risk of type 2 error, due to the use of secondary data analysis and not powered to detect differences in cognitive outcome as a primary outcome variable. In addition, many post-hoc studies excluded participants after randomisation and in some cases multiple variable adjustments were required to address the significant characteristic differences between the groups, suggesting the assumptions of randomisation had been breached. Most post-hoc studies did not perform intention-to-treat analysis to evaluate the randomisation of the study groups. This approach is likely to introduce bias and systematic error that will reduce the power of the study. Therefore, it is important that future evidence is derived from adequately powered primary studies to reduce the risk of a type 2 error and systematic bias in post-hoc studies and determine whether there are any effects of nutritional interventions during pregnancy on child cognition.

Measuring Child Cognition
There are many methodological challenges when measuring child cognition using psychological assessment tests. The cognitive performance of infants and children will be affected by their mood, motivation, amount of sleep and personal effort [138]. Environmental factors, including who conducts the cognition tests, who else is present during testing and where the tests are held can also influence a child's performance ability [138]. In addition, most cognition tests were devised in developed countries, using stimuli relevant to their culture, therefore this stimuli is likely to be unfamiliar or unacceptable to children from developing countries.
Many studies in this review used global development assessments (e.g., Bayley's Scales of Infant Development) to determine the children's overall cognitive ability. However, broad measures of cognitive function may only detect highly significant nutritional effects, especially if the study is underpowered [138]. Specific cognitive skills (e.g., attention, memory) that contribute to overall function are regulated by distinct neural mechanisms, that may be differentially affected by nutritional manipulations [139]. Therefore, Cognitive Assessment Tests that analyse specific skills, may be more sensitive to detect subtle effects of nutritional interventions on cognition. For this reason, future trials may be improved by the use of assessment tests that measure specific cognitive skills, as well as global measures of cognition. The choice of specific cognition test should be based on the hypothesised neural mechanisms that are expected to be altered by nutrients.
To determine cognitive function later in childhood and adolescence, future studies should consider linking and analysing school performance data. Pilot data should be used to determine the appropriateness of the sample size and ensure that the study is adequately powered. This approach also has the advantage of the outcome assessors being blinded to the randomisation.

Nutrition Intervention Design
Due to methodological and ethical challenges, most trials analysed the effects of nutrient supplementation during pregnancy on child cognition rather than using food-based interventions. This may reflect the perception that obtaining adherence to food-based interventions is a challenge and therefore nutrient supplementation is a more feasible strategy/study design. However, performing a meta-analysis for the supplementation trials proved to be difficult because many studies used a supplemented control group rather than an unsupplemented placebo group. Consequently, only the effect of the additional nutrient supplemented in the intervention group compared to the control group could be accessed. In Australia, evidence-based guidelines recommend the use of folic acid and iodine supplementation during pregnancy, therefore it is unethical to use an unsupplemented control group within this population. Li et al. [60] and Caulfield et al. [57] attempted to address this dilemma by providing the same dosage of folic acid across all study groups while women in the intervention groups received a multiple micronutrient supplement (including folic acid) or folic acid with iron. In addition, 88% of the studies conducted in high-income countries provided a single nutrient supplement to women during pregnancy. Since the effect of a single nutrient supplement on child cognition is likely to be modest, especially in a population that is well-nourished, this is likely to explain why a significant summary effect was not found. While, 88% of middle and low-income country studies, provided multiple micronutrient supplementation in a population that is likely to be undernourished, but no significant effect on child cognition was found. However, since child cognition assessment test were validated in a western population, this may have resulted in summary effect that was underestimated.
By convention, many randomised control trials analyse the effect of nutrition on cognition by distilling complex dietary patterns into a single agent using a nutrient supplement. A single nutrient supplement is However, the bioavailability and efficacy of nutrient supplements can differ compared to nutrients contained in food [140]. Nutrient bioavailability is affected by the specific nutrient, absorptive capacity of intestinal mucosa, physiological states (e.g., pregnancy) and nutrient-nutrient interactions [140]. For example, the bioavailability of inorganic iron is enhanced by an interaction with vitamin C and increasing the bioavailability of iron can decrease magnesium and calcium absorption [140]. Huang et al. [141] reported that the bioavailability of natural vitamin was approximately 50 percent greater than synthetic vitamin E (α-tocopherol) in adults. In addition, synthetic vitamin E supplementation was also associated with a significant reduction in the bioavailability of other forms of vitamin E [141]. This trial highlights the limitations of analysing the effects of nutrition via a single agent without considering the biochemical pathways in which nutrients interact that is likely to be important for cognitive development [142][143][144]. Given the limited evidence available, there is a clear need for trials that are designed from a sound biologically plausible basis.

Public Health Implications
Based on the meta-analyses presented, there is a lack of robust evidence on which to base recommendations for maternal diet or for the use of nutrient supplements as a strategy to enhance cognition in children. Given that the synergistic effect of micronutrients in metabolic pathways is important for enhancing cognitive development of children [145], it is important that future trials focus on food-based interventions that aim to improve the overall diet quality of pregnant women. In developing countries, pregnant women are more likely to experience protein-energy malnutrition, as well as multiple micronutrient deficiencies in comparison to high-income countries [146]. Therefore, further evidence is required to determine whether protein and energy as well as micronutrients interventions are effective for improving child cognition in the long-term in developing countries.

Conclusions
In conclusion, this systematic review suggests that the infant and child cognitive outcomes were not significantly improved by nutritional interventions (dietary counselling, education, nutrition supplementation, fortified foods and/or whole foods). However, the possible association between LCPUFA supplementation and child crystallised intelligence warrants further investigation in high quality RCTs powered to detect difference in cognitive variables as the primary outcome. Further evidence from primary research is required to test whether maternal nutritional interventions during pregnancy, in particular food-based interventions, are beneficial for improving child cognition. This review emphasises the need for well-designed pilot studies to inform adequately powered RCTs and data linkage studies. This evidence is important because gains in child cognition are likely to have extended reach to other areas of public health and carry economic benefit.
Supplementary Materials: The following are available online at www.mdpi.com/2072-6643/9/11/1265/s1, Table S1: Search strategy used for systematic review; Table S2: Computing a combined effect across outcomes; Table S3: Summary of quality assessment; Table S4: Summary of the nutritional interventions for included studies; Table S5