Maternal pre-pregnancy obesity and child neurodevelopmental outcomes: A Meta-analysis

Study Selection

We conducted a comprehensive search of Pubmed (1946 to April 2017) and PsycInfo (1597 to April 2017) using terms (see S-Table 1) related to prenatal, obesity, offspring and neurodevelopment. The term ‘neurodevelopmental’ included ASD, ADHD, cognitive and intellectual delay and emotional/behavioral problems. We included observational studies irrespective of publication status, language, sample size, follow-up duration or BMI classification standard.

Three coders (CES, KR, AM) examined each article’s title, abstract, and keywords from the search results. Each coder worked independently to determine whether an article met inclusion criteria. If disagreements between coders could not be resolved, the senior author (BFF) was consulted. If relevant articles were misclassified during the initial coding process, two other techniques (see below for more detail) were used to help uncover the reports again. In total, 1483 abstracts were examined − 57 (4%) were deemed to meet inclusion criteria.

Two other types of searches were conducted on any empirical study report and/or literature review that was determined to be relevant. We examined reference lists of all reports that met inclusion criteria to determine whether they cited any potentially relevant article (backward search). We then conducted a cited reference search to determine whether the reports that met inclusion criteria had been later cited by any potentially relevant article (forward search). For backward and forward searches, titles and abstracts were initially reviewed by the first author (CES), and if deemed potentially relevant, the full-text was obtained. These search strategies yielded an additional 6 reports.

To minimize publication bias, we contacted authors whose articles reported maternal obesity and neurodevelopmental outcomes but did not provide adjusted or unadjusted ORs, and requested raw data so that we could calculate unadjusted ORs. For 12 articles (with publication dates within the past 10 years), we received 4 responses that allowed us to include the report in the meta-analysis and 2 responses that provided information to include in the synthesis.

Data Extraction

Numerous characteristics of each study were retrieved using a standardized data collection form. These characteristics encompassed six broad domains: (a) the research report included basic information about authorship and date of publication; (b) study characteristics included information about the cohort, source of study population, and study design; (c) quality indicators to assess observational studies (see below); (d) sample information included demographic characteristics of the cohorts; (e) outcome measures included information pertaining to the neurodevelopmental outcomes and covariates; and (f) estimate of effect size detailed the information needed to derive an unadjusted OR or adjusted OR and 95% confidence intervals (CIs).

Data extraction was completed by 4 reviewers (CB, KR, AS, AM), with each article being independently coded twice. If there was a discrepancy in coding, the two coders discussed each disagreement until agreement was reached. If the disagreement could not be resolved, the first author was consulted. This method results in high effective reliability.²²

Risk of Any Adverse Neurodevelopmental Outcome

For overweight mothers, 22 cohorts from 23 articles contributed to the analysis. The average cohort contributed more than two effect sizes (M = 2.59, MEDIAN = 1.00, MIN = 1.00, MAX = 15.00), with a total of 57 ORs (range 0.58 – 2.00). Across these studies, mothers who were overweight prior to pregnancy were 17% more likely to have a child with any adverse neurodevelopmental outcome (OR, 1.17; 95% CI, 1.11, 1.24; P < .001, I² = 65.51). Analysis for publication bias suggests that this OR is underestimated by 0.003.

For obese mothers, 25 cohorts from 27 articles contributed to the analysis. The average cohort contributed three effect sizes (M = 3.00, MEDIAN = 2.00, MIN = 1.00, MAX = 15.00). The 75 ORs ranged from 0.26 to 5.20. Results across these studies indicated that mothers who were obese prior to pregnancy were 51% more likely to have a child with any adverse neurodevelopmental outcome (OR, 1.51; 95% CI, 1.35, 1.69; P < 0.001, I² = 79.63). Publication bias analysis suggested that this OR is overestimated by 0.006.

An additional 9 articles were not included in the meta-analyses because of methodological differences in either the quantification of pre-pregnancy BMI, the outcome, or the association: 1 article reported continuous BMI data, 7 articles reported continuous outcome data and 1 article reported hazard ratios. In the study using a continuous BMI data⁶² the investigators found no association between ASD risk at 8 years of age and pre-pregnancy BMI (though they did find ASD risk was associated with pregnancy weight gain).^62,63 With respect to the studies that examined outcome data continuously, 2 studies investigated cognitive development during infancy and 5 others reported on cognitive and intellectual development in children 3 to 7 years. Among infants, one study reported an association between maternal pre-pregnancy obesity and reduced cognitive development (an adjusted mean difference between normal weight and obesity of −2.7 in one cohort and −3.7 in another) and the other reported no significant associations.^64,65 Among children 3 to 7 years of age, a significant negative association between maternal pre-pregnancy BMI and child IQ (beta values ranged from −0.10 to −4.03), academic achievement (a mean difference between normal weight and obesity of −3 points or 0.23 SD units) and emotional/behavioral dysregulation (beta values ranged from 0.36 to 4.28) was reported.^66–70 Finally, Chen et al.⁷¹ investigate maternal pre-pregnancy BMI and offspring’s risk of ADHD using a sibling-comparison design and Stratified Cox proportional hazards models. Findings suggested that sibling comparison attenuated the dose-dependent increased risk of offspring’s ADHD, indicating that the association may be due to unmeasured familial confounding.

Risk of Specific Neurodevelopmental Outcomes

Compared to children born to mothers who were normal weight prior to their pregnancy, children of overweight mothers were 30% more likely to have ADHD (OR, 1.30; 95% CI, 1.10 – 1.54; P = 0.01, I² = 52.97), 10% more likely to have ASD (OR, 1.10; 95% CI, 1.01 – 1.21; P = 0.026, I² = 0.00) and 23% more likely to have a cognitive/intellectual developmental delay (OR, 1.19; 95% CI, 1.09 – 1.29; P = .003, I² = 40.05). Children of overweight mothers did not exhibit a statistically significant increased odds of emotional or behavioral problems (OR, 1.14; 95% CI, 0.93 – 1.39; P = .16, I² = 70.78).

Compared to children born to mothers who were normal weight prior to their pregnancy, children of obese mothers were 62% more likely to have ADHD (OR, 1.62; 95% CI, 1.23 – 2.14; P = 0.006, I² = 70.15; Figure 2), 36% more likely to have ASD (OR, 1.36; 95% CI, 1.08 – 1.70; P = 0.015, I² = 60.52; Figure 3), 58% more likely to display cognitive or intellectual delay (OR, 1.58; 95% CI, 1.39 – 1.79; P < 0.001, I² = 75.77; Figure 4), and 42% more likely to have an emotional or behavioral problems (OR, 1.42; 95% CI, 1.26 – 1.59; P < 0.001, I² = 87.74; Figure 5).

Sensitivity Analysis

Studies ranked as having a higher quality score produced an overall effect size that was lower than those with lower quality scores. Slightly attenuated effects sizes were observed for those studies with a propective study design, a larger sample size and methods to control for maternal/paternal or genetic effects in multivariate regression or in study design. (see Table 3).

Study Limitations and Strengths

An underlying assumption here was that measured outcomes in the various studies were comparable with respect to the quality. Studies were not excluded based on method of assessment (e.g., health professional diagnosis vs parent self-report). Also, some studies did not use standardized scales, relying instend on selected items from different scales. Those that used standardized scales differed in defining levels of problematic functioining, with some studies using cutoff score defined by normative data and others using data-driven cutpoints (e.g., higher than the 90^th percentile). To advance the literature it would be useful for future studies to use standardized scales as well as include individually-administered measures of child cognitive and neurodevelopmental functioning.

A broad limitation of this field is that pre-pregnancy BMI is almost always based on maternal report of weight prior to last menstrual period, ascertained either via medical records or via self-report. There is some evidence that women who are obese underreport their weight status;^72,73 however, data from birth cohort studies find that the discrepancy between maternal self-reported weight and weight measured by trained staff in the clinic is highly correlated.⁷⁴ It is difficult to know what effect this potential misclassification has on the associations observed across the various studies reviewed, but if the trend was toward under-reporting then the associations may underestimate the effect size.

The methods for addressing confounding differed across studies. One critique of this literature is that the associations between pre-pregnancy obesity and adverse outcomes may be due to unmeasured confounding by genetic or shared family environmental factors and that the association may not be causal. A method proposed to mitigate sources of confounding shared within families has been the use of sibling designs where the analysis involve comparing outcomes among siblings that are discordant for the exposure. We identified five studies that used this design.^{13,34,44,59,71} Two were included in the meta-analysis and three in the systematic review. All but one⁴⁴ found that the sibling comparison design nullified the dose-dependent increased risk of child neurodevelopmental outcome in relation to pre-pregnancy weight. Such designs, while informative, are not without potential limitations (e.g., carryover effects)⁷⁵ and do not completely rule out a causal effect of pre-pregnancy obesity. Continued research using a variety of study designs such as these and others (e.g., propensity score methods, Mendelian randomization) would be helpful in further clarifying the magnitude of the overall effect size.

Another caveat to this review was that we did not focus on gestational weight gain in relation to neurodevelopment due to the limited number of studies that examined this variable. Gestational weight gain, whether it is insufficient or excessive, may be important to neurodevelopment. Some studies that did assess gestational weight gain found it was not related to neurodevelopmental outcomes beyond pre-pregnancy weight,^59,66 whereas others found significant positive associations with neurodevelopmental outcomes.^44,52,58,62 Continued research is needed that will help elucidate the degree to which gestational weight gain and pre-pregnancy obesity independently or synergistically relate to an increased risk of adverse neurodevelopmental outcomes in children.

Strengths of this review were the inclusion of methods to evaluate sources of potential bias through missing studies estimation (publication bias) and sensitivity analysis. Our analyses of publication bias suggested that the summary estimates are approximately correct with very little evidence of publication bias influencing the results. Sensitivity analyses evaluating overall quality were also conducted. In general, these sensitivity analyses show that the overall effect of pre-pregnancy obesity on neurodevelopment ranged from an OR of 1.90 in the lower quality studies to 1.33 in the higher quality studies. This suggests that in order to produce more accurate assessments of the effect size of this relationship between pre-pregnancy weight and neurodevelopment, future studies should consider quality indicators evaluated here such as, the use of prospective designs, sufficient sample size for analysis, and inclusion of covariates in multivariate models. Another strength of this study was the use of the robust variance estimation (RVE) method, which allowed for analyses of all collected effect sizes to be used in the same meta-analysis. This was important since many studies reported on multiple outcomes. In addition, we also contacted researchers who reported results with regression coefficients only and obtained additional information to calculate an OR. This resulted in 6 additional studies being included in our synthesis and strengthened our estimation of the overall effect size.

Potential Mechanisms

There are a number of proposed mechanisms to help account for the observed association between pre-pregnancy obesity and neurodevelopment outcomes. A prevailing hypothesis is that the inflammatory milieu that accompanies an obese state before and during pregnancy leads to a cascading series of events that affect brain development and subsequent neurodevelopmental functioning.^5,54,76,77 Preclinical models support this hypothesis and show that gestational obesity induced by a high dietary fat produces changes in inflammatory signaling that have downstream effects on offspring microglial functioning, brain development and subsequent offspring neurobehavioral patterns.^5,76 Other processes may include metabolic hormone-induced changes, such as exposure to increased circulating levels of leptin or insulin, or suppression of serotoninergic functions by disrupting the normal development of the serotonin (5HT) system.⁵ Maternal pre-pregnancy obesity may also be a marker for nutritional differences. For example, maternal high-fat diet during pregnancy has been linked with dopamine functioning and impulsive responding of offspring.^78,79

Although preclinical studies have been useful in explaining some of the potential mechanisms, fewer studies have examined similar mechanisms in human population studies. Notably, there are methodological challenges in human studies that make it difficult to disentang the effects of maternal diet and maternal pre-pregnancy weight on offspring neurodevelopmental outcomes. Some preliminary evidence, however, does suggest a link between aberrant prenatal inflammatory cytokines levels and early childhood intelligence and risk for ASD.^80,81 A critical next step could be to begin to link prenatal exposure to maternal obesity to childhood cognitive capacity via inflammatory signaling.

Implications

This meta-analysis supports the notion established in pre-clinical and observational studies that offspring born of mothers who are obese prior to and during their pregnancy are at increased risk for neurodevelopmental problems, including a higher risk of ADHD-related symptoms. The effect size across these studies (e.g., OR of 1.62 or Cohen’s d = 0.27 for maternal pre-pregnancy obesity and ADHD) is comparable to the effect size seen in studies of lead exposure and ADHD (Cohen’s d = 0.32)⁸² and studies of prenatal smoking and ADHD (Cohen’s d = .20).⁸³ Further confirmation of these associations could have significant clinical implications by 1) allowing for the identification of a group women who may at risk for delivering children with neurodevelopmental problems and 2) increasing maternal awareness of the harmful health effects of obesity not only on her pregnancy but her child’s brain development. Current 2016 estimates of obesity for women in the US are 40.4% which has increased from previous estimates in 2005–2014.⁸⁴ At a population level, arresting the obesity epidemic among women in the childbearing age not only has the potential to improve delivery outcomes, but, if these findings continue to bear out, has the potential to have downstream effects on the prevalence of neurodevelopmental problems in children.