Maternal thyroid hormone insufficiency during pregnancy and risk of neurodevelopmental disorders in offspring: A systematic review and meta‐analysis

Summary Background In the last 2 decades, several studies have examined the association between maternal thyroid hormone insufficiency during pregnancy and neurodevelopmental disorders in children and shown conflicting results. Aim This systematic review aimed to assess the evidence for an association between maternal thyroid hormone insufficiency during pregnancy and neurodevelopmental disorders in children. We also sought to assess whether levothyroxine treatment for maternal thyroid hormone insufficiency improves child neurodevelopment outcomes. Methods We performed systematic literature searches in MEDLINE, EMBASE, PSYCinfo, CINAHL, AMED, BNI, Cochrane, Scopus, Web of Science, GreyLit, Grey Source and Open Grey (latest search: March 2017). We also conducted targeted web searching and performed forwards and backwards citation chasing. Meta‐analyses of eligible studies were carried out using the random‐effects model. Results We identified 39 eligible articles (37 observational studies and 2 randomized controlled trials [RCT]). Meta‐analysis showed that maternal subclinical hypothyroidism and hypothyroxinaemia are associated with indicators of intellectual disability in offspring (odds ratio [OR] 2.14, 95% confidence interval [CI] 1.20 to 3.83, P = .01, and OR 1.63, 95% CI 1.03 to 2.56, P = .04, respectively). Maternal subclinical hypothyroidism and hypothyroxinaemia were not associated with attention deficit hyperactivity disorder, and their effect on the risk of autism in offspring was unclear. Meta‐analysis of RCTs showed no evidence that levothyroxine treatment for maternal hypothyroxinaemia or subclinical hypothyroidism reduces the incidence of low intelligence quotient in offspring. Limitations Although studies were generally of good quality, there was evidence of heterogeneity between the included observational studies (I 2 72%‐79%). Conclusion Maternal hypothyroxinaemia and subclinical hypothyroidism may be associated with intellectual disability in offspring. Currently, there is no evidence that levothyroxine treatment, when initiated 8‐ to 20‐week gestation (mostly between 12 and 17 weeks), for mild maternal thyroid hormone insufficiency during pregnancy reduces intellectual disability in offspring.


| INTRODUCTION
Thyroid hormone is essential for optimum neurological development of the foetus. However, the foetal thyroid gland is not functional until the 12-14th week of gestation, 1 and during that period, the foetus is solely dependent on thyroxine from the mother. Therefore, it is plausible that maternal thyroid hormone insufficiency, particularly in early pregnancy, could impair foetal neurodevelopment. Indeed, in recent years, several studies have shown that even mild maternal thyroid hormone insufficiency (including subclinical hypothyroidism and isolated hypothyroxinaemia) during pregnancy is associated with various types of neurodevelopmental disorders in children, including reduced intelligence quotient (IQ) scores, 2 autism 3 and attention deficit hyperactivity disorder (ADHD). 4 However, other studies have shown conflicting results, 5,6 and there remains an uncertainty whether levothyroxine treatment in mild maternal thyroid hormone insufficiency improves neurodevelopmental outcomes in children. 7 These observations have led to an ongoing debate about whether all pregnant women should be screened and treated for thyroid dysfunction.
We therefore conducted a systematic review and meta-analysis to assess the evidence for an association between maternal thyroid hormone insufficiency during pregnancy and neurodevelopmental disorders in childhood, with a focus on intellectual disability, autism spectrum disorders and ADHD, which show substantial comorbidity. 8 We also assessed the evidence from randomized controlled trials (RCTs) of the effects of levothyroxine treatment in pregnant mothers with thyroid hormone insufficiency on child neurodevelopmental outcomes.

| METHODS
A systematic review was conducted according to the best practice guidelines recommended by the Centre for Research Dissemination 9 and is reported in accordance with the PRISMA reporting guidelines. 10 A predefined protocol was developed and is registered with PROSPERO (2016: CRD42016032790).

| Literature search
A comprehensive search syntax using MeSH and free text terms was developed for MEDLINE (Appendix S1) and adapted as appropri-

| Eligibility criteria
Observational studies were included if they investigated the association of maternal thyroid hormone insufficiency (including overt hypothyroidism, subclinical hypothyroidism and maternal hypothyroxinaemia) in mothers during pregnancy with neurodevelopmental outcomes in their children. Overt hypothyroidism was defined as high serum thyroid-stimulating hormone (TSH) levels and low-serum-free thyroxine (fT4) or total thyroxine (tT4) , subclinical hypothyroidism as high TSH levels with normal fT4 or tT4 levels and hypothyroxinaemia as normal TSH with low fT4 or tT4 levels. Studies measuring TSH, fT4 and tT4 as continuous variables were also included. Studies were classified as indicating intellectual disability where IQ, language delay or global developmental delay was measured, and were classified as ADHD or autism studies where a diagnosis or a validated scale for measuring these conditions were reported as an outcome measure in the offspring. We excluded studies that focussed on psychomotor development rather than neuropsychological development. Studies were included if the children studied were between 0 and 18 years old. RCTs were included if they studied pregnant women with thyroid hormone insufficiency, and randomly assigned them to treatment with levothyroxine or to a control group and reported children's neurodevelopmental outcomes.
Only studies published after 1994 were included in the analysis, coinciding with the publication of DSM-IV which introduced changes in terminology used to describe neurodevelopmental outcomes.

K E Y W O R D S
autism, hypothyroxinaemia, intelligent quotient, pregnancy, subclinical hypothyroidism, thyroid

| Study selection
The search results were uploaded to reference management software (Endnote X7). Titles and abstracts were screened for relevance independently by 2 reviewers (WT and JTC or GB), with any disagreements being resolved by discussion and involvement of a third reviewer (JTC or GB) where necessary. The full text of potentially relevant papers was retrieved and screened in the same way using the prespecified inclusion and exclusion criteria. All duplicate papers were double-checked and excluded. "Sibling" papers derived from the same parent study were identified and linked.

| Data collection
For each study, data on study and participant characteristics, intervention/exposure, thyroid status and relevant neurodevelopmental outcomes were extracted by 1 reviewer (WT) into a bespoke data extraction form, which was piloted on a random sample of studies and refined. All data extraction was checked by a second reviewer (JTC or GB) with discrepancies resolved by discussion and involvement of a third reviewer (JTC or GB) where necessary. Authors were contacted to provide clarification or additional data if necessary.

| Quality assessment
The quality of the design and reporting of included studies was assessed using the Downs and Black checklist 13 by 1 reviewer (WT) and checked by a second (JTC or GB), a reliable and valid quality index for the appraisal of both RCTs and nonrandomized studies. Discrepancies were resolved through discussion and involvement of a third reviewer (JTC or GB) where necessary.

| Data analysis
Studies were tabulated and described narratively in the first instance.
Studies were only included in the meta-analysis with the following criteria: (i) they used a fixed predictor, (ii) that the fixed predictor used was at a threshold of clinical relevance, (iii) that they provided numerical data and (iv) used an outcome assessment that was well known and validated. Where meta-analysis was possible, the following prespecified principles were applied to the data prior to analysis: (i) for studies using multiple tests for the same outcome, the most commonly used test was reported, (ii) where available, the total test score was used rather than subscales, (iii) where a trait was measured over multiple time periods, the latest point of measuring would be chosen, as neurodevelopmental traits at an older age tend to be more stable, (iv) where multiple cut-offs of TSH and fT4 were measured, the most extreme cut-off was used and when an outcome was measured both against a continuous thyroid hormone measure and against a cut-off value (eg cut-off of the 10th percentile fT4 for hypothyroxinaemia) the cut-off value was used, (v) where thyroid hormone samples were collected at multiple times in pregnancy, the earliest time was chosen as hypothyroidism in early pregnancy is believed to be more problematic than in later pregnancy, (vi) where both fT4 and tT4 were measured, the value for fT4 was used and (vii) adjusted measures were used when provided.
Where possible, continuous results including those derived from regression coefficients were converted to odds ratios using the method described by Chinn and colleagues. 14,15 Continuous results were also converted to odds ratios using the method described by Suissa 16

and
Whitehead 17 and explored the impact of the conversion method used in sensitivity analyses. Meta-analyses were also performed using studies only reporting odds ratios. We also carried out sensitivity analysis (post hoc) by carrying out meta-analyses where the results were split by the point in gestation when maternal thyroid hormones were measured, to test the potential adverse effects of thyroid dysfunction in early pregnancy vs late pregnancy.
Some studies gave results that were split by a covariate, which were not comparable with other studies. Data from the study reported by Päkkilä et al 18 for ADHD, for example, were split by gender, hence the authors were contacted for results for both genders combined.
Meta-analysis was conducted using STATA using the randomeffects model.

| Search results
In total, 39 original articles were found to be eligible for this systematic review, with 37 observational studies and 2 RCTs 2,3,5-7,18-51 (Table   S1). Overall 909 176 people participated in these studies (857 014 of them being from one study 37 ). A flow diagram summarizing search results is shown in Figure 1.

| Risk of bias and heterogeneity
Observational studies have an inherent risk of selection bias and confounding. Quality assessment using the Downs and Black checklist 13 (Table S2) showed that some of these studies also had noticeable risk of bias relating to attrition bias (Table S3). One study had a high risk of bias, as it had inadequate information on participant demographics. 23 One of the RCTs 7 was not placebo-controlled and suffered from a high attrition rate (Table S4).
There was variation between the included studies in terms of timing of the assessment of thyroid function during pregnancy, definitions of thyroid dysfunction and child age at the time of assessment of neurodevelopment outcome measures (Table S1).

| Observational studies excluded from meta-analysis
In total, 26 studies (24 observational studies and both RCTs) were included in the meta-analysis. Five studies presenting regression analyses were excluded 26,29,30,35,45 because they did not provide regression information in a form that would allow straightforward conversion into odds ratios: for example, 2 studies 29,30 provided linear regression results with continuous not binary predictors. A further 8 articles were excluded from the meta-analysis for the following reasons: sibling papers with no additional data (n = 2), 21,51 lacking numerical data (n = 1), 44 using a novel outcome measure that was not validated or suitable for meta-analysis (n = 1), 34 reporting odds ratios (using logistic regression) but using continuous predictors (n = 1) 39 and reporting psychomotor outcomes only (n = 3). 19,24,42 Details on why studies were excluded from the meta-analysis are shown in Table S5.

| Overt hypothyroidism
There were inadequate data to perform a meta-analysis on the association between overt hypothyroidism and indicators of intellectual disability. Two studies claimed to measure overt hypothyroidism, 2,48 but in F I G U R E 1 The PRISMA diagram showing search strategy and exclusion criteria at each step practice, the exposed groups in these studies consisted of a mixture of overt and subclinical hypothyroidism cases as the hypothyroidism diagnosis was based entirely on TSH levels rather than TSH and fT 4 combined.

| Subclinical hypothyroidism
Eleven studies were included in the meta-analysis ( Figure 2). There  Figure S1). When only studies reporting odds ratios directly (6 studies) were included in the meta-analysis, maternal subclinical hypothyroidism was not significantly associated with indicators of intellectual disability, though the magnitude of the point estimate was similar (OR 2.37, 95% CI 0.96 to 5.85) (Table S6). When only studies (n = 4) that measured TSH before 12 weeks were included in the meta-analysis, maternal subclinical hypothyroidism was not significantly associated with indicators of intellectual disability, and the magnitude of the point estimate was reduced (OR 1.11, 95% CI 0.66 to 1.88, P = .7) ( Figure S5). The 5 studies are not included in the subclinical hypothyroidism and risk of intellectual impairment meta-analysis 26,27,29,35,39 found no association between maternal subclinical hypothyroidism and impaired neuropsychological development in children.

| Hypothyroxinaemia
Eleven studies were included in the meta-analysis (Figure 3).
Compared to those children born to euthyroid mothers, children born to mothers with hypothyroxinaemia were significantly more likely to show signs of intellectual impairment (OR 1.63, 95% CI 1.03 to 2.56, P = .04). The 3 studies excluded from the hypothyroxinaemia and risk of intellectual impairment meta-analysis 26,29,35 found no association between maternal hypothyroxinaemia and impaired neurodevelopment in children. Our sensitivity analyses showed that the method used to convert continuous measures to odds ratios had little impact on interpretation (Table S6). When only studies reporting odds ratios (4 studies) were included in the meta-analysis, maternal hypothyroxinaemia was not significantly associated with indicators of intellectual disability, though the magnitude of the point estimate was similar (OR 2.11, 95% CI 0.92 to 4.83) (Table S6). When only studies (n = 4) that measured fT 4 before 12 weeks were included in the meta-analysis, maternal hypothyroxinaemia was not significantly associated with indicators of intellectual disability, though the magnitude of the point estimate was similar (OR 1.22, 95% CI 0.55 to 2.74, P = .62) ( Figure S6).

| Association between maternal thyroid hormone insufficiency and autism
Three studies investigated the link between maternal overt hypothyroidism and autism, 3 studies investigated the link between maternal subclinical hypothyroidism and autism, and 2 studies investigated the link between maternal hypothyroxinaemia and autism. Of the 3 studies that investigated the link between maternal overt hypothyroidism and autism, 2 studies 37,50 are based on overt hypothyroidism status on hospital records, whilst one study 40 measured the mother's thyroid hormones directly. As there was no information on thyroid hormone levels on the mothers with hypothyroidism based on hospital records, 37,50 we analysed the results from these studies separately from the study with the direct thyroid hormone measurement in participants. 40 There was no statistically significant effect of maternal overt hypothyroidism based on hospital records on autism in children (OR 2.12, 95% CI 0.75 to 6.00, P = .12) ( Figure S2). The one study that  one used a continuous predictor 40 thus was excluded, thus no metaanalysis could take place. The study that used a continuous predictor 40 found no association between maternal hypothyroxinaemia and autism, whilst the one that used a binary predictor 3 did. Therefore, the effect of maternal thyroid hormone insufficiency on the risk of autism in the offspring is unclear.

| Association between maternal hypothyroidism and ADHD
One study investigated the link between maternal overt hypothyroidism and ADHD, 5 studies investigated the link between maternal subclinical hypothyroidism and ADHD (2 were included in the metaanalysis), and 5 studies investigated the link between maternal hypothyroxinaemia and ADHD (2 were included in the meta-analysis). The one study that investigated the link between maternal overt hypothyroidism and ADHD in children found no significant association (hazard ratio 1.10, 95% CI 0.98 to 1.25). 37 Our meta-analysis found no association between maternal subclinical hypothyroidism and ADHD in children (OR 1.58, 95% CI 0.5 to 5.0, P = .44) ( Figure S3)  Ghassabian (2014) Pop (2003) Craig (2012) Suarez−Rodriguez (2012) Williams (2012) Kasatkina (2006) Henrichs (2010) Pakkila (2015) Li (2010) Grau (2015) Noten ( CI 0.17 to 10.47, P = .78) ( Figure S4). Of the 3 studies not included in meta-analysis, 2 studies 29,45 found no association between either subclinical hypothyroidism or hypothyroxinaemia and ADHD, whilst one study 30 found an association between high TSH and externalizing symptoms.

| Effect of levothyroxine treatment in maternal thyroid hormone insufficiency on neurodevelopment outcomes in children
We

| DISCUSSION
This systematic review and meta-analysis show an association between maternal subclinical hypothyroidism and hypothyroxinaemia during pregnancy and indicators of intellectual disability. There was no association between maternal thyroid hormone insufficiency and the risk of ADHD in offspring, and the effect of maternal thyroid hormone insufficiency on the risk of autism in the offspring was unclear.
We also found no evidence that levothyroxine treatment of mothers with subclinical hypothyroidism or hypothyroxinaemia during pregnancy reduces the incidence of low IQ in children, although the number of studies with data able to contribute to this was small (n = 2).
During the course of our study, 2 systematic reviews with metaanalyses on the association between maternal thyroid hormone insufficiency and neurodevelopmental disorders in offspring were published. 52,53 Compared to these previous systematic reviews, our study had a wider scope as we analysed multiple neurodevelopmental disorders (including autism and ADHD) in addition to indicators of intellectual disability, and performed meta-analysis on RCTs of levothyroxine treatment for pregnant women with mild thyroid hormone insufficiency to improve neurological outcomes in offspring. We This systematic review found no association between maternal thyroid hormone insufficiency and offspring risk of ADHD while its impact on the offspring risk of autism was unclear. As some researchers have suggested autism appears to be linked to a lack of neuronal migration, 66 and some studies suggest ADHD is linked to a lack of functional connectivity 67 (hinting at a lack of white matter formation), it seems possible that hypothyroxinaemia in early pregnancy could lead to autism whilst hypothyroxinaemia in later pregnancy could lead to ADHD. As most of the studies measured fT4 between the first and second trimesters, this could explain the lack of an association found between maternal thyroid hormone insufficiency and ADHD. Future studies should explore the effects of maternal thyroid hormone insufficiency at different stages of pregnancy.
We were unable to demonstrate evidence of a beneficial effect of levothyroxine treatment in pregnant women with mild thyroid hormone insufficiency on neurodevelopmental outcomes in offspring.  5,20 whilst others explicitly excluded participants with known thyroid disease. 28,32 Finally, several studies report thyroid function as continuous outcomes rather than dichotomous. Where possible, we converted results from such studies into odds ratios using 2 different methods, [14][15][16][17] and there was no difference in interpretation between the 2 results generated. However, we excluded 5 studies from meta-analysis as it was not possible to convert their results to odds ratios.
In conclusion, our systematic review and meta-analysis showed an association between mild maternal thyroid hormone insufficiency (subclinical hypothyroidism and hypothyroxinaemia) in pregnancy and impaired neuropsychological development in offspring. We found no evidence that levothyroxine treatment for maternal mild thyroid hormone insufficiency, including maternal hypothyroxinaemia and subclinical hypothyroidism, improves neurodevelopmental outcomes in children.