Association of maternal iodine status with child IQ: a meta-analysis of individual-participant data

Précis: This meta-analysis of individual-participant data showed that a lower urinary iodine-to-creatinine ratio during pregnancy was associated with lower verbal IQ up to the 14th week of gestation. Abstract Context: While the consequences of severe iodine deficiency are beyond doubt, the effects of mild-to-moderate iodine deficiency in pregnancy on child neurodevelopment are less well established. Objective: To study the association between maternal iodine status during pregnancy and child IQ and to identify vulnerable time-windows of exposure to suboptimal iodine availability. Design: Meta-analysis of individual-participant data from three prospective population-based birth cohorts: Generation R (The Netherlands), INMA (Spain), and ALSPAC (United Kingdom); pregnant women were enrolled between 2002-2006, 2003-2008, and 1990-1992, respectively. Setting: General community. Participants: 6180 mother-child pairs with measures of urinary iodine and creatinine concentrations in pregnancy and child IQ. Exclusion criteria were multiple pregnancy, fertility treatment, medication affecting the thyroid, and pre-existing thyroid disease. Intervention(s): None. Main Outcome Measure: Child non-verbal and verbal IQ assessed at 1.5-8 years of age. Results: There was a positive curvilinear association of the urinary iodine-to-creatinine ratio (UI/Creat) with mean verbal IQ only. UI/Creat < 150 µg/g was not associated with lower non-verbal IQ [-0.6 points, 95% CI -1.7 to 0.4, P =0.246] or lower verbal IQ [-0.6, 95% CI -1.3 to 0.1, P =0.082]. Stratified analyses showed that the association of UI/Creat with verbal IQ was only present up to 14 weeks of gestation. Conclusions: Fetal brain development is vulnerable to mild-to-moderate iodine deficiency, particularly in the first trimester. Our results show that any potential randomized, controlled trial investigating the effect of iodine supplementation in mild-to-moderate iodine deficient women on child neurodevelopment, should start with supplementation not later than the first trimester.


Introduction
Iodine is an essential trace element required for the production of thyroid hormones; optimal thyroid hormone availability is important for normal fetal brain development (1,2). During pregnancy there is a higher demand for maternal iodine intake (3,4). This is due to: i) the increased maternal thyroid hormone synthesis required to ensure adequate thyroid hormone availability to the fetus, ii) greater urinary iodine loss due to an increased glomerular filtration rate, and iii) placental transfer of iodine to the fetus to facilitate fetal thyroid hormone production. While severe iodine deficiency is no longer common in Europe, mild-to-moderate iodine deficiency is still common, especially in pregnant women (5). Severe iodine deficiency in pregnancy results in a higher risk of goiter, hypothyroidism, and mental retardation in the offspring (6). However, the consequences of mild-to-moderate iodine deficiency in pregnancy on child neurodevelopment are less well established (4).
Differences in results between studies might be related to methodological differences (e.g. measurement of iodine status, selected reference group and available data on confounders), the age of assessment of the neurodevelopmental outcome of interest, the timing of the iodine measurements, and the relative severity of iodine deficiency in the population. While the main focus in the literature has been on the effects of iodine deficiency, some studies have suggested adverse effects of supplementary intake or excess iodine on either maternal thyroid function (17,18), fetal thyroid function (19,20), or child neurodevelopment (10,15,16).
International health authorities have similar recommendations to ensure optimal iodine status in pregnancy (21)(22)(23). It is universally recognized that any necessary iodine supplementation should be commenced before, or as early as possible in pregnancy to achieve adequate iodine intake, owing to the susceptibility of the fetal brain to iodine deficiency (23). However, whether the effect of iodine on child cognition varies during different stages of pregnancy is unknown. We therefore assessed the association between maternal iodine status in pregnancy and child IQ across three cohorts of differing iodine status, and investigated potential effect modification by gestational age.

Study design and populations
This study was embedded in three cohort studies: Generation R (The Netherlands), INfancia y Medio Ambiente Project (INMA; Spain, three regions), and the Avon Longitudinal Study of Parents and Children (ALSPAC, United Kingdom). Their study designs have been described elsewhere (24)(25)(26)(27); the ALSPAC study website contains details of all the data that are available through a fully searchable data dictionary and variable search tool (28). For the current study, mother-child pairs were included if a measure of urinary iodine and creatinine concentration during pregnancy, and child IQ were available. Exclusion criteria were multiple pregnancy, fertility treatment, medication affecting the thyroid and pre-existing thyroid disease. Ethical approval was obtained from: the Medical Ethical approval by participants and/or parents or guardians of the children was given by signed informedconsent.

Maternal iodine status
Urinary iodine concentration (UIC) and creatinine concentration were measured in spot urine samples stored at -20°C after collection. As part of this study, additional urine samples were analysed for iodine and creatinine concentration and existing measurements from each cohort (7,19,29) were also included.
The additional measurements were performed in the same laboratories as where the existing measurements were performed. The laboratories were registered with EQUIP and used certified reference materials (Seronorm urine Levels one and two, Nycomed, Norway) for the verification of results. In Generation R, UIC was measured by the Sandell-Kolthoff method. In INMA, UIC was measured using paired-ion reversed-phase, high-performance liquid chromatography with electrochemical detection at a silver working electrode (Waters Chromatography, Milford, MA). In ALSPAC, UIC was measured on a dynamic reaction cell inductively-coupled plasma mass spectrometer. Urinary creatinine concentration was determined by the Jaffe rate method in all cohorts.
More information on the measurement methods and the variability between assays can be found in the Supplemental material (30, page 2).
In a subset of women repeated measures of urinary iodine and creatinine were available; we used the earliest available sample as an indicator of iodine status. The urinary iodine-to-creatinine ratio (UI/Creat) was used as a measure of iodine status. Due to possible contamination of UIC by the use of iodine-containing test strips in ALSPAC (31), UIC above 500 µg/L and/or UI/Creat >700 µg/g were excluded from the analyses in this cohort (N=363). These cut-offs were based on previous work in ALSPAC and from other studies of pregnant women in the United Kingdom (7,32,33). We grouped women's results by UI/Creat: (i) <150 µg/g, (ii) 150 to < 500 µg/g, and (iii) ≥500 µg/g; based on WHO classification these groups broadly relate to iodine deficiency, sufficiency, and excess, respectively.

Maternal thyroid function
Thyroid stimulating hormone (TSH) and free thyroxine (FT4) were measured according to different methodologies between cohorts, which are described in detail elsewhere (34)(35)(36). For the analysis, FT4 and TSH concentrations were logarithmically transformed and cohort-specific SD scores were calculated with a mean of 0 and a SD of 1 based on the data of TPOAb-negative women (TPOAb measurements were available in Generation R and ALSPAC). A TPOAb titre ≥ 60 IU/mL and ≥ 6 IU/mL was considered as positive in Generation R and ALSPAC, respectively. These cut-offs were determined by the assay manufactures.

Non-verbal and verbal IQ
In Generation R, non-verbal IQ was assessed at a median age of 5.9 years using a subset of the Snijders Oomen Nonverbal Intelligence Test (2.5-7-Revised) (37) (39). In ALSPAC, non-verbal and verbal IQ were assessed at a median age of 8.6 years using the Wechsler Intelligence Scale for Children, 3rd UK edition (40). Except for verbal IQ ascertainment in Generation R, which was a parental questionnaire, all other measurements were performed by psychologists or trained staff. To homogenize the different scores, raw cohort-specific scores were standardized to a mean of 100 and a SD of 15. Children with IQ scores below 50 or above 150 (n=3) were considered as outliers and excluded from analyses. Suboptimal IQ was defined as an IQ score below 85.

Potential confounding variables
Information on maternal age, educational level (low, middle, high), ethnicity/country of birth (cohortspecific categories), parity (zero, one, two or more), pre-pregnancy body mass index, and smoking during pregnancy (never smoked, smoked in the beginning or until pregnancy confirmed, continued smoking) was collected by questionnaires administered during pregnancy. Gestational age at urine sampling was defined using ultrasound or last menstrual period. Child sex and age at time of the IQ assessment was obtained during the study visits.

Statistical Analyses
UI/Creat was not normally distributed and was therefore transformed using the natural logarithm; backtransformed values are shown in plots for better interpretation. We studied the associations of UI/Creat with child non-verbal and verbal IQ by using a one-step and a two-step approach. In the one-step approach, data from the cohorts were pooled and we performed standard multivariable linear regression models with and without a quadratic term to investigate the possible non-linear nature of the associations. Non-linearity was also investigated by using ordinary least-squares linear regression models with restricted cubic splines with three knots. With ANOVA we tested the null hypothesis that child mean IQ was similar across the full range of the natural logarithm of UI/Creat. The decision for linear regression models instead of multilevel models for the one-step analyses was made because we found no difference between multilevel models with random intercepts and/or slope per cohort versus standard linear regression correcting for cohort (e.g. cohort-specific variable ethnicity/country of birth) when assessed using the Akaike information criterion and log-likelihood tests. In the two-step approach we first studied the associations UI/Creat < 150 µg/g and UI/Creat ≥ 500 µg/g with child IQ by using linear regression models in each cohort separately. In these analyses the reference group consisted of women with UI/Creat 150-500 µg/g. We then combined the cohort-specific effect estimates using random effects meta-analyses.
Potential effect modification according to gestational age was analyzed by adding a product interaction term between UI/Creat and gestational age to the one-step approach models. Due to the known constraints of statistical power for interaction analyses, a P-value below 0.15 for interaction terms was used to screen for potential relevant modification (41). We further quantified potential relevant differences by performing stratified analyses by tertile of gestational age (≤12 weeks, >12 to ≤14 weeks, and >14 weeks). We also studied associations with suboptimal IQ (score<85) by combining cohort-specific estimates from logistic regression models into random effects meta-analyses.
Sensitivity analyses aimed to study: i) the associations of UI/Creat with verbal IQ in motherchild pairs from INMA and ALSPAC only, as verbal IQ was assessed at a pre-school age in Generation R; ii) the association between UI/Creat and maternal TSH and FT4 SD scores within the ± 4 SD range around the mean, as TSH and FT4 values outside this range were considered as outliers (n=19 for TSH, n=5 for FT4); and iii) whether the association between UI/Creat and IQ could potentially be explained by maternal thyroid function by adjusting for FT4 and TSH in the models.
Heterogeneity between cohorts was assessed using the Cochran Q test and the I 2 statistic (42).
All models were adjusted for the potential confounding variables. However, due to collinearity between maternal ethnicity/country of birth, child age at IQ ascertainment, and cohort, we only adjusted for maternal ethnicity/country of birth in the one-step approach models.
We applied inverse probability weighting to take into account the potential differential loss to follow-up (eTable 1) (30), i.e. to account for selection bias that potentially arises when only the population with available data on iodine status and child IQ is included as compared to a full initial cohort recruited at pregnancy (43). Briefly, we used information available for all participants at recruitment to predict the probability of participation in the study, and used the inverse of those probabilities as weights in the analyses so that results would be representative for the initial populations of the cohorts. Additionally, missing values in the potential confounding variables were imputed using chained equations (44). A total of 25 datasets were generated. A P-value < 0.05 was defined as statistically significant. Statistical analyses were performed in STATA (version 14.0; StataCorporation, College Station, TX) and R statistical software (version 3.3.2, package rms).

Non-verbal IQ
Using pooled data in the one-step approach, we observed a positive linear association between the UI/Creat and mean non-verbal IQ (Figure 2a and eTable 2) (30), although this association was not statistically significant. Using the two-step approach where we combined cohort-specific effect estimates using random effects meta-analysis, neither UI/Creat < 150 µg/g nor UI/Creat ≥ 500 µg/g

Verbal IQ
Using the one-step approach, we observed a positive curvilinear association between UI/Creat and verbal IQ (Figure 3a and eTable 2) (30). There was a positive linear association when excluding the measures in pre-school children from Generation R. Using the two-step approach, neither UI/Creat < 150 µg/g nor UI/Creat ≥ 500 µg/g was associated with verbal IQ (-0.6 points, 95% CI -1.

Effect modification according to gestational age
The continuous association of UI/Creat with non-verbal IQ did not differ according to gestational age at measurement (P for interaction term=0.306). By contrast, we identified possible effect modification by gestational age in the association with verbal IQ (P for interaction term=0.078). Stratification by tertile of gestational age showed a positive curvilinear association of UI/Creat with mean child verbal IQ with an overall effect of roughly 5 IQ points during the first 12 weeks of pregnancy (Figure 4a

Discussion
This meta-analysis of individual-participant data showed that a lower UI/Creat during pregnancy was associated with lower verbal IQ. The association of UI/Creat with verbal IQ was only seen up to the start of the second trimester (up to the 14 th week of gestation). By contrast, we observed no associations between IQ and UI/Creat below 150 µg/g or above 500 µg/g. Only a few of the previous single-centre studies, i.e., the Generation R and ALSPAC cohort studies, focused on child non-verbal IQ (7,14). They found no association between UI/Creat < 150 µg/g and non-verbal IQ. It was suggested that iodine deficiency in the Generation R cohort might not have been severe enough for an association to be identified (14). After combining these two cohorts of contrasting iodine status with a third mildy-deficient population (INMA), there was still no effect of iodine deficiency on non-verbal IQ.
Our meta-analysis using pre-defined cutoffs showed that UI/Creat < 150 µg/g was not associated with lower verbal IQ. The estimates we found for ALSPAC are in contrast to the strong negative association of maternal UI/Creat < 150 µg/g in the first trimester (defined as ≤13 weeks gestation) with child verbal IQ found in a previously published study from that cohort (fully adjusted: -2.9, 95CI% -5.0 to -0.8, P=0.006) (7). However, there are a few important differences between the studies. Compared to the previous publication, the ALSPAC data in our study included a larger number The importance of iodine status in the preconceptional stage for child IQ has recently been shown (45). In early pregnancy the fetus is fully dependent on the placental transfer of thyroid hormone to support the crucial processes of brain development (2). There is a need for optimal iodine supply from the initiation of conception implying that sufficient intrathyroidal iodine stores at the preconception stage may well be critical. Indeed, our results suggest that the fetus is particularly sensitive to suboptimal iodine status in the early stages of pregnancy (e.g. ≤14 week of gestation) for optimal development of verbal IQ. Effects on verbal IQ could possibly be explained by the impact of mild iodine deficiency, via thyroid hormone, on the auditory system (13,46). In our study, we did not find evidence that the association between UI/Creat and verbal IQ was mediated via maternal thyroid function. Possible explanations could be that urinary iodine excretion is a highly volatile and crude measurement of individual iodine status and/or a crude marker of thyroidal iodine availability. Alternatively, it is also possible that the effects are (in part) mediated via fetal thyroid function.
This study confirms that low iodine status is associated with a reduction in verbal IQ scores, putting these children potentially at risk of poorer academic achievement (47). Furthermore, on a national level, our findings may have implications, for example by negatively affecting economic growth (48). However, evidence for the benefits of supplementation of mild-to-moderate iodine deficient women on child neurodevelopment is still inconclusive (11,15,16,(49)(50)(51)(52)(53). A recent randomized, placebo-controlled trial showed no benefit on child non-verbal or verbal IQ of daily supplementation with 200 µg iodine (as potassium iodide) of mildly iodine-deficient women (52). In addition to the already-mentioned limitations of that trial (54), our results provide an additional explanation for the null finding; the trial randomized women up to 14 weeks of gestation, while we showed that maternal iodine status is particularly important in the first trimester. Although our study needs replication, it suggests that the trial might have missed a critical period of vulnerability in iodinedeficient women. Our results clearly suggest that further randomized controlled trials should start with iodine supplementation early in the first trimester, or preferably even before pregnancy.
The strengths of our study are the following: the same approach of analysis and harmonization of potential confounding variables across cohorts optimized comparisons; advanced statistical methods were used to overcome selection bias due to loss to follow up and missing data; UI/Creat was used as a marker of iodine status -this has been shown to be a more valid measure of iodine excretion when used in the same groups of age and sex (55), though we recognize that a single measure may not be reflective of overall iodine status in an individual. A limitation of the study is that the assessment of IQ was performed with different tools at different ages. Nevertheless, the tools measured the same construct (non-verbal or verbal IQ) and the standardization of IQ scores in each cohort facilitated comparison of results across cohorts. Sensitivity analysis in older children only (e.g. excluding children from Generation R), thus reducing the age range at which verbal IQ was assessed, confirmed the association between UI/Creat and verbal IQ. Another limitation is that UIC was measured in different laboratories using different assays; it is known that urinary iodine measurements vary between laboratories (56). We used laboratories that were registered with EQUIP and the use of certified reference materials enabled us to ensure the accuracy of results.
In conclusion, this study confirms that iodine status in pregnancy is associated with child IQ and results indicate that the development of verbal IQ of the fetus is particularly vulnerable to suboptimal iodine concentration during early pregnancy up until the start of second trimester. As such, our results suggest that iodine supplementation after the first fourteen weeks of pregnancy could be outside of the critical period during which iodine availability affects fetal brain development. However, further studies should replicate these data and study the effects of iodine supplementation.     a) continuous association, depicted as the mean child non-verbal IQ (black line) with 95% confidence interval (grey area) using pooled data. Models are adjusted for gestational age, child sex, maternal ethnicity/country of birth, maternal education, parity, maternal age, pre-pregnancy BMI, and smoking during pregnancy. The p-value was provided by an ANOVA test of the null hypothesis that child mean non-verbal IQ was similar across the whole range of the natural logarithm of UI/Creat. Forest plots, b) UI/Creat < 150 µg/g ("deficiency"), and c) UI/Creat ≥500 µg/g ("excess"), compared to the reference group of UI/Creat ≥150 to <500 µg/g ("sufficient") depicted as effect estimate (dot) with 95% confidence interval per cohort, and overall, as estimated by random-effects meta-analysis (diamond).

Figure 3 Association of maternal iodine status during pregnancy with child verbal IQ.
a) continuous association, depicted as the mean child verbal IQ (black line) with 95% confidence interval (grey area), using pooled data. Models are adjusted for gestational age, child sex, maternal ethnicity/country of birth, maternal education, parity, maternal age, pre-pregnancy BMI, and smoking during pregnancy. The p-value was provided by an ANOVA test of the null-hypothesis that child mean verbal IQ was similar across the whole range of the natural logarithm of UI/Creat. Forest plots, b) UI/Creat < 150 µg/g ("deficiency"), and c) UI/Creat ≥500 µg/g ("excess"), compared to the reference group of UI/Creat ≥150 to <500 µg/g ("sufficient"), depicted as effect estimate (dot) with 95% confidence interval per cohort and overall as estimated by random-effects meta-analysis (diamond).

Figure 4 Association of maternal iodine status during pregnancy with child verbal IQ, stratified by tertiles of gestational age.
Continuous association, depicted as the mean child verbal IQ (black line) with 95% confidence interval (grey area) restricted to: a) the first 12 weeks of gestation (lowest tertile, median UI/Creat 116 µg/g, N=2209), b) from week 12 to week 14 of gestation (middle tertile, median UI/Creat 147 µg/g, N=1776), and c) later than week 14 of gestation (highest tertile, median UI/Creat 157 µg/g. N=1879). Models are adjusted for gestational age, child sex, maternal ethnicity/country of birth, maternal education, parity, maternal age, pre-pregnancy BMI, and smoking during pregnancy. The p-value was provided by an ANOVA test of the null hypothesis that child mean verbal IQ was similar across the whole range of the natural logarithm of UI/Creat. Reported beta and 95%CI are increase in mean child IQ per natural log increase in UI/Creat. UI/Creat^2 refers to the addition of a squared logtransformed UI/Creat variable in the model. Analyses were performed using linear regression models. Analyses were adjusted for gestational age at urine sampling, child sex, maternal ethnicity/country of birth, maternal education, parity, maternal age, pre-pregnancy BMI, and smoking during pregnancy. eFigure 1 Association of UI/Creat < 150 µg/g and UI/Creat ≥ 500 µg/g during pregnancy with child suboptimal non-verbal and verbal IQ. eFigure 1 Association of UI/Creat < 150 µg/g and UI/Creat ≥ 500 µg/g during pregnancy with child suboptimal non-verbal and verbal IQ.

Supplemental material eTable 1 Distribution and comparison of maternal and child characteristics in the included and excluded population
Figure shows the association of UI/Creat <150 µg/g ("deficiency") with a) non-verbal IQ score <85 and c) verbal IQ score <85, and UI/Creat >=500 µg/g ("excess") with b) non-verbal IQ score <85 and d) verbal IQ score <85 as compared to the reference group (UI/Creat ≥150 to <500 µg/g). Associations are depicted as effect estimate (dot) with 95%CI per cohort and overall as estimated by random-effects meta-analysis (diamond). Models are adjusted for gestational age, child sex, maternal ethnicity/country of birth, maternal education, parity, maternal age, pre-pregnancy BMI, smoking during pregnancy, child age and region in INMA. eTable 3 Continuous analysis of the association of maternal iodine status during pregnancy with child verbal IQ stratified by tertile of gestational age using standard linear regression models.