Maternal thyroid disorder in pregnancy and risk of cerebral palsy in the child: a population-based cohort study

One study population is register-based and comprises children born in the eastern part of Denmark in 1979–1996, and children born in all of Denmark in 1997–2007. In total, 1,277,101 live births were identified through the Danish Civil Registration System [19]. Individual-level data from nationwide registers on the children and their mothers were linked by use of the unique personal identifier to which all live-born children in Denmark are assigned. We excluded infant deaths (n = 7022), leaving 1,270,079 children for the final analyses.

The other study population is derived from the MOthers and BAbies in Norway and Denmark (MOBAND) collaboration cohort [20], which consists of pooled data from the Danish National Birth Cohort (DNBC) and The Norwegian Mother and Child Cohort study (MoBa) [21, 22]. MOBAND includes 210,400 live-born children born in 1996–2009 on whom detailed information on prenatal exposures was obtained around gestational week 16 and 31 by telephone interviews in DNBC, and around gestational week 17 and 30 by self-administered questionnaires in MoBa, as described in more detail elsewhere [20]. We excluded children without any prenatal information from the earliest data collection (n = 16,691), infant deaths (n = 634), and children with incomplete information on maternal thyroid disorder (n = 157), leaving 192,918 children for the final analyses.

Diagnoses of thyroid disorder in the Danish mothers were derived from the Danish National Patient Register [23], which keeps information on hospital admissions since 1977 and outpatient visits since 1995. Until 1994, all diagnoses were coded using the eighth version of the Internal Classification of Diseases (ICD-8), and since 1995, the tenth version (ICD-10). The Danish National Prescription Registry [24], established in 1995, provided the Anatomical Therapeutic Chemical (ATC) codes indicating redeemed prescriptions dispensed from Danish pharmacies. The Danish Medical Birth Registry [25] and population registers in Statistics Denmark [26] provided information on characteristics of the Danish participants, while the Medical Birth Registry of Norway [27] provided information on the Norwegian participants.

Maternal hypothyroidism and hyperthyroidism in the register-based study population were defined by a hospital diagnosis and at least one redeemed prescription of the appropriate medication, i.e. thyroid hormone (ACT-code: H03A) for hypothyroidism and anti-thyroid medication (ACT-code: H03B) for hyperthyroidism, to enhance the validity of the identified disorders. Hypothyroidism was identified by ICD-8 codes 243.99 and 244.00–244.09, and by ICD-10 codes E00, E03.0-E03.9 and E89.0, excluding 244.02, E03.0A, E03.1B, and E03.4. Hyperthyroidism was identified by ICD-8 as 242.00–242.29 and by ICD-10 as E05.0-E05.9, excluding E05.4, E05.8A, and E05.9A. There were some exceptions; when a first-time diagnosis was recorded before the establishment of the Danish National Prescription Registry in 1995, the diagnosis code exclusively determined exposure status. Further, as the Danish National Patient Register does not keep information on diagnoses made by general practitioners, we also defined hypothyroidism by at least two redemptions of thyroid hormone prescriptions and no redemptions of anti-thyroid medication prescription, and we defined hyperthyroidism by at least two redemptions of anti-thyroid prescriptions; regardless of thyroid hormone prescriptions, in case of no record of thyroid diagnosis [28‐30] (see Additional file 1: eMethod 1 for information on additional coding). We included all thyroid disorders recorded until 5 years subsequent to pregnancy until 2010 inclusive. Time of identification of thyroid disorder was defined as the day of the first diagnosis code or redeemed prescription, whichever was recorded first, and categorized into; ‘before pregnancy’, ‘during pregnancy’, and ‘within 5 years after pregnancy’.

In MOBAND, information on maternal thyroid disorder in pregnancy was based on self-reports from the earliest data collection in pregnancy, which was around 16–17 weeks of gestation. The data collected in MoBa only distinguished thyroid disorder overall and could not separate hypo- and hyperthyroidism. We used information from the second pregnancy interview of DNBC and the earliest questionnaire of MoBa to define the use of thyroid medication in pregnancy.

Danish CP cases were identified through the Danish National Cerebral Palsy Registry that includes all children surviving the first year of life with a neuro-pediatrician validated diagnosis of CP at age five-six years [31, 32]. In 1979–1996 the register covered the eastern part of Denmark, and in 1997–2007 it was nationwide. Approximately 80% of the Norwegian cases were identified though the Cerebral Palsy Registry of Norway [33]. The remaining cases were identified in record linkage with the Norwegian Patient Registry and verified by neuro-pediatricians’ examinations of medical records [34]. We assessed all CP subtypes combined and the two major subtypes: unilateral and bilateral spastic CP.

We used register-based data to form the variables: the year of child’s birth, maternal age (< 25, 25–29, 30–34, ≥35 years), child’s sex (boy, girl), gestational age (≥37, < 37 weeks of gestation), and maternal diabetes (no, type 1, type 2) for both study populations. For the register-based study population, we also obtained information from administrative registers about maternal educational level (basic, intermediate, higher). In MOBAND, we used self-reported information on maternal occupational status (employed, unemployed, student, receiving benefits or pension), maternal alcohol consumption per week (0, 0.5, 1–2.5, ≥3 units), and number of cigarettes smoked per day (0, 1–9, ≥10); all reported in the earliest data collection.

We used logistic regression to estimate odds ratios (OR) with 95% confidence intervals (95% CIs) for the relationship of CP with maternal thyroid disorder. Robust standard errors were used to take into account the potential dependency between siblings. To guide our decision about which potential confounders we should adjust for, we used Directed Acyclic Graphs [35]. The adjusted models included the child’s birth year, maternal diabetes, maternal age, and maternal socioeconomic position (education/occupation), and in addition smoking and alcohol consumption in pregnancy in the MOBAND study population. We imputed missing values of covariates by use of multiple imputations (see Additional file 1: eMethod 2).

In the register-based study population, we stratified by child’s sex and gestational age, respectively, to explore whether exposure to maternal thyroid disorder has a greater impact on risk of CP in boys and children born at term. Potential misclassification of exposure was assessed by examining the agreement of self-reported and register-based information on maternal thyroid disorder in the 90,088 Danish children included in both the register-based and MOBAND study population. We subsequently applied the calculated positive and negative agreement to adjust estimates for non-differential misclassification error by using a probabilistic approach [36] (for more details see Additional file 1: eMethod 3).

To assess the sensitivity of our findings to diagnostic errors or incomplete data, we ran several secondary analyses. In the register-based study population, we restricted caseness to hospital-diagnosed maternal thyroid disorders, and also to diagnoses and redeemed prescriptions recorded from 10 years before to 5 years after pregnancy. In both study populations, we also restricted analyses to children with complete data on covariates. Finally, we included children who died within the first year of life in the study population and assessed the relationship of maternal thyroid disorder to infant death in case infant death linked to maternal thyroid disorder (perhaps with brain damage) precluded the possibility of a CP diagnosis. All analyses were performed using StataSE 14 (64-bit).

In the register-based study population, 23,625 (1.9%) of the mothers had a thyroid disorder. Hypothyroidism was recorded in 12,929 (1.0%) and hyperthyroidism in 9943 (0.8%) of the mothers; we were unable to classify the condition in the remaining 753 (< 0.1%) mothers with thyroid disorders. In MOBAND, 3042 (1.6%) of the mothers reported a thyroid disorder in pregnancy, of whom 2229 (73.3%) reported use of thyroid medication. More than twice as many mothers in MoBa (2.1%) than in DNBC (1.0%) reported a thyroid disorder. Mothers with thyroid disorders were more likely to be older, have diabetes, and deliver preterm than mothers without the disorder. Also, mothers with thyroid disorders were more likely to have intermediate or higher education in the register-based study population, were more likely to be unemployed or receive benefits/pension, and to consume less alcohol and smoke less in pregnancy in MOBAND (Table 1).

CP was diagnosed in 2798 children in the register-based study population. Bilateral spastic CP was the most common subtype with 1490 cases, while 912 children had unilateral spastic CP. Maternal thyroid disorder diagnosed or treated for the first time before pregnancy until 5 years subsequent to pregnancy was not associated with CP overall or either of the subtypes: unilateral or bilateral spastic CP (Table 2). Maternal thyroid disorder identified during pregnancy was associated with increased risk of unilateral spastic CP (adjusted OR 3.1 (95% CI: 1.2–8.4) (Table 2). Sufficient statistical power was unavailable to address the timing of identification of hypo- and hyperthyroidism separately. In the stratified analyses, estimates were similar across strata of sex and gestational ages, respectively, and no interaction was suggested (p-values for interaction > 0.2, Additional file 1: eTable 1–2), though the estimates were imprecise.

In MOBAND, 402 children were diagnosed with CP of whom 47.8% had bilateral, and 37.1% had unilateral spastic CP. Six of the children with bilateral spastic CP were exposed to maternal thyroid disorder, while only one child with unilateral spastic CP was exposed, making it unfeasible to confirm or deny the register findings for unilateral spastic CP. The estimates for maternal thyroid disorder and CP overall did not suggest an association (adjusted OR 1.2 (95% CI: 0.6–2.4), Table 2), although an high, but imprecise estimate of risk for bilateral spastic CP (adjusted OR 1.9 (95% CI: 0.8–4.3)) was noted. A higher estimate of overall CP risk (adjusted OR 2.5 (95% CI: 0.8–7.9)) was seen in children exposed to an untreated thyroid disorder than to a treated disorder, but these estimates were also imprecise. Adjustment for maternal lifestyle factors did not alter the results (Table 2).

The proportion of positive agreement between self-reported and register-based information on maternal thyroid disorder was 60%, while the proportion of negative agreement was 99%. Estimates adjusted for systematic bias were unstable but suggested that non-differential misclassification of exposure biased towards the null (Table 3). Further, changing the categorization of maternal thyroid disorder and restricting the analyses to complete cases did not alter the results. Finally, we found an increased risk of infant death in children exposed to both maternal hypothyroidism (unadjusted OR 1.4 (95% CI: 1.1–1.9)) and hyperthyroidism (unadjusted OR 1.9 (95% CI. 1.4–2.4)).

Results from the register part of this study show that thyroid disorder in pregnancy is not related to bilateral spastic CP, but may possibly be related to unilateral CP. Though the statistical power of MOBAND data was limited, information on lifestyle during pregnancy enabled us to perform more thorough control for potential confounders, which did not influence the results.

We studied unilateral and bilateral spastic CP separately, because they may have distinct etiological profiles. It has been hypothesized that thyroid hormone deficiency can cause CP by altering myelination, differentiation, and migration of nerve cells [8], which would likely be reflected in bilateral damage to the brain. An increased risk of bilateral spastic CP was suggested in MOBAND data only, but the estimate was unstable, and the finding was not replicated in the larger register-based study. The most striking finding in this study, a three-fold increase in risk of unilateral CP in association with thyroid disorder identified in pregnancy, is biological plausible. Maternal thyroid disorder may affect the coagulation system and increase the risk of thrombosis (leading to ischemia) and bleeding [16, 17], and such vascular events most likely cause unilateral spastic CP [37]. In support of this line of reasoning, markers of coagulation abnormalities including Factor V Leiden mutations, which implying an increased risk of thrombosis, have been linked to spastic CP [38]; especially, the unilateral subtype, though the evidence is sparse [39].

The male excess of CP indicates perhaps a heightened vulnerability to brain injury in boys [40], and abnormal thyroid hormone levels may affect boys differently than girls [41], but we were unable to find any sex differences in the association between maternal thyroid disorder and the risk of CP. Further, we hypothesized that the risk of CP in association with maternal thyroid disorder would be elevated mainly in children born at term, as we expect prenatal factors to play a greater role in the etiology of CP in children born at term than in children born preterm [40]. The estimates were unstable after stratification by gestational age, and there was no indication of any differences in risk.

The syndrome of neurological cretinism provides a convincing indication of a link between maternal thyroid disturbances and CP. Children born to women with severe iodine deficiency have a substantial risk of impaired cognitive and motor function, and clinical findings and brain imaging compatible with CP has been observed in children with the neurologic form of endemic cretinism [42]. Maternal thyroid diseases have only been investigated in relation to risk of CP in few studies. Nelson et al. found in a cohort study of 45,559 children, of whom 189 had CP, an increased risk of CP in infants with a birth weight ≥ 2500 g who were born to women with hyperthyroidism and in infants exposed to maternal thyroid hormone and estrogen supplementation in pregnancy [7]. In another study of 183 children with CP and 549 controls without CP, more cases than controls were born to women who were treated with thyroid hormone; however, the difference was not statistically significant [12]. Recent register-based studies from Denmark by Andersen et al. [28, 43] have indicated that thyroid disorders identified subsequent to pregnancy, especially within 5 years after pregnancy, are correlated with increased risk of attention deficit hyperactivity disorder, autism spectrum disorder, and seizures. We found a tendency to increased risk of unilateral spastic CP in children born to women with thyroid disorders identified during pregnancy. These children will probably have been exposed to abnormal thyroid hormone levels in utero, as abnormal levels may be present for a period before the disorder is diagnosed and treated for the first time.

The large scale register data enabled examination of different subtypes of thyroid disorder and CP, but lacked information on maternal lifestyle factors. Although the statistical power of MOBAND was limited, among prospective cohort studies, MOBAND holds by far the largest sample of CP cases with detailed information on lifestyle collected during pregnancy, which allowed us to address such potential confounders.

CP was verified by neuropediatricians based on clinical presentation when the children were five-six years old, unaware of maternal illness during pregnancy, which enhances the validity of the CP diagnoses. Likewise, measures of maternal thyroid disorder were not affected by knowledge of whether the child had CP, because of the register recording or self-reporting during pregnancy. We restricted our analyses to children surviving to age 1 year as CP cannot reliably be diagnosed before this age. We saw an increased risk of infant death in children prenatally exposed to maternal thyroid disorders. If children with brain damage compatible with CP also are more likely to die before CP can be diagnosed, this restriction might have biased our results towards the null.

We did not have information on causes of thyroid disorders, which may be important for understanding the mechanism by which unregulated thyroid disorder is associated with risk of CP. Moreover, it is possible that autoimmune conditions confound the association. Thyroid disorders in reproductive-age women are most often autoimmune in origin, and may be associated with other autoimmune manifestations [40]. Autoimmune disorders might lead to CP either because autoantibodies are themselves pathogenic, or because of the presence of inflammation in autoimmunity, which is an established risk factor for spastic CP [44]. We adjusted for maternal diabetes, but it is possible that confounding by other autoimmune diseases and other abnormalities that coexist with thyroid disorder, for which we did not have information, had occurred.

The comparison of measures of thyroid disorder from different sources revealed that some non-differential misclassification had occurred, which may have led to an underestimation of the association of thyroid disorder and CP. In the register data, we were unable to identify women who had a thyroid disorder diagnosed outside hospital settings and who did not redeem any thyroid medication before the establishment of the Danish National Prescription in 1995. Moreover, we may have categorized some women as exposed to maternal thyroid disorder in pregnancy, even though the mother may have recovered from the disease before pregnancy, e.g. postpartum thyroiditis is often transient. In MOBAND, some women may have been unaware of their diagnoses, which also may have led to misclassification. However, in the register-based study population, we used thyroid disorder diagnosed within 5 years after pregnancy as a proxy for subclinical and asymptomatic thyroid diseases in pregnancy, since recent findings from DNBC based on blood samples drawn in early pregnancy have shown that abnormal thyroid function may be present for a period before the disorder is identified and that asymptomatic thyroid diseases are common [45]. We anticipate that much initially asymptomatic thyroid disorders would have come to medical attention within 5 years, and that the milder and most subclinical forms of thyroid disorders would be less likely to have an impact on risk of CP. Moreover, as we did not have access to information on the actual thyroid hormone level, we cannot know whether women with an identified thyroid disorder actually had abnormal thyroid hormone values during pregnancy or whether the women had a hypo- or hyperthyroid condition due to overtreatment. The natural next step is to make use of maternal blood samples collected during pregnancy to study the link between maternal thyroid disorder and CP.