Associations Between Prenatal Exposure to Air Pollution and Congenital Hypothyroidism

Abstract

Abbreviations

 
• CHT

  congenital hypothyroidism

 
• IQR

  interquartile range

 
• NICU

  neonatal intensive care unit

 
• NO₂

  nitrogen dioxide

 
• NO_x

  nitrogen oxide

 
• PM

  particulate matter

 
• PM₁₀

  particulate matter with an aerodynamic diameter ≤10 μm

 
• PM_2.5

  particulate matter with an aerodynamic diameter ≤2.5 μm

 
• TT4

  total thyroxine

Adequate availability of thyroid hormones is required for optimal brain development (1–3). Thyroid hormone deficiency present at birth can result from an underlying condition such as congenital hypothyroidism (CHT), causing intellectual disability when left untreated (4–12). Permanent CHT refers to a persistent thyroid hormone deficiency and requires lifelong treatment, while transient CHT represents a temporary deficiency, with expected remission within months and up to 3 years (3, 13–15).

Fetal thyroid gland formation occurs mainly up to week 14 of gestation; the thyroid begins to produce thyroid hormones around week 14 and reaches maturity by weeks 18–20. After this period, there remains a substantial transfer of thyroid hormones from the mother to the fetus (10, 11, 16, 17). Associations between air pollution exposures and CHT found in the first trimester might indicate a potential effect of air pollution on the formation or early development of the thyroid gland. In contrast, associations in the third trimester probably result from effects on newborn thyroid function, fetal thyroid hormone metabolism, or the placental transfer of thyroid hormones to the fetus.

Heritability studies estimate that heritable factors explain only 30%–40% of the variability in free thyroxine and total thyroxine (TT4) concentrations at birth (18, 19). Additional studies show that a broad range of chemicals to which humans are exposed bind to thyroid hormone receptors and produce complex effects on the signaling of thyroid hormones, affecting thyroid hormone function during brain development (20). Additionally, studies show associations between prenatal exposure to environmental chemicals and thyroid hormone concentrations (21–24).

Ambient air pollution is a widespread environmental risk factor for various health conditions. Two previous studies observed that prenatal exposure to particulate matter (PM), especially during early and mid-pregnancy, was associated with higher newborn TT4 concentrations and was inversely associated with newborns’ thyroid-stimulating hormone and free thyroxine (25, 26). In addition, 4 studies found that PM with a diameter of ≤2.5 μm (PM_2.5) and nitrogen dioxide (NO₂) could interfere with maternal thyroid function during early pregnancy (26–29). These studies presented pioneering results, but the issue deserves further research given that they were limited in their sample size, did not examine associations with clinical diagnoses of thyroid disease, and did not specifically assess residual confounding. In this study, we aimed to examine associations of trimester-specific exposure to 4 main criteria air pollutants with CHT, using data from the national program for neonatal screening in Israel, and assess possible residual confounding.

METHODS

Study design and study population

This is a historical cohort study based on the Israeli national program for neonatal screening, which is managed by the Israeli Ministry of Health. Our initial study population included all term newborns who were born in Israel during 2009–2015, resided in localities within the exposure models’ boundaries, and were not hospitalized in a neonatal intensive care unit (NICU) (n = 1,016,261). Of this cohort we excluded neonates with addresses that could not be geocoded (n = 126,527) and further excluded those who were born after March 28, 2015 (n = 102,350), because we could not estimate their 9-month postnatal exposure (our exposure models contain data up to the end of 2015). Last, we excluded neonates with missing data on gestational age (n = 90,923), resulting in a sample of 696,461 neonates for analysis with PM exposure models. Because NO₂ and nitrogen oxide (NO_x) exposure models cover only the central coastal areas of Israel, where 54% of the study population resides, a population of 377,203 neonates was available for analysis with data on these pollutants (Web Figure 1, available at https://doi.org/10.1093/aje/kwab187).

We received neonatal data from the Israeli Ministry of Health, including hospital department (NICU/non-NICU), sex, birth weight, gestational age, ethnicity, CHT diagnoses, and residential addresses. Ethical permission for the study was received from the supreme ethics committee of the Israel Ministry of Health. Informed consent was waived by the ethics committee.

Outcome assessment

Newborn screening tests in Israel are routinely performed for each newborn 48–72 hours after birth. Heel-stick blood spots are sent to the national neonatal screening laboratory in Sheba Medical Center, Be’er Sheva, Israel. Non-NICU newborns are screened for CHT by measuring TT4 blood concentrations, with a subsequent measurement of thyroid-stimulating hormone for newborns in the lowest daily TT4 decile (30). We further describe the clinical diagnosis process that follows in Web Figure 2. Final CHT diagnoses are reported back to the national lab. In newborns initially diagnosed with permanent CHT, the diagnosis might be changed to transient CHT if the newborn is taken off treatment, up to age 3 years. Such changes are diagnosed at pediatric endocrinology units and are also reported to the national lab.

In an initial examination of the data, we detected 1 case of transient CHT born during 2009–2010 (Web Figure 3) and an excess number of permanent CHT cases compared with those born after 2010. Therefore, we concluded that before 2011, the distinction between these 2 conditions was not documented correctly in the database. Consequently, we combined permanent and transient CHT cases into a single outcome variable: CHT.

Exposure assessment

We examined 2 fractions of PM, PM_2.5 and PM_10–2.5 (PM with aerodynamic diameter of 2.5–10 μm), and 2 markers of traffic-related pollution, NO₂ and NO_x. PM_2.5 and PM₁₀ exposure estimates were generated from hybrid spatiotemporal models, with a daily 1-km resolution, which use satellite data along with various meteorological and spatial PM predictors. We examined PM_2.5 and PM_10–2.5 (the difference between PM₁₀ and PM_2.5), which differ in their ability to penetrate the human body. The PM models were validated using standard methods, with out-of-sample cross-validation R² values of 0.79 and 0.72 for PM₁₀ and PM_2.5, respectively (31, 32). To estimate exposure to NO₂ and NO_x, we used half-hourly 500-m-resolution output of the optimized dispersion models, incorporating emission input from various source types. These models use land use, air pollution monitoring, meteorological, and traffic data, and cover the central coastal areas of Israel (33–35). The half-hourly data was further lumped as mean daily exposure data. The leave-one-out cross-validated performance in estimating the spatiotemporal NO_x and NO₂ on the weekly scale in 2015, for example, was of R² = 0.67 and R² = 0.75, respectively (36–38).

Exposure data were available to us from January 2008 until the end of 2015. We aggregated mean daily exposure estimates for each air pollutant at each grid point into mean weekly exposure estimates based on calendar weeks.

Residential addresses were based on official maternal addresses at the date of birth, originating from the Ministry of Interior, and were geocoded with ArcGIS version 10.6 (ESRI, Redlands, California) and street reference data from HERE (HERE technologies, Tel Aviv, Israel) (39). We have successfully mapped 62% of the addresses to their exact residential location and additional 9% to their street. The other addresses were geocoded with lower accuracy: 3% based on neighborhood center in Jerusalem, 17% based on large locality centers (localities with populations of >10,000 residents), and 9% based on small locality centers (localities with populations of ≤10,000 residents). We considered house, street, and small locality geocoding to be higher accuracy for the purpose of sensitivity analyses. We have further generated a proportional sample of 100 geocoded addresses, randomly selected after stratification by level of geocoding, and performed an accuracy assessment by manually comparing point geocoded locations with 2 base maps: OpenStreetMap and World Street Layer. All sample addresses were found to be geocoded correctly.

We assigned each neonate to the grid area within which the neonate resided, based on the residential address. Because the grids for PM and NO₂/NO_x models are different, we gave each neonate 2 grid identifiers, one for each grid. Next, we set neonates with their gestational weeks based on date of birth and gestational age and merged neonate data and mean weekly exposure estimations, based on grid identifier and gestation calendar weeks. Finally, we aggregated, for each neonate, weekly mean air pollutant exposure into mean trimester exposures, based on gestational weeks: first trimester, 1–13; second trimester, 14–27; and third trimester, week 28 until birth. For addresses that were geocoded to the locality or neighborhood center, we have assigned the mean pollutant concentrations over the entire locality or neighborhood.

Negative control exposures

To assess residual confounding, we used postnatal exposures to the same pollutants in the same address as negative control exposures (40–43). These exposures cannot affect CHT status at birth because they occur after birth, but they are presumably affected by the same potential unobserved confounders as the prenatal exposures, as demonstrated in the directed acyclic graph in Web Figure 4. According to causal inference theory, in multivariable models that include the prenatal exposure, the negative control exposure will be associated with the outcome beyond random error if and only if there is residual confounding. We used Pearson correlations among trimester and postnatal 3-month exposure periods (first 3 months, 4–6 months, or 7–9 months after birth) to determine the postnatal exposure period for each mutually adjusting model. The postnatal 3-month period with the lowest absolute correlation value for each pregnancy trimester was selected, in order to avoid multicollinearity and give more accurate point estimates (44) (Web Table 1).

Statistical analyses

We used multivariable logistic regression models to estimate odds ratios and 95% confidence intervals of CHT and trimester-specific exposures to PM_2.5, PM_10–2.5, NO₂, and NO_x, in separate models for each pollutant. The odds ratios were reported per 1-interquartile-range (IQR) increase in each pollutant. We examined associations of trimester-specific exposures with CHT using models adjusted for covariates (see below). We used logistic regression models that included 3 (mutually adjusted) exposure periods, representing the 3 trimesters, in order to account for correlations between trimester exposures (45).

We identified potential confounding and collider bias (due to the exclusion of NICU neonates), which might induce noncausal associations, using a directed acyclic graph (Web Figure 5). Accordingly, the models adjusted for the following covariates: ethnicity, socioeconomic status, geographical area, conception season, conception year, gestational age, birth weight, and child’s sex (Web Figure 6). Our directed acyclic graph shows that birth weight might be a partial mediator in the associations between air pollution exposures and CHT.

Subdistrict data were aggregated into 4 geographical areas: North, Center, Tel Aviv, and South. We have aggregated ethnicity into Jewish, Arab, and other/missing. We extracted the 2008 socioeconomic status index from the Israel Central Bureau of Statistics (46) as a continuous variable attributed to small statistical areas, and aggregated it into tertiles. Additionally, we adjusted for the following birth weight categories: <3,000, 3,000–3,500 and >3,500 g.

To assess residual confounding, we used logistic regression models with 2 exposure periods each: one pregnancy trimester period and one 3-month postnatal period (used as a negative control, as explained above). To estimate whether birth weight acts as a mediator, we fitted logistic regression models that further adjusted for birth weight and compared them with models that did not adjust for birth weight. To evaluate whether child’s sex modifies the association, we fitted stratified and interaction models by sex. In addition, we conducted sensitivity analyses to evaluate the effects of exposure estimation errors that were introduced by inaccurate geocoding, by excluding newborns with addresses geocoded to large locality centers.

All analyses were performed using R, version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria).

RESULTS

Our study population included 696,461 infants; the majority were Jewish (75%), and 20% were Arabs. We identified 622 newborns with CHT (52% were female), resulting in CHT prevalence of 8.9 per 10,000 live births (Table 1). The average entire-pregnancy exposures to PM_2.5, PM_10–2.5, NO₂, and NO_x in our study population were 21.1 (standard deviation, 2.0) μg/m³, 29.9 (standard deviation, 5.0) μg/m³, 13.3 (standard deviation, 3.6) ppb, and 20.9 (standard deviation, 7.2) ppb, respectively (Web Table 2).

In models with single-trimester exposures, adjusted for potential confounders, we found that higher third-trimester NO_x and NO₂ exposures were positively associated with CHT (for example, per NO_x IQR, odds ratio = 1.22, 95% confidence interval: 1.07, 1.39) (Figures 1K–1L; Web Figure 7 (color version)). Mutually adjusting for the 3 trimester periods yielded results comparable to the single trimester periods, for associations of third-trimester periods (Figures 2C–2D; Web Figure 8 (color version)). In models that included both pregnancy trimester exposure and 3-month postnatal exposures (as negative controls), we found that the associations of third-trimester exposures with NO_x and NO₂ with CHT were almost unchanged relative to single-period models. The matching negative control exposure periods for the first- and second-trimester NO_x and NO₂ exposures resulted in null associations with CHT. However, the matching negative control exposure periods for the third trimester exposures resulted in negative associations of NO_x and NO₂ with CHT (e.g., per NO_x IQR, odds ratio = 0.86, 95% confidence interval: 0.73, 1.02) (Figures 3K–3L; Web Figure 9 (color version)).

Figure 1

Adjusted odds ratios (and 95% confidence intervals) per 1-interquartile-range (IQR) increase in exposure levels for associations between 4 air pollutants and congenital hypothyroidism from separate models with 1 air pollution exposure period (trimester or postnatal 3-months), among Israeli newborns, 2009–2015. A–D) First trimester; E–H) second trimester; I–L) third trimester. A, E, I) Particulate matter (PM) with an aerodynamic diameter ≤2.5 μm (IQR = 2.6 μg/m³); B, F, J) PM_10–2.5 (IQR = 6.9 μg/m³); C, G, K) nitrogen dioxide (IQR = 5.2 parts per billion (ppb)); D, H, L) nitrogen oxides (IQR = 9.7 ppb). All models adjusted for ethnicity, socioeconomic status, geographical area, conception season, conception year, gestational age, birth weight, and child’s sex.

Open in new tabDownload slide

Figure 2

Mutually adjusted odds ratios (and 95% confidence intervals) per 1-interquartile-range (IQR) increase in exposure levels for associations between 4 air pollutants and congenital hypothyroidism from joint trimester models with 3 pregnancy-trimester air pollution exposure periods, Israeli newborns, 2009–2015. A) Particulate (PM) matter with an aerodynamic diameter ≤2.5 μm (IQR = 2.6 μg/m³); B) PM_10–2.5 (IQR = 6.9 μg/m³); C) nitrogen dioxide (IQR = 5.2 parts per billion (ppb)); D) nitrogen oxides (IQR = 9.7 ppb). All models adjusted for ethnicity, socioeconomic status, geographical area, conception season, conception year, gestational age, birth weight, and child’s sex.

Open in new tabDownload slide

Figure 3

Mutually adjusted odds ratios (and 95% confidence intervals) per 1-interquartile-range (IQR) increase in exposure levels for associations between 4 air pollutants and congenital hypothyroidism with 2 air pollution exposure periods (trimester and postnatal 3-months), Israeli newborns, 2009–2015. A–D) First trimester; E–H) second trimester; I–L) third trimester. A, E, I) Particulate matter (PM) with an aerodynamic diameter ≤2.5 μm (IQR = 2.6 μg/m³); B, F, J) PM_10–2.5 (IQR = 6.9 μg/m³); C, G, K) nitrogen dioxide (IQR = 5.2 parts per billion (ppb)); D, H, L) nitrogen oxides (IQR = 9.7 ppb). All models adjusted for ethnicity, socioeconomic status, geographical area, conception season, conception year, gestational age, birth weight and child’s sex.

Open in new tabDownload slide

In mutually adjusting models with trimester exposures and 3-month postnatal exposures without adjusting for birth weight, we found associations comparable to those from models that also adjusted for birth weight (Web Figure 10). In models that were separate for permanent and transient CHT conditions, for neonates born in 2011–2015, we found comparable associations, with the main difference being in the width of the confidence intervals (Web Figure 11). Stratification by sex for mutually adjusting models revealed comparable associations of air pollution exposures for female versus male children (Web Figure 12). Interaction models by sex of third-trimester NO₂ and NO_x exposure showed no interactions (P value = 0.63 and 0.33, respectively). Excluding neonates with addresses geocoded to large localities (n = 138,993) resulted in comparable associations with CHT (Web Figure 13).

DISCUSSION

In this historical cohort study of 696,461 term Israeli newborns, we found that third-trimester traffic-related exposures were positively associated with CHT. Comparable results of the mutually adjusting trimester periods strengthened our findings by reducing the likelihood that correlations bias the associations we found among exposure periods.

Using a negative control approach, we further conclude that the main associations we detected are not a result of residual confounding, given that the matching postnatal exposures show null or negative associations in mutually adjusting models. These analyses show that, if residual confounding is present, it actually works in the opposite direction of the causal effect, producing a negative association for the negative control exposure, and therefore that the real causal effect might even be more substantial.

These findings suggest that exposure to traffic pollution might impede the function of the fetal thyroid gland or might affect the neonates’ thyroid hormone levels by obstructing its thyroid hormone metabolism or the placental transfer of thyroid hormone to the fetus.

Recent studies observed associations of prenatal PM and traffic-related pollution exposures on the function of maternal and neonatal thyroid glands (25–29). Our results extend these observations by exploring clinical CHT diagnoses. Even though no study has specifically explored associations of air pollution with CHT, the disease will cause a low neonatal free thyroxine concentration. An increase in neonatal TT4—as observed in a study from California (25)—is expected if prenatal air pollution increases neworn thyroxine-binding globulin concentrations, which in turn could lead to higher TT4 but lower free thyroxine in neonates. Henceforth, the existing literature is consistent with our findings on CHT, while our research expands on the specific pollutants and critical windows for which the associations are more likely to be causal.

The differences in the findings among studies can be related to the fact that in California, traffic is a major source of PM_2.5, whereas in Israel NO₂ and NO_x are good traffic markers, but traffic is a minor contributor to the overall PM_2.5 (38, 47–49). Accordingly, the correlation of PM_2.5 with traffic-related measures in Israel is weak (r = 0.48 with NO₂ and r = 0.50 with NO_x), given that most of the PM contents is transported to the country from remote locations such as Europe, North Africa, and the nearby Middle East deserts, and has a chemical composition different from traffic-related PM (50).

Our study has several limitations. We had no data on 2 key maternal factors: iodine status and thyroid function. Lack of access to health care is one of the determinants of persistent maternal thyroid dysfunction, and is somewhat related to residential address, which determines much of the exposure in our study. However, every Israeli citizen is covered by the national health insurance throughout life, including access to health-care services with very limited fees. Actual use of health-care resources in Israel does vary by geographical area, ethnicity, and socioeconomic status despite the national coverage; yet, our statistical models adjusted for these variables, so residual confounding by access to health care, if exists, should be minimal. In addition, such confounding would have been detected in our negative control analyses.

The lack of data on maternal thyroid function did not allow us to exclude pregnant women with overt hyper- or hypothyroidism. However, maternal thyroid function is a potential mediator of the possible effect of prenatal traffic-related pollution on CHT, and not a potential confounder (Web Figure 6), as suggested previously (28). One of the determinants of maternal thyroid function is maternal iodine status. Recent studies have suggested that iodine deficiency might be prevalent among pregnant women in Israel (51, 52). Iodine concentrations could be associated with the exposures we have examined mainly through geographical area—for example, it was suggested that the increased use of desalinated drinking water in Israel might promote iodine deficiency—but also through demographic factors. However, our analyses adjusted for demographic factors and geographical areas, and any residual confounding left from such factors would have affected the associations with the negative control in a similar way.

Another limitation is that our exposure estimates were based on maternal residential addresses and ignored potential mobility throughout pregnancy, which might cause exposure estimation errors. Yet, we assume these potential errors to be nondifferential with respect to CHT, and thus expect any introduced bias to be toward the null. Incomplete geocoding also affected the quality of our exposure estimates. However, we have shown in our sensitivity analysis that this concern cannot account for the main findings. In addition, we had no detailed clinical data or biological samples, so other indicators of thyroid function—such as thyroxin-binding globulin—could not be examined in this study. This definitely limited our ability to study possible mechanisms, and we suggest that such mechanisms be studied in smaller samples that are richer in clinical data. Finally, despite the very large sample size of our cohort, CHT (and especially transient CHT) are not common conditions, and the absolute number of cases limited our statistical power.

Our study has several unique strengths as well. Our data set covered almost all births in Israel over a long period, thus minimizing selection bias and allowing us to adjust for geographical, social, and seasonal factors. Furthermore, we used a negative control approach to assess residual confounding, which is a major concern in observational studies. Finally, we examined clinical CHT diagnoses, giving context and elaborating on findings of previous studies, which examined thyroid hormone concentrations in pregnant women and their offspring.

In conclusion, this study suggests that third-trimester exposures to traffic-related pollution are related to CHT, and that this relationship is not a result of residual confounding. Further studies are needed to replicate these results, examine the most critical gestational weeks, evaluate effect modification and interactions, and inspect the underlying biological mechanisms.

ACKNOWLEDGMENTS

Author affiliations: Braun School of Public Health and Community Medicine, The Hebrew University—Hadassah, Israel (Ruthie Harari-Kremer, Ronit Calderon-Margalit, Raanan Raz); The Advanced School for Environmental Studies, The Hebrew University, Jerusalem, Israel (Ruthie Harari-Kremer); Department of Internal Medicine, Academic Center for Thyroid Diseases, Erasmus University Medical Center, Rotterdam, the Netherlands (Tim I. M. Korevaar); Department of Statistics and Operations Research, Tel Aviv University, Tel Aviv, Israel (Daniel Nevo); Civil and Environmental Engineering, Technion, Haifa, Israel (David Broday); Technion Center of Excellence in Exposure Science and Environmental Health, Haifa, Israel (David Broday); Department of Geography and Environmental Development, Ben-Gurion University of the Negev, Be’er Sheva, Israel (Itai Kloog, Alexandra Shtein); Public Health Department, Ben-Gurion University of the Negev, Be’er Sheva, Israel (Itamar Grotto); Israeli Ministry of Health, Jerusalem, Israel (Isabella Karakis); and Pediatric Endocrinology Unit, Soroka University Medical Center, Faculty of Health Sciences, Ben-Gurion University of the Negev, Be’er Sheva, Israel (Alon Haim).

This study was funded by a research grant from the Environment and Health Fund to R.R. (grant RPGA 1601).

We thank Dr. Shlomo Almashanu, the director of the Israel National Center for Newborn Screening.

Conflict of interest: none declared.

REFERENCES