Associations between gestational residential radon exposure and term low birthweight in Connecticut, USA

Abstract

Background:

Studies suggest biologic mechanisms for gestational exposure to radiation and impaired fetal development. We explored associations between gestational radon exposure and term low birthweight, for which evidence is limited.

Methods:

We examined data for 68,159 singleton full-term births in Connecticut, USA, 2016–2018. Using a radon spatiotemporal model, we estimated ZIP code-level basement and ground-level exposures during pregnancy and trimesters for each participant’s address at birth or delivery. We used logistic regression models, including confounders, to estimate odds ratios (ORs) for term low birth weight in four exposure quartiles (Q1 to Q4) with the lowest exposure group (Q1) as the reference.

Results:

Exposure levels to basement radon throughout pregnancy (0.27–3.02 pCi/L) were below the guideline level set by the US Environmental Protection Agency (4 pCi/L). The ORs for term low birth weight in the second-highest (Q3; 1.01–1.33 pCi/L) exposure group compared to the reference (<0.79 pCi/L) group for basement radon during the first trimester was 1.22 (95% confidence interval [CI]: 1.02, 1.45). The OR in the highest (Q4; 1.34–4.43 pCi/L) quartile group compared to the reference group during the first trimester was 1.26 (95% CI: 1.05, 1.50). Risks from basement radon were higher for participants with lower income, lower maternal education levels, or living in urban regions.

Conclusion:

This study found increased term low birth weight risks for increases in basement radon. Results have implications for infants’ health for exposure to radon at levels below the current national guideline for indoor radon concentrations and building remediations.

Introduction

Radon is a radioactive ubiquitous gas without color, odor, or taste. Levels of radon can be high indoors especially underground, whereas levels are low outdoors.¹ Radon progeny decayed from uranium 238 in rock and soil can enter homes with other gases through openings in foundations such as cracks.² Radon progeny exist in two forms. The radioactive progeny reacting with ambient water molecules and atmospheric gases and forming highly mobile clusters (generally with a diameter >5 nm) is referred to as ‘attached’ radon progeny and the others as ‘unattached’ radon progeny. Attached progeny continue to decay and emit radiation after inhalation.³ Due to its smaller size, unattached radon progeny (usually <5 nm diameter) can more effectively deposit in bronchial regions, consequently giving a higher exposure compared to attached progeny.⁴ Ambient fine particles may serve as a vector for radon progeny, increasing respiratory and cardiovascular effects.^3,5

Residential radon exposure is estimated to be the second leading cause of lung cancer, after tobacco use.^1,2 Inflammation and oxidative stress due to radon have been associated with pancreatic cancer and non-cancerous respiratory diseases.^2,6,7 Pregnant women and fetuses can be sensitive to radon due to rapid prenatal development. Exposure to radiation even at low dose may harm cells, which may result in adverse embryonic and fetal outcomes.⁸ Although exact mechanisms are unknown, animal models suggest biologically plausible mechanisms for radon’s effect on fetal anomalies, low fetal growth, and cardiovascular effects in pregnant women.^9,10 The mechanisms for developing fetuses include cell death, cell depletion, mitotic delay, and inhibition of cell communication.⁹

Some US epidemiologic studies suggested associations of prenatal radon exposure with gestational hypertensive disorders, congenital anomalies in newborns, and childhood asthma.^9,11–13 However, population-based evidence is limited for offspring health, especially for low birthweight, and gestational radon exposure. Furthermore, evidence is limited for trimester-specific prenatal effects of radon.

Previous population-based US studies were largely dependent on area-based radon prediction data published in the 1990s.^14–17 While informative, a limitation of those datasets is their temporally invariant county-level radon levels, which hinder capturing temporal changes in exposure over time due to mitigation. Furthermore, indoor radon concentrations are generally higher in cold seasons than in warm seasons due to the larger variance in temperature and pressure caused by heating.¹⁸ Climate factors such as ambient temperature and precipitation can non-periodically affect radon concentrations, leading to complexity in the time series of radon levels at various temporal resolutions (e.g., daily to annually).¹⁸ Therefore, temporally invariant exposure data may not accurately assess exposure levels during a crucial period of the development of disease or health outcomes. Spatiotemporally resolved radon predicting models can be useful in exploring the prenatal impacts of radon exposure on birth outcomes, benefiting studies targeting the entire population to reduce potential selection bias.

This population-based longitudinal study assessed associations between women’s residential radon exposure during pregnancy and term low birth weight of infants born in Connecticut, USA. We utilized machine learning-based radon modeling data using numerous radon test sampling results over two decades to substantially improve on the previous exposure datasets. We investigated potential associations between radon and risk of term low birth weight and its variation by socioeconomic status (SES) and urbanicity. The findings based on statewide birth data and high-resolution radon data can inform local environmental health policies by aiding understanding of the health impacts of radon and who is most affected.

Data and Methods

Study population

This observational study used individual-level electronic birth certificate data from the Connecticut Department of Public Health for births in 2016–2018. We restricted analysis to term singleton births. The outcome was term low birth weight. Using the birthweight records in the birth certificates, we identified term low birth weight as infants with birthweight <2,500 grams and ≥37 weeks of completed gestational weeks estimated by clinical gestational age. Connecticut birth certificates record how long the mother has lived at the residence at delivery. We excluded participants without this information or those who had lived at the residence at delivery <10 months. This avoids concerns of exposure misclassification from residential mobility, a common limitation in studies of environmental exposures and pregnancy outcomes.^19,20 We excluded pregnant women aged <18 years and mothers with missing values for the included covariates. As a result, 68,159 pregnant women were available for analysis (eFigure 1). Ethical approval was obtained from the Institutional Review Boards of Yale University.

Exposures

We assessed trimester-specific radon exposure using a monthly model for each ZIP Code Tabulation Area (ZCTA) based on nearly three million short-term radon measurements in the Northeast and Midwest USA. Most such measurements were collected through home inspections during property transactions, spanning 2–7 days. A national comparison found these short-term measurements to be a good proxy (R²=0.78) for seasonal concentrations.²¹ We developed a spatial random forest model to estimate interactions among 81 geologic, meteorologic, architectural, and socioeconomic predicting factors, and to predict the average monthly radon concentrations for all basement (i.e., below-ground floors) and ground-level floors (i.e., first level above-ground floors) by ZCTA. Mean absolute error of the predictions in New England, which includes Connecticut, is 0.46 pCi/L (i.e., 17.1 Bq/m³) with a 24% relative error. The Pearson’s correlation (r²) between the observed and predicted radon concentrations was 0.47 for New England. More details of the model can be found in the study by Li et al. and eAppendix.¹⁴ This research used the unit of pCi/L (picocuries/liter of air) for radon exposure assessments (1 pCi/L equals 37 Bq/m³) to match the unit used in US Environmental Protection Agency (US EPA) radon guidelines.²² ZIP codes of the mother’s residential address at delivery were used to assign exposure levels during pregnancy, using the ZCTA-level monthly radon model. We assumed this to be the residence throughout pregnancy as we excluded participants living <10 months at that address. We defined radon exposure during pregnancy as the aggregated radon levels from the month including the conception date to the previous month of delivery. We also calculated average radon levels during each trimester (<12, 13–26, and >27 gestational weeks). Given that ZIP codes and ZCTAs are highly consistent, we excluded ZIP codes that did not have identical matching ZCTAs. Of the 273 ZIP codes of the mothers’ addresses, seven ZIP codes did not have identical ZCTAs, resulting in exclusion of seven participants with these seven ZIP codes from our analysis.

Covariates

We adjusted for individual-level covariates based on prior knowledge of relation to term low birth weight²³: mother’s age, infant’s sex, infant’s race–ethnicity (non-Hispanic White, non-Hispanic Black, American Indian and Alaska Native, Asian, Hispanic, and other as defined by the electronic birth data), Apgar five score, whether the mother was married to the biological father at any time between conception and birth (yes vs. no), smoking during pregnancy (yes vs. no), alcohol consumption during pregnancy (yes vs. no), parity (≥1, 0), insurance type (public, private, self-pay, and other), maternal education (9–12th grade, completed high school, some college, associate degree, bachelor’s degree, postgraduate degree), gestational diabetes (yes vs. no), gestational hypertension (yes vs. no), eclampsia (yes vs. no), and prenatal care (yes vs. no). Models included ZCTA-level socioeconomic and other demographic variables, assigned based on ZIP codes of participants’ address, for tertiles of the percent of the population that is Black, quartiles of percent of total housing units that are single-family houses, tertiles of percent of population in poverty, and tertiles of median household income (USD), obtained from the 5-year estimates of the American Community Survey, 2014–2018. Our model included a linear term for ZIP code-level average ambient temperature during pregnancy for each participant.^24,25 Models adjusted for an indicator variable of urbanicity; ZIP codes located in or overlapping with census blocks classified as Urban Areas in the 2010 US Census²⁶ were defined as urban regions and other ZIP codes were defined as rural.

Statistical analysis

We used logistic regression models to assess associations between radon exposure and term low birth weight. We estimated the exposure–response associations as odds ratios (ORs) and 95% confidence intervals (CI) in four quartiles (Q1–Q4) of exposure with the lowest exposure group (Q1) serving as the reference. Our models included all individual-level covariates listed above. We analyzed associations with term low birth weight for both basement and ground-level radon exposures, assessed as mean levels over the entire length of pregnancy as well as for each trimester separately. For the latter, we considered two approaches to estimate trimester-specific associations as exposures are often correlated across trimesters (eFigure 2). In our first approach for different trimesters, we grouped exposure levels into quartile groups based on the mean radon levels for each trimester. Then, we used the second approach with trimester-specific radon exposures adjusted for the correlation across trimesters using published methods.²⁷ For this, we used a linear regression: $E\left[P_c^i\right]={\beta }_1+{\beta }_2{P}_a^i+{\beta }_3{P}_b^i$. The $P^i$ is exposure over trimester a, b, or c for birth i, ${\beta }_1$ is intercept, and ${\beta }_2$ and ${\beta }_3$ are the associations between exposures. We repeated this regression using each trimester as the reference trimester. Then, we grouped the residuals of this regression for each trimester into quartiles and considered them as correlation-adjusted trimester-specific exposure levels. As the range of radon levels across participants differed by trimester, we categorized exposures into quartiles for each specific analysis (eTable 2).

We examined whether associations between radon and term low birth weight differ by urbanicity, infant’s race–ethnicity, household income, and maternal education, separately. Residential radon levels are affected by complex and interconnected contributing factors such as housing conditions, urbanicity of the residence, and socioeconomic characteristics.²⁸ A previous study showed that the correlation between house values and residential radon levels was not consistent by urbanicity; lowest-value houses had higher exposure levels than expensive houses in urban regions, but the former had the lowest exposure levels in rural houses.²⁹ Given the inconsistent correlation between socioeconomic status and radon exposure by urbanicity in previous studies and our study region’s mixed distribution of urbanicity and income, which can affect indoor radon levels simultaneously, we examined the interplay between urbanicity and income with radon exposure levels. To do this, we examined potential effect modification by urbanicity and household income, comparing risks among four groups: high-income (annual household income >$75,000) in urban area; low-income (annual household income ≤$75,000) in urban area; high-income in rural area; and low-income in rural area. Effect modification was tested by the ratio of relative risks (RRR).³⁰ The RRR is calculated by exponentiating the difference of OR (d) between the two comparison groups. The standard error of the estimated difference in log ORs is calculated as $\sqrt{{SE}_1^2+{SE}_2^2}$, where ${SE}_1$ and ${SE}_2$ reflect the standard error of each log OR of the two comparison groups. When RRR and lower CI boundary (i.e., $d-1.96\times \sqrt{{SE}_1^2+{SE}_2^2}$) >1, there is a strong risk difference between the two groups.

We applied a generalized additive model (GAM) to examine potential non-linearity between radon and term low birth weight.³¹ The GAM used penalized regression splines with automatic smoothness estimation for the continuous variable of basement-level radon exposure after we incorporated all covariates included in the main logistic regression models.

The missing data for covariates were small. Ambient temperature during pregnancy had the highest missing data prevalence (1%), followed by payment source (1%). We did not perform imputation due to expected minimal impact.

Sensitivity analysis

Lower humidity during pregnancy may be a risk factor for term low birth weight.³² As a sensitivity analysis, we conducted separate logistic regressions including the mean monthly ZIP code-level dew point temperature³³ during the whole pregnancy of each participant in addition to the included covariates in the main logistic regressions. We also conducted separate logistic regressions applying a non-linear term for gestational exposure to ambient temperature, using a natural spline, instead of a linear term.

We evaluated potential unmeasured confounding by calculating an E-value, indicating the minimum strength needed for an unmeasured confounder to explain the association.³⁴

We used the Z-score, a commonly used metric of birth weight, for full-term births (Z-score = [individual weight – cohort mean]/cohort standard deviation) for particular gestational age and infant sex.³⁵ We estimated the exposure–response relationship between basement radon exposure during pregnancy and the Z-score birth weight using a natural cubic spline of radon concentration with knots fitted at the 20th, 50th, and 80th percentiles of the radon ranges. This model was the same GAM as applied to term low birth weight except that penalized splines were applied to radon exposure and term low birth weight. We added the same set of confounders in the main statistical model for term low birth weight to this model.

Results

Among the 68,159 included infants, we categorized 1,324 (2%) as term low birth weight. Of the mothers, 67% were married to the infant’s biological father at any time between conception and birth (Table 1). Exposure levels for basement radon ranged from 0.27 to 3.02 pCi/L, averaging 1.07 pCi/L (eTable 1). The range of exposure levels for ground-level radon was 0.12–2.35, with an average of 0.70 pCi/L.

Table 1.

Descriptive statistics of individual-level and community-level (ZIP code) variables (2016–2018) of study participants (n = 68,159).

	Total
Individual-level variables
Mother’s age at delivery (years) (mean ± SD)	30.8 ± 5.3
Maternal education, n (%)
8th grade or less	1,520 (2)
9–12th grade	4,306 (6)
High school completed	23,761 (35)
Bachelor’s or associate degree	23,096 (34)
Postgraduate degree	15,476 (23)
Married to the biological father at any time between conception and birth, n (%)
Yes	45,651 (67)
No	22,508 (33)
Insurance type, n (%)
Self-pay	2,736 (4)
Public	24,468 (36)
Private	40,251 (59)
Other	704 (1)
Urbanicity of residence, n (%)
Urban	20,599 (30)
Rural	47,560 (70)
Apgar five score (mean ± SD)	8.8 ± 0.6
Infant’s sex, n(%)
Male	34,398 (51)
Female	33,761 (50)
Smoking during pregnancy, n (%)
Yes	2,251 (3)
No	65,908 (97)
Alcohol use during pregnancy, n (%)
Yes	216 (0)
No	67,943 (100)
Parity, n (%)
≥1	41,958 (62)
0	26,201 (38)
Gestational diabetes, n (%)
Yes	4,621 (7)
No	63,538 (93)
Gestational hypertension, n (%)
Yes	3,687 (5)
No	64,472 (95)
Eclampsia, n (%)
Yes	135 (0)
No	68,024 (100)
Prenatal care, n (%)
Received prenatal care	68,090 (100)
No prenatal care	69 (0)
Season of birth, n (%)
Spring	17,445 (26)
Summer	18,158 (27)
Fall	17,184 (25)
Winter	15,372 (23)
ZIP code-level variables
Percent of population in poverty (mean, SD)	9 (8.0)
Percent of Black population (mean, SD)	12 (14)
Percent of single-family houses (mean, SD)	55 (24)
Annual median household income ($, mean, SD)	71,087 (30,086)

Figure 1-A shows correlations among population density, basement-level radon, and ground-level radon during pregnancy. Basement-level and ground-level radon levels were highly and positively correlated (r = 0.82). Basement-level radon exceeded ground-level radon (Figure 1-B). Levels were lower in areas with higher population density.

Figure 1.

Distribution of exposure levels to radon throughout pregnancy: (A) correlations among radon exposures and ZIP code-specific population density (2010) and (B) geographical distribution of average radon exposure levels (pCi/L) during 2016–2018 for study ZIP codes.

We examined basement-level radon by demographics (Figure 2). We observed a trend that gestational exposures for basement-level radon were higher in rural areas, higher income households, and those with higher maternal education, although the differences were not large. The highest exposure was among mothers of Non-Hispanic-White infants.

Figure 2.

Basement radon exposure levels (pCi/L) throughout pregnancy of study participants (n = 68,159) by (A) urbanicity, (B) household income, and (C) race–ethnicity of infant, and (D) maternal education levels.

In the analysis of associations, the risk of term of birth weight was not generally observed with ground-level radon, but only with basement radon. The adjusted ORs for term low birth weight in the second-highest (Q3) quartile group compared to the reference (Q1, lowest) group for basement radon during the entire pregnancy was 1.19 (95% CI: 0.99, 1.43) (Table 2). The OR in the highest (Q4) quartile group compared to the reference group during the entire pregnancy was 1.16 (95% CI: 0.97, 1.39) (Table 2). Radon exposures were moderately correlated across trimesters (r = 0.61–0.77) (eFigure 2). Table 2 shows the trimester-specific risks from the models with and without adjustments of the correlations of radon levels across trimesters. In the model without adjustments for correlations, we found increased risks of term low birth weight with increased basement radon exposure during the first trimester; the OR was 1.22 (95% CI: 1.02, 1.45) for the second-highest (Q3; 1.01–1.33 pCi/L) and 1.26 (1.05, 1.50) for the highest (Q4; 1.34–4.43 pCi/L) quartile groups of exposure, respectively, compared to the first quartile (Table 2). During the second trimester, the second-highest (Q3; 1.01–1.33 pCi/L) exposure group had a positive OR (OR = 1.21, 95% CI: 1.02, 1.45). Increased risks for the first trimester were also found in the second-highest (Q3) and highest (Q4) exposure groups for basement radon in the model adjusted for the correlation of radon across trimesters. When we adjusted for the correlation of radon across trimesters, we found a suggestion of association during the second trimester for the Q2 exposure group for basement-level radon.

Table 2.

Odds ratios (ORs) and 95% confidence intervals of term low birthweight by the quartile groups of residential radon levels and exposure windows (n = 68,159).

	OR (95% CI)
		Model without adjustment for the correlation across trimesters			Model adjusted for the correlation across trimesters
	Pregnancy	Trimester 1	Trimester 2	Trimester 3	Trimester 1	Trimester 2	Trimester 3
Basement radon levels (pCi/L) a
Q1 (<0.81)	Reference	Reference	Reference	Reference	Reference	Reference	Reference
Q2 (0.81–0.99)	1.01 (0.85, 1.21)	1.07 (0.91, 1.27)	1.07 (0.91, 1.26)	1.02 (0.87, 1.21)	1.13 (0.96, 1.34)	1.17 (0.99, 1.38)	0.89 (0.76, 1.05)
Q3 (1.00–1.31)	1.19 (0.99, 1.43)	1.22 (1.02, 1.45)	1.21 (1.02, 1.45)	1.15 (0.96, 1.37)	1.22 (1.03, 1.44)	1.12 (0.95, 1.33)	1.00 (0.85, 1.18)
Q4 (≥1.32)	1.16 (0.97, 1.39)	1.26 (1.05, 1.50)	1.16 (0.97, 1.39)	1.08 (0.90, 1.30)	1.30 (1.08, 1.55)	1.02 (0.85, 1.22)	0.98 (0.82, 1.18)
Ground-level radon levels (pCi/L) a
Q1 (<0.46)	Reference	Reference	Reference	Reference	Reference	Reference	Reference
Q2 (0.46–0.67)	1.00 (0.83, 1.20)	1.03 (0.87, 1.22)	0.97 (0.81, 1.15)	1.05 (0.88, 1.25)	1.07 (0.92, 1.25)	1.06 (0.90, 1.24)	1.02 (0.87, 1.19)
Q3 (0.68–0.87)	1.14 (0.95, 1.36)	1.09 (0.91, 1.30)	1.03 (0.86, 1.23)	1.08 (0.90, 1.29)	0.98 (0.83, 1.15)	1.02 (0.86, 1.20)	0.92 (0.78, 1.08)
Q4 (≥0.88)	0.99 (0.81, 1.21)	1.05 (0.87, 1.27)	1.07 (0.88, 1.29)	1.03 (0.85, 1.25)	1.06 (0.89, 1.27)	1.01 (0.84, 1.21)	1.00 (0.83, 1.19)

Note. Models adjusted for mother’s age, sex of infant, race/ethnicity of infant, Apgar five score, marital status, smoking during pregnancy, alcohol consumption during pregnancy, parity, the season of birth, insurance type, maternal education level, average ambient temperature exposure during pregnancy, gestational diabetes, gestational hypertension, eclampsia, prenatal care, community-level percent Black population, percent of the single-family houses, percent population in poverty, community-level median household income, and urbanicity.

^a:radon exposure levels throughout pregnancy for exposure quartile groups.

The non-linear model of basement-level radon and term low birth weight indicated a potential non-linear relationship (Figure 3). Risk of term low birth weight was the highest around 1.5 pCi/L (i.e., 85^th percentile), below which the risk increased with higher radon exposures. The risk decreased with exposure for levels above 1.5 pCi/L.

Figure 3.

Generalized additive model (GAM) plots for exposure–response relationships between radon (basement) exposure throughout pregnancy and risk of term low birthweight (n = 68,159). The grey dotted lines represent the three quartile values (0.81, 1.00, and 1.32 pCi/L) of radon exposure levels.

We compared risks among quartile groups of basement radon exposure, stratified by urbanicity and income (eFigure 3). Associations between basement-level radon and term low birth weight tended to be higher in the low-income groups than high-income groups in both urban and non-urban (e.g., semi-urban and rural) areas. Sub-group analysis for basement-level radon by income, race–ethnicity, and education separately showed that risks tended to be higher for those with low- and moderate-income, non-Hispanic Black persons, and those with lower education levels (eFigure 4). We calculated the RRR for the ORs for the second-highest exposure group across each subgroup of the SES factor. The results showed that for the second-highest (Q3) exposure level, the OR for participants living in urban areas was higher than the OR for the same exposure level of participants living in rural areas (RRR = 1.22, 95% CI: 1.02, 1.46). Also, at the second-highest exposure level, the ORs for those with low income (RRR = 1.36, 95% CI: 1.10, 1.68) and moderate income (RRR = 1.42, 95% CI: 1.15, 1.75) were higher than the OR for those with high income. There were no substantial differences for the risks at the second-highest exposure level by maternal education level, while the OR at the highest (Q4) exposure level for participants with an undergraduate or associate degree was higher than the OR at the highest level for participants with a postgraduate degree (RRR = 1.26, 95% CI: 1.02, 1.56). However, estimates for most sub-groups were uncertain, possibly due to the reduced number of participants in each stratum. The ORs for the second-highest exposure level of ground-level radon were higher for participants living in urban areas, those with low income, and those with an undergraduate or associate degree, but the estimated RRRs were suggestive (eFigure 5).

Non-linear exposure–response relationships by urbanicity and household income are shown in eFigure 6. For low-income groups, the exposure range was wider for those in rural areas. The associations tended to be higher in the low-income groups (rural-low income and urban-low income compared to rural-high income and urban-high income).

The ORs of term low birth weight for each quartile exposure group were robust to additional adjustment for the mean monthly dew point temperature during pregnancy (eTable 3). The model applying a non-linear relationship to gestational exposure to ambient temperature did not change the results of ORs of term low birth weight. The sensitivity analysis for the Z-score birth weight showed lower Z-scores at the exposure levels below the 50th percentile of basement radon concentrations during pregnancy (eFigure 7). This result was consistent with the increased risks of term low birth weight at radon exposure levels below the 85th percentile of radon concentrations (Figure 3). Despite a trend of lower Z-scores at a lower exposure level within the range below the 50th percentile exposure concentration, the difference in the Z-score was not substantial. For example, the difference between the Z-score birth weight at the 50th percentile radon concentration and the Z-score at the 25th percentile radon concentration was small (difference=−0.006, 95% CI: −0.030, 0.018).

The E-values of the estimated OR of term low birth weight during pregnancy for the third and fourth exposure quartile groups were 1.00. The E-values of the OR for the third and fourth exposure quartile groups during the first trimester were 1.16 and 1.28, indicating that a hypothetical unmeasured confounder would need to be associated with radon exposure by a risk ratio of 1.16-fold and term low birth weight by a risk ratio of 1.28-fold each. Given the prior knowledge of the environmental exposure and outcome, there may be potential unmeasured confounders for the term low birth weight risk during early pregnancy.

Discussion

This longitudinal study examined the associations between gestational radon exposure and term low birth weight using the birth certificate data of Connecticut, USA, 2016–2018. In our analysis, basement radon exposure in the first and second trimesters showed some exposure quartile groups (e.g., the second-highest [Q3] and the highest [Q4] groups) with increased term low birth weight risks, whereas risk estimates during the third trimester were not substantially different among the exposure groups. Previous research suggests that pregnant women and fetuses are more sensitive during the first trimester when placenta forms and organs and systems adapt to hormonal and cardiovascular changes.^11,36 The placenta modulates supplying oxygen and nutrients, and decreased placental weight is associated with term born low birthweight³⁷ and birth defects.³⁸

To our knowledge, there are only a few studies on radon exposure during pregnancy and adverse birth outcomes. For example, a study using the Birth Defects Registry in Texas, USA, 1999–2009, reported increased risks of cleft lip (prevalence ratio=1.16, 95% CI: 1.08, 1.26) and limb reduction defects (1.28, 95% CI: 1.06, 1.54) for 1 pCi/L increase in regional mean radon.⁹ A previous study in Massachusetts, USA, 2011–2016, estimated adverse impact of radioactive components on lower birth weight, consistent with our results. In this previous study, an interquartile range increase in the radioactive component of particulate matter during pregnancy was associated with lower mean birth weight Z-scores (−0.05, 95% CI: −0.11, −0.001).¹²

Our results for non-linear associations between basement radon and term low birth weight risks indicated the highest risk around 1.5 pCi/L, with decreasing risks for exposures above 1.5 pCi/L. This result suggests the hypothesis that lower exposure (<1.5 pCi/L) is more harmful than higher exposure for risk of term low birth weight. A few biologic hypotheses can explain the higher risk of low birth weight under lower radon exposure compared to higher exposure levels. Non-irradiated cells can suffer radiation-induced damages (e.g., DNA damage, chromosomal instability, mutation, apoptosis) through intercellular communication mechanisms.³⁹ This effect may be more important at lower exposure doses than moderate or high doses because the effectiveness of repairing DNA in cells may be limited at low radon exposure levels.^40,41 For example, experimental studies showed that the DNA repair after irradiation was lower in low-dose irradiated cells than in cells irradiated with higher doses, indicating lower efficiency of DNA repair at lower doses.⁴² Some researchers hypothesize that the efficiency of DNA repair in a single cell may increase when damage is induced in a larger number of adjacent cells due to higher-dose exposures and repair processes are activated.⁴² Yet, further evidence is needed for the impact of radiation exposure at low doses. On the other hand, a potential reason for decreasing risks above 1.5 pCi/L is that people may be more likely to apply household radon mitigation measures when radon is higher, although we used modeled radon data that did not capture all individual time-varying practices regarding radon. Also, wealthy people who are more likely to live in single-family homes may have better general health due to their higher socioeconomic status and better access to health care despite higher radon exposure. This assumption may explain the lower associations seen in our study among participants with the highest exposures in both our analyses assessing risks in exposure quartiles and by continuous exposure levels. In a previous meta-analysis of a total of 60 epidemiologic studies (1976–2013) for residential or occupational radon exposure, the exposure–response relationship curves for other health outcomes (e.g., lung cancer) showed a non-linear relationship with linearly increasing risks at low exposure and decreasing risks at moderate to high exposures.⁵² Therefore, a statistical model assuming the linear–no threshold hypothesis of the risk of radon⁴¹ may not be able to detect or fully capture the impact of radon.

Because they have more ground contact than other building parts and often have limited ventilation, basements have increased potential for higher radon concentrations compared with other parts of buildings.⁴³ For our participants, average ground-level radon (0.7 pCi/L) was lower than the average basement-level radon (1.1 pCi/L). In our analysis, basement-level radon was associated with term low birth weight, but ground-level radon was not. There were fewer ground measurements than basement measurements in our radon modeling; the number of above-ground radon measurements (32,958) used to estimate radon in New England, 2001–2020, was about 12% the number of basement radon measurements (277,868), which may lead to larger exposure errors for ground-level radon estimates.¹⁴

US EPA recommends low-cost radon reduction techniques for households (e.g., radon-proof barriers, underfloor ventilation, soil suction) when indoor ambient radon level ≥4 pCi/L, but notes that radon levels <4 pCi/L still pose potential health risks (e.g., lung cancer).²² In our study, residential basement-level radon throughout pregnancy was 0.27 to 3.02 pCi/L. This range is lower than the levels reported in EPA’s 1993 Map of Radon Zones,⁴⁴ but corresponds to ranges (0.7–3.8 pCi/L) in a previous North American study combining results from the 1990s for six states including Connecticut.⁴⁵ Our results found higher basement-level radon exposures in rural areas than in urban areas. Single-story homes tend to have higher indoor radon levels than multi-story buildings.⁴⁶ Residential radon concentrations are usually lower in urban areas as much of the population lives on higher floors, displaced from the ground.^1,47,48 Also, the underlying rock in urban areas is typically sedimentary rock, emitting less radon than other rock types.⁴⁹

In our research, residential radon exposures were higher among participants with higher income and education and among non-Hispanic White participants. More affluent people are more likely to live in single-family houses than in multi-family or high-rise buildings in our study population. Epidemiologic evidence is currently limited for potential effect modifications for radon and pregnancy. Maternal SES may modify associations between radon and health during pregnancy, potentially due to disparities in radon exposure and vulnerability to effects of radon. The few studies yield inconclusive findings on the influence of socioeconomic disparities on radon exposure. Some showed a negative income–radon correlation⁵⁰ whereas others suggested lower exposures for low-income families than higher-income families.⁵¹ One such study found no disparities in radon levels throughout the USA by education.⁵¹ Disparities in radon risks may be simultaneously associated with complex factors including geographic location, urbanicity, SES, and housing.

Our stratified analysis considering quartiles of exposure levels found potential effect modifications for the associations by urbanicity, income, and maternal education. Higher levels of income and education were associated with higher radon levels, but the association of radon with term low birth weight was higher for those with lower income or education. In particular, for non-linear associations our analysis indicated that the associations may be explained by income, which was correlated with radon exposure levels. More study is required to understand radon exposure disparities, fetal health impacts, and vulnerable subgroups.

While our models in general found increased risks in higher exposure groups, we did not observe associations between basement radon concentration during pregnancy and the Z-score birth weight for full-term births. Reduced birth weight is a relatively rare outcome. Term low birth weight, occurring infrequently (2% in our study), challenges risk prediction models due to sparse outcome distribution. We considered the binary variable of term low birth weight a more suitable metric than birth weight distribution (Z-score) to examine the effects of chronic radon exposure during pregnancy.

A strength of this study is that we examined a non-linear exposure–response relationship between residential radon and fetal growth. Our analysis utilized spatiotemporally resolved (monthly ZIP code-level) radon exposures. Another strength is the use of individual-level data rather than aggregated population data. We controlled for various health-related factors including smoking, alcohol consumption, and diseases during pregnancy. We considered both individual- and community-level SES to explore disparities in exposure–response associations by urbanicity and income. We had information on the period of residence at the address at delivery, which is uncommon in birth certificate data and helps prevent exposure misclassification from residential mobility. We estimated radon exposures for various exposure windows including gestation and trimesters.

This study has limitations. First, we did not perform radon sampling at the households of the included participants, and instead used high-resolution radon modeling data. Despite the small mean absolute error (0.46 pCi/L) for the radon estimates, the radon modeling data in our study had errors in estimates with 24% mean relative errors and relatively low correlations between the observed and predicted radon values (r² =0.47) in New England.¹⁴ Also, the radon modeling data we used had improved precision for radon concentrations in areas where there were more measurement samples available for use in modeling. Although a higher number of radon measurement samples in buildings is not an ultimate indicator of non-urban status, these are likely correlated. The measurement errors for radon exposure for our participants may be differential depending on the status of the urbanicity of study areas. Therefore, it can be assumed that some degree of uncertainty in the risk estimates may be attributable to such measurement errors in exposure. Yet, there is a considerable lack of understanding regarding the extent to which exposure measurement errors in residential radon can bias the effect estimates among epidemiologic studies. We also did not have information on time–activity information of individuals, which would affect exposure levels, although this is a common limitation of environmental epidemiologic studies. Further, we did not have information on radon mitigation systems in the study regions, which would further help investigate risk modifications. The quality or modernity of housing may influence indoor radon levels, but our study did not have information on housing conditions or construction year. Analyzing compiled data from property tax assessor records on construction years could aid future work. Despite these important limitations, our estimates are substantial improvements over earlier exposure estimates for radon. Second, our data did not have information on some critical risk factors of term low birth weight including prenatal body mass index (BMI), poor nutrition, or history of giving birth to infants with low birth weight. Also, due to the lack of individual-level information for some variables of interest, we relied on ZIP code-level indicators as a proxy for some variables (e.g., income), although we included individual-level information on educational attainment. Third, although our analysis considered smoking status during pregnancy for each participant (i.e., tobacco use vs. no tobacco use), there may be false negative misclassification for this variable (i.e., errors that a pregnant woman reports no use of tobacco even tobacco was used during pregnancy). Last, we used statistical models stratified by potential effect modifiers, which results in non-identical radon exposure levels for the reference group and hinders comparison of the degree of effect modification across different factors.

Results from this study are consistent with the hypothesis that maternal exposure to residential radon in basements increases risk of term low birth weight. Earlier gestational radon exposure in pregnancy was associated with higher risks than exposure in the later trimester. The potential non-linear exposure–response relationship may contribute to pronounced risks at lower radon levels (e.g., <1.5 pCi/L = 85th percentile). Exposure levels and the estimated risks differed by maternal SES. In particular, the risks observed at lower radon levels were more pronounced for those with lower household income. Further studies should investigate whether effects of gestational radon exposure on low fetal growth occur in other regions and populations.

Data Availability Statements

Deidentified individual participant data will not be made available. The data of modeled residential radon in the USA will be made available to researchers who provide a methodologically sound proposal for use in achieving the goals of the approved proposal. Proposals should be submitted to LL [lol087@mail.harvard.edu].