Ambient Air Pollution, Preconceptional Thyrotropin Abnormalities, and the Risk of Preterm Birth: A Nationwide Chinese Cohort Study

Abstract

Air pollution has been associated with preterm birth (PTB); however, evidence remains limited about preconception exposure. Additionally, it remains unknown whether air pollution and thyrotropin abnormalities may synergistically contribute to incident PTB. We conducted a nationwide cohort study of 6.1 million singleton births, based on the National Free Prepregnancy Checkups Project across 29 provinces in China, 2014–2020. We observed that interquartile range increases in particulate matter with diameters ≤2.5 and ≤10 μm, nitrogen dioxide, and sulfur dioxide during 12 weeks before conception through pregnancy were associated with increased risks of PTB, with the hazard ratios of 1.25 (95% confidence intervals [CI]: 1.24, 1.25), 1.13 (95% CI: 1.13, 1.14), 1.10 (95% CI: 1.09, 1.10), and 1.24 (95% CI: 1.24, 1.25), respectively. Participants with high air pollution exposure (>75th percentile) and abnormal thyrotropin levels (<0.37 or ≥4.88 mIU/L) showed higher risks of PTB, compared to those with low air pollution exposure (<25th percentile) and normal thyrotropin levels. Additive interactions were also identified between air pollution and thyrotropin abnormalities (relative excess risk due to interaction >0, attributable proportion due to interaction >0, and synergy index >1). Our study showed that exposure to air pollution during preconception through pregnancy was associated with increased risks of PTB and that air pollution and thyrotropin abnormalities may synergistically contribute to incident PTB. These findings highlight the importance of air pollution control and thyroid management.

Introduction

Preterm birth (PTB, <37 completed gestational weeks) is the primary cause of neonatal and infant morbidity and mortality worldwide, leading to a variety of lifelong health complications, such as neurodevelopmental impairments, gastrointestinal disorders, and respiratory and cardiovascular diseases. − Previous estimates showed that PTB rates have increased significantly in most countries over the past decade. Globally, PTB affects approximately 4–16% of pregnancies and accounts for an estimated 740,000 infant deaths in 2021, which imposes a substantial economic burden on health systems. Addressing the burden of preterm-related disease is essential for achieving the United Nations’ Sustainable Development Goal of ensuring healthy lives and promoting well-being for all ages. Better understanding of the key determinants of PTB is crucial for developing effective intervention and policy strategies.

An increasing number of studies have reported that prenatal exposure to air pollutants, including particulate matter with an aerodynamic diameter ≤2.5 μm (PM_2.5) and ≤10 μm (PM₁₀), nitrogen dioxide (NO₂), sulfur dioxide (SO₂), and ozone (O₃), is associated with an elevated risk of PTB. , However, most studies have focused on exposure timing during pregnancy, − with less attention given to the preconception period, especially the 12 weeks prior to conceptiona critical window for preantral follicle maturation and ovulation. Preconception exposure to air pollution may disrupt follicular development, early embryonic cell lineage segregation, hormone homeostasis, and intrauterine inflammation, all of which adversely affect fetal development. Mechanistic and epidemiological studies have revealed that preconception exposure to air pollutants can influence epigenetics and DNA methylation in germ cells , and alter placental mitochondrial DNA copy number, potentially affecting embryonic development. In addition, evidence shows that preconception exposure to air pollution may contribute to the development of gestational hypertension, gestational glucose intolerance, and gestational diabetes, thereby disrupting fetal growth. To date, few studies have investigated the association between preconceptional exposure to air pollution and PTB, and the findings remain inconsistent. ,−

Maternal thyrotropin (thyroid-stimulating hormone, TSH) is a key regulator of thyroid function and plays a critical role in fetal development and wellness during pregnancy. , Epidemiological studies have linked alterations in maternal TSH levels during preconception and pregnancy to an elevated risk of multiple adverse pregnancy outcomes, including PTB. − Our previous study among reproductive-aged women from the China National Free Prepregnancy Checkups Project (NFPCP) found that fluctuations in preconception TSH levels, even within the reference range, can significantly elevate the risk of PTB. Evidence suggests that TSH abnormalities may contribute to PTB by inducing inflammation and oxidative stress, endocrine disorders, and placental dysfunction. These pathways are common underlying mechanisms linking air pollution exposure to PTB, , which supports the biological plausibility of a synergistic effect between air pollution and TSH abnormalities. Given the evidence linking air pollution, TSH abnormalities, and PTB, the potential interactive effects of these factors on PTB remain an open question of significant interest and importance. To this extent, in this nationwide cohort study of more than 6.1 million births from the NFPCP, we examined the effects of ambient air pollution exposure from preconception through pregnancy and preconceptional TSH abnormalities on the risk of PTB. We also assessed whether air pollution and TSH abnormalities synergistically contribute to incident PTB.

Materials and Methods

Design and Study Population

This nationwide cohort study was conducted based on the NFPCP, with the design details including sampling, data acquisition, and data quality assessment previously reported. , Briefly, the NFPCP is a nationwide preconceptional healthcare service launched in 2010, covering 31 provinces in China, with the aim of providing free preconceptional health examinations and counseling, as well as early gestation and postnatal follow-ups for rural and urban reproductive-aged couples who meet the fertility policy and plan to conceive within the next 6 months. For each NFPCP participant, data were collected and archived in three main stages: preconceptional examination, early gestation follow-up, and postnatal follow-up, respectively (see the Supporting Information). We collected NFPCP database data regarding the preconceptional examination, early gestation follow-up, and postnatal follow-up for 6,542,908 Chinese women who participated in the NFPCP and successfully conceived between January 1, 2014, and April 21, 2020. A flowchart of the exclusion criteria for the study population is shown in Figure S1. We excluded participants with missing information on age, and those aged <16 years or >49 years. Participants with multiparous pregnancies (n = 48,078) were excluded. Among the remaining 6,483,804 participants with singleton pregnancy, we further excluded those missing information on preconception TSH data (n = 88,959), those with certain pregnancy outcomes, including ectopic pregnancy, spontaneous abortion, or medically induced abortion (n = 15,190), and those having prepregnancy diseases, including diabetes, hypertension, heart disease, chronic nephritis, cancer, thyroid disorder, or sexually transmitted disease (n = 36,781). Participants who resided in Xinjiang or Tibet provinces (n = 78,687) were excluded due to missing air pollutant data. Each participant’s address information during the preconception examination, early gestation follow-up, and postpartum follow-up was compared, excluding those who relocated during follow-up (n = 60,945). Participants with stillbirth (n = 9460), birth defects (n = 2742), or unclear infant sex (n = 9563) were also excluded. Additionally, we further excluded the births with an implausible gestational age (n = 39,428). Finally, a total of 6,142,049 participants were included in this study. This study was approved by the Institutional Research Review Board of the National Health Commission (IRB-201001). Written informed consents were obtained from all NFPCP participants. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.

Exposure Assessment

Daily concentrations of ambient air pollutants, including PM_2.5, PM₁₀, NO₂, and SO₂, at a 15 × 15 km spatial resolution across China from 2000 to 2020 were derived from the Nested Air Quality Prediction Model System (NAQPMS). The detailed information on NAQPMS can be found elsewhere. , Briefly, NAQPMS is a 3-D regional Eulerian chemical transport model that integrates chemical transport, emissions data, and meteorological conditions, encompassing chemical reactions, advection, diffusion, and dry and wet deposition processes. Cross-validation results indicated a strong correlation between NAQPMS estimates and monitored values, with an R ² > 0.90. We assigned the daily average concentration from the grid data set for each participant according to the longitude and latitude of their residential addresses and then calculated average concentrations of pollutants for five exposure windows: preconception (12 weeks before conception), first trimester (1–13 weeks), second trimester (14–27 weeks), third trimester (28 weeks of gestation to delivery), and preconception through pregnancy (12 weeks before conception to delivery). Daily ambient temperature (AT) and relative humidity (RH) data were obtained from 2423 weather monitoring stations of the China Meteorological Data Service Center (http://data.cma.cn/). Participants’ addresses were matched to the nearest weather monitoring station within 50 km, and average AT and RH were calculated for each participant from preconception through pregnancy.

Preconceptional TSH Measurement

During the preconception examination, fasting blood samples were collected from NFPCP participants and immediately sent to accredited laboratories affiliated with local authorized maternal and child health service centers. Serum samples were centrifuged and stored at −30 °C until analysis. TSH concentrations were measured using enzyme-linked immunosorbent assay kits or an electrochemiluminescence immunoassay in accordance with the National Guide to Clinical Laboratory Procedures. Given the vast participant pool across 2753 counties in 31 provinces, a uniform TSH detection kit or standardized detection method was not feasible; however, samples from the same county were measured using the same TSH assay. In our previous study, a population-specific TSH reference range has been established. Preconception TSH levels were classified into subnormal (<0.37 mIU/L), normal (0.37–4.87 mIU/L), and supranormal levels (≥4.88 mIU/L) based on the established reference range. In addition, we calculated multiples of the median (MoM) of TSH levels by dividing each participant’s TSH value by the median value of TSH in the reference population of the county in which they participated in the NFPCP. The TSH MoM value served as an alternative indicator to absolute TSH measurement, mitigating the variability and systematic differences among various assays and laboratories.

Outcome Definition

Gestational age was estimated based on the time since the first day of the last menstrual period. PTB was defined as a gestational age from 20 to 36 completed weeks. We further classified PTB into three categories according to the criteria recommended by the World Health Organization: moderate PTB (a gestational age from 32 through 36 completed weeks), very PTB (a gestational age from 28 through 31 completed weeks), and extremely PTB (20 through 27 completed weeks).

Covariates

According to the prior knowledge and existing literature, ,, the following covariates were selected: age, preconception body mass index (BMI), ethnicity, education level, residence, occupation, alcohol drinking, smoking, passive smoking, psychological pressure, infant sex, season of conception, parity, supplementation of folic acid, geographical region, and average levels of AT (°C) and RH (%) from preconception through pregnancy. To control the potential long-term trends, the year of birth (2014, 2015, 2016, 2017, 2018, 2019, and 2020) was included as a categorical covariate. Since all covariables had missing data rates below 7.0%, the missing indicator method was used to handle these missing values, and missing indicators were included for all categorical adjustment variables.

Statistical Analysis

Descriptive statistics were calculated for participants’ characteristics and outcomes. Spearman correlation analysis was used to examine the correlation among air pollution, AT, and RH.

We applied a Cox frailty model with random effects to account for clustering by counties to estimate the associations of air pollutants during different exposure windows and PTB. Gestational age was fitted as the underlying time scale, with PTB defined as a binary outcome. A directed acyclic graph was employed to identify the minimal sufficient adjustment set in the model: maternal age, preconceptional BMI, education, occupation, ethnicity, folic acid supplementation, geographic region, season of conception, year of birth, AT, and RH (Figure S2). Based on the Akaike Information Criterion (AIC), AT and RH were modeled using natural cubic splines with degrees of freedom (df) from 3 to 10 and the df for minimum AIC was determined as 6 for AT and 7 for RH. Effect estimates are presented as hazard ratios (HRs) and 95% confidence intervals (CIs) for per interquartile range (IQR) increase in each pollutant (i.e., 33.1 μg/m³ for PM_2.5, 48.5 μg/m³ for PM₁₀, 22.1 μg/m³ for NO₂, and 16.2 μg/m³ for SO₂).

To assess the possible interactive effects between preconceptional TSH and air pollutants on PTB, air pollution exposures over preconception through pregnancy were classified into low (<25th percentile), intermediate (25th–75th percentile), and high levels (>75th percentile) according to quartiles. A dummy variable with 3 × 3 levels was generated to represent the combination of exposure to air pollutants and TSH levels. Additive interaction was evaluated using three measures: the relative excess risk due to interaction (RERI), the attributable proportion due to interaction (AP), and the synergy index (SI), calculated as follows

$$\mathrm{R}\mathrm{E}\mathrm{R}\mathrm{I}={\mathrm{H}\mathrm{R}}_{ii}-{\mathrm{H}\mathrm{R}}_{i0}-{\mathrm{H}\mathrm{R}}_{0i}+1$$

$$\mathrm{A}\mathrm{P}=\frac{\mathrm{R}\mathrm{E}\mathrm{R}\mathrm{I}}{{\mathrm{H}\mathrm{R}}_{ii}}$$

$$\mathrm{S}\mathrm{I}=\frac{{\mathrm{H}\mathrm{R}}_{ii}-1}{({\mathrm{H}\mathrm{R}}_{0i}-1)+({\mathrm{H}\mathrm{R}}_{i0}-1)}$$

where HR _i0, HR_0i , and HR _ii refer to the HR of the i category of air pollution and TSH in comparison with the reference group (low pollution and normal TSH level). RERI > 0, AP > 0, and SI > 1 denote a synergistic interaction, whereas RERI < 0, AP < 0, and SI < 1 indicate an antagonistic interaction. The corresponding 95% CIs of the RERI, AP, and SI were estimated using the delta method.

To quantify the disease burden of PTB due to exposure to air pollution over preconception through pregnancy and TSH abnormalities, we further estimated the population-attributable fraction (PAF) and attributable number (AN) of PTB based on air pollution and TSH categories using the following equations

$$\mathrm{P}\mathrm{A}\mathrm{F}=\frac{\sum _i^{N}1-{\mathrm{e}}^{-\beta \times {\alpha }_i}}{N}\times 100\%$$

$$\mathrm{A}\mathrm{N}\mathrm{o}\mathrm{f}\mathrm{P}\mathrm{T}\mathrm{B}=\mathrm{P}\mathrm{A}\mathrm{F}\times N$$

where β represents the effect estimate of air pollutant exposure (intermediate or high levels) or TSH abnormalities, or their combinations in the Cox frailty model; e represents the nature logarithm; α indicates whether exposure to intermediate or high levels of air pollutants, or TSH abnormalities (0 = no, 1 = yes); and N represents the total number of PTB cases.

A series of sensitivity analyses were performed to examine the robustness of our main analyses: (a) we fitted two-pollutant models by simultaneously adding each of the other air pollutants in the models to test the possible confounding from coexposure of other pollutants. Due to the high correlation and collinearity between PM_2.5 and PM₁₀, these two parameters were not mutually adjusted in the two-pollutant analyses. (b) Multiple imputation using the chained equation polytomous logistic regression method, with 1000 imputed samples, was employed to handle missing data on covariates including BMI, education, occupation, ethnicity, residence, smoking, passive smoking, drinking, psychological pressure, and folic acid supplementation (Table S1). The imputation model included primary outcomes and all covariates previously described. All variables were iteratively used as auxiliary variables to predict missing values in the others. We generated 5 data sets with imputed missing values. The association between air pollution and PTB was estimated separately in each data set using the Cox frailty model, and the results were then pooled according to Rubin’s rules to obtain overall estimates and account for the uncertainty due to missing data. (c) The associations were reanalyzed by excluding missing data. (d) Analyses were restricted to participants with gestational weeks <42 weeks. (e) The mode of delivery was further adjusted in the models. (f) We applied distributed lag models incorporating the Cox frailty model to further estimate the exposure–lag–response associations between weekly specific exposure of air pollution with PTB. The exposure–response relationships were modeled using a linear function, and the lag–response relationships were modeled using natural cubic splines with dfs selected to minimize AIC. (g) A restricted cubic spline model with four knots was used to assess the dose–response relationship of preconceptional TSH levels or TSH MoM values and the risk of PTB. (h) TSH abnormalities were defined using a clinic reference range (0.75–5.60 mIU/L) recommended by the National Health Commission of China. (i) To examine the potential effects of live birth bias, we fitted Fine–Gray subdistribution hazard models that accounted for the competing risk of pregnancy losses using the GFR function in the riskRegression package (version 2023.12.21). In the Fine–Gray models, PTB was defined as the primary outcome (binary), with pregnancy loss cases defined as the competing risk, adjusted for the same covariates as those in the Cox frailty models. The associations between air pollution and PTB were expressed as subdistribution HRs and corresponding 95% CIs. (j) Given the evidence linking iodine intake to TSH levels and PTB risk, we further adjusted for iodine intake (adequate versus excess) at the area level in the Cox frailty model.

Statistical analyses were performed in R software (version 4.0.2). Geostatistical analyses were performed in the ArcGIS platform (version 10.08).

Results

Study Participants and Environmental Exposures

A total of 6,142,049 birth records were included in this study, 588,208 (9.58%) defined as PTB, including 440,596 moderate PTB (7.17%), 89,504 very PTB (1.46%), and 58,108 extremely PTB (0.95%). The population characteristics by air pollution exposure levels are shown in Table . For PM_2.5 and PM₁₀, participants with higher exposure levels were more likely to have a longer gestational age, be younger than 30 years, be overweight before pregnancy, have a junior high school education or below, be of Han Chinese ethnicity, live in rural areas, work as farmers, smoke or consume alcohol, experience psychological pressure, have standardized folic acid supplementation, conceive during spring or winter, deliver a female infant, reside in eastern, northern, or central regions of China. For NO₂ and SO₂, participants with higher exposure levels were more likely to be overweight, be multiparous, be of Han Chinese ethnicity, work as farmers, smoke, experience psychological pressure, have standardized folic acid supplementation, conceive during spring or winter, undergo cesarean delivery, and reside in northern regions of China.

1. Characteristic of the Study Population in China, 2014–2020 .

	PM2.5 (μg/m3)			PM10 (μg/m3)			NO2 (μg/m3)			SO2 (μg/m3)
characteristics	low (<36.1)	intermediate (36.1–69.3)	high (>69.3)	low (<52.2)	intermediate (52.2–100.7)	high (>100.7)	low (<18.5)	intermediate (18.5–40.6)	high (>40.6)	low (<11.4)	intermediate (11.4–27.5)	high (>27.5)
number	1,534,805	3,072,437	1,534,807	1,534,805	3,072,439	1,534,805	1,534,805	3,072,439	1,534,805	1,534,805	3,072,439	1,534,805
gestational age, mean (SD) (week)	39.2 (2.4)	39.1 (2.7)	39.4 (2.5)	39.14 (2.52)	39.14 (2.68)	39.48 (2.31)	39.11 (2.71)	39.26 (2.52)	39.26 (2.47)	39.20 (2.39)	39.24 (2.56)	39.22 (2.70)
preconceptional TSH level, median (IQR) (mIU/L)	1.60 (1.31)	1.62 (1.30)	1.64 (1.29)	1.57 (1.30)	1.63 (1.31)	1.65 (1.28)	1.54 (1.29)	1.63 (1.31)	1.67 (1.32)	1.61 (1.30)	1.60 (1.29)	1.64 (1.31)
<0.37	71,547 (4.7)	102,879 (3.3)	38,988 (2.5)	77,847 (5.1)	97,824 (3.2)	37,743 (2.5)	73,748 (4.8)	95,955 (3.1)	43,711 (2.8)	64,939 (4.2)	102,868 (3.3)	45,607 (3.0)
0.37–4.87	1,422,192 (92.7)	2,883,724 (93.9)	1,460,872 (95.2)	1,417,896 (92.4)	2,889,718 (94.1)	1,459,174 (95.1)	1,421,916 (92.6)	2,895,980 (94.3)	1,448,892 (94.4)	1,426,032 (92.9)	2,891,179 (94.1)	1,449,577 (94.4)
≥4.88	41,066 (2.7)	85,834 (2.8)	34,947 (2.3)	39,062 (2.5)	84,897 (2.8)	37,888 (2.5)	39,141 (2.6)	80,504 (2.6)	42,202 (2.7)	43,834 (2.9)	78,392 (2.6)	39,621 (2.6)
age at last menstrual period, year
16–24	533,438 (34.8)	1,118,167 (36.4)	625,575 (40.8)	564,545 (36.8)	1,151,681 (37.5)	560,954 (36.5)	629,384 (41.0)	1,125,489 (36.6)	522,307 (34.0)	515,086 (33.6)	1,173,952 (38.2)	588,142 (38.3)
25–29	655,134 (42.7)	1,322,621 (43.0)	652,466 (42.5)	643,147 (41.9)	1,309,363 (42.6)	677,711 (44.2)	631,713 (41.2)	1,317,889 (42.9)	680,619 (44.3)	666,202 (43.4)	1,300,398 (42.3)	663,621 (43.2)
30–49	346,233 (22.6)	631,649 (20.6)	256,766 (16.7)	327,113 (21.3)	611,395 (19.9)	296,140 (19.3)	273,708 (17.8)	629,061 (20.5)	331,879 (21.6)	353,517 (23.0)	598,089 (19.5)	283,042 (18.4)
parity
0	801,560 (52.2)	1,504,231 (49.0)	829,952 (54.1)	800,313 (52.1)	1,549,902 (50.4)	785,528 (51.2)	812,258 (52.9)	1,559,975 (50.8)	763,510 (49.7)	785,898 (51.2)	1,571,731 (51.2)	778,114 (50.7)
≥1	616,509 (40.2)	1,351,044 (44.0)	639,889 (41.7)	615,311 (40.1)	1,304,320 (42.5)	687,811 (44.8)	611,669 (39.9)	1,315,910 (42.8)	679,863 (44.3)	633,798 (41.3)	1,299,465 (42.3)	674,179 (43.9)
missing data	116,736 (7.6)	217,162 (7.1)	64,966 (4.2)	119,181 (7.8)	218,217 (7.1)	61,466 (4.0)	110,878 (7.2)	196,554 (6.4)	91,432 (6.0)	115,109 (7.5)	201,243 (6.5)	82,512 (5.4)
prepregnancy BMI, kg/m2
≤18.5	271,695 (17.7)	401,801 (13.1)	147,422 (9.6)	272,841 (17.8)	410,350 (13.4)	137,727 (9.0)	250,098 (16.3)	395,011 (12.9)	175,809 (11.5)	253,566 (16.5)	405,751 (13.2)	161,601 (10.5)
18.6–23.9	1,050,306 (68.4)	2,144,972 (69.8)	1,095,969 (71.4)	1,054,424 (68.7)	2,158,588 (70.3)	1,078,235 (70.3)	1,083,441 (70.6)	2,153,473 (70.1)	1,054,333 (68.7)	1,053,184 (68.6)	2,167,224 (70.5)	1,070,839 (69.8)
≥24	203,332 (13.2)	512,746 (16.7)	289,565 (18.9)	199,318 (13.0)	488,972 (15.9)	317,353 (20.7)	199,378 (13.0)	509,300 (16.6)	296,965 (19.3)	218,053 (14.2)	488,182 (15.9)	299,408 (19.5)
missing data	9472 (0.6)	12,918 (0.4)	1851 (0.1)	8222 (0.5)	14,529 (0.5)	1490 (0.1)	1888 (0.1)	14,655 (0.5)	7698 (0.5)	10,002 (0.7)	11,282 (0.4)	2957 (0.2)
education
junior high school or below	672,432 (43.8)	1,628,385 (53.0)	1,012,462 (66.0)	725,492 (47.3)	1,623,584 (52.8)	964,203 (62.8)	816,964 (53.2)	1,668,813 (54.3)	827,502 (53.9)	673,558 (43.9)	1,690,099 (55.0)	949,622 (61.9)
senior high school	330,832 (21.6)	695,922 (22.7)	291,341 (19.0)	327,143 (21.3)	688,467 (22.4)	302,485 (19.7)	349,731 (22.8)	673,619 (21.9)	294,745 (19.2)	338,392 (22.0)	669,286 (21.8)	310,417 (20.2)
college or above	426,533 (27.8)	606,446 (19.7)	178,056 (11.6)	386,642 (25.2)	613,841 (20.0)	210,552 (13.7)	303,871 (19.8)	583,505 (19.0)	323,659 (21.1)	412,161 (26.9)	568,230 (18.5)	230,644 (15.0)
missing data	105,008 (6.8)	141,684 (4.6)	52,948 (3.4)	95,528 (6.2)	146,547 (4.8)	57,565 (3.8)	64,239 (4.2)	146,502 (4.8)	88,899 (5.8)	110,694 (7.2)	144,824 (4.7)	44,122 (2.9)
ethnicity
Han Chinese	1,342,612 (87.5)	2,857,489 (93.0)	1,491,766 (97.2)	1,319,671 (86.0)	2,889,928 (94.1)	1,482,268 (96.6)	1,313,266 (85.6)	2,906,043 (94.6)	1,472,558 (95.9)	1,375,529 (89.6)	2,845,420 (92.6)	1,470,918 (95.8)
ethnic minority	154,551 (10.1)	149,379 (4.9)	15,786 (1.0)	179,409 (11.7)	116,701 (3.8)	23,606 (1.5)	196,616 (12.8)	97,984 (3.2)	25,116 (1.6)	116,247 (7.6)	164,154 (5.3)	39,315 (2.6)
missing data	37,642 (2.5)	65,569 (2.1)	27,255 (1.8)	35,725 (2.3)	65,810 (2.1)	28,931 (1.9)	24,923 (1.6)	68,412 (2.2)	37,131 (2.4)	43,029 (2.8)	62,865 (2.0)	24,572 (1.6)
residence
rural	1,259,864 (82.1)	2,769,367 (90.1)	1,477,643 (96.3)	1,294,568 (84.3)	2,758,892 (89.8)	1,453,414 (94.7)	1,382,807 (90.1)	2,778,797 (90.4)	1,345,270 (87.7)	1,275,106 (83.1)	2,791,597 (90.9)	1,440,171 (93.8)
urban	274,484 (17.9)	302,465 (9.8)	56,771 (3.7)	240,075 (15.6)	312,558 (10.2)	81,087 (5.3)	151,994 (9.9)	293,403 (9.5)	188,323 (12.3)	259,302 (16.9)	280,285 (9.1)	94,133 (6.1)
missing data	457 (0.0)	605 (0.0)	393 (0.0)	162 (0.0)	989 (0.0)	304 (0.0)	4 (0.0)	239 (0.0)	1212 (0.1)	397 (0.0)	557 (0.0)	501 (0.0)
occupation
farmer	663,351 (43.2)	1,955,471 (63.6)	1,251,405 (81.5)	715,709 (46.6)	1,938,623 (63.1)	1,215,895 (79.2)	907,049 (59.1)	1,959,647 (63.8)	1,003,531 (65.4)	684,834 (44.6)	1,996,314 (65.0)	1,189,079 (77.5)
worker	255,314 (16.6)	348,208 (11.3)	101,060 (6.6)	246,800 (16.1)	351,822 (11.5)	105,960 (6.9)	204,577 (13.3)	354,919 (11.6)	145,086 (9.5)	243,474 (15.9)	340,593 (11.1)	120,515 (7.9)
other	486,432 (31.7)	604,016 (19.7)	128,539 (8.4)	449,739 (29.3)	615,624 (20.0)	153,624 (10.0)	338,554 (22.1)	591,900 (19.3)	288,533 (18.8)	471,674 (30.7)	570,073 (18.6)	177,240 (11.5)
missing data	129,708 (8.5)	164,742 (5.4)	53,803 (3.5)	122,557 (8.0)	166,370 (5.4)	59,326 (3.9)	84,625 (5.5)	165,973 (5.4)	97,655 (6.4)	134,823 (8.8)	165,459 (5.4)	47,971 (3.1)
smoking
no	1,484,924 (96.8)	3,008,622 (97.9)	1,508,166 (98.3)	1,489,441 (97.0)	3,005,700 (97.8)	1,506,571 (98.2)	1,499,823 (97.7)	2,999,445 (97.6)	1,502,444 (97.9)	1,486,561 (96.9)	3,002,980 (97.7)	1,512,171 (98.5)
yes	10,717 (0.7)	16,434 (0.5)	4445 (0.3)	10,619 (0.7)	15,802 (0.5)	5175 (0.3)	9030 (0.6)	16,225 (0.5)	6341 (0.4)	11,548 (0.8)	15,640 (0.5)	4408 (0.3)
missing data	39,164 (2.6)	47,381 (1.5)	22,196 (1.4)	34,745 (2.3)	50,937 (1.7)	23,059 (1.5)	25,952 (1.7)	56,769 (1.8)	26,020 (1.7)	36,696 (2.4)	53,819 (1.8)	18,226 (1.2)
passive smoking
no	1,521,618 (99.1)	3,045,475 (99.1)	1,525,414 (99.4)	1,521,336 (99.1)	3,045,453 (99.1)	1,525,718 (99.4)	1,521,684 (99.1)	3,048,098 (99.2)	1,522,725 (99.2)	1,521,822 (99.2)	3,046,713 (99.2)	1,523,972 (99.3)
yes	13,187 (0.9)	26,962 (0.9)	9393 (0.6)	13,469 (0.9)	26,986 (0.9)	9087 (0.6)	13,121 (0.9)	24,341 (0.8)	12,080 (0.8)	12,983 (0.8)	25,726 (0.8)	10,833 (0.7)
drinking
no	1,471,735 (95.9)	3,004,380 (97.8)	1,507,818 (98.2)	1,478,354 (96.3)	2,999,716 (97.6)	1,505,863 (98.1)	1,495,945 (97.5)	2,991,016 (97.3)	1,496,972 (97.5)	1,474,447 (96.1)	2,997,911 (97.6)	1,511,575 (98.5)
yes	22,583 (1.5)	18,756 (0.6)	3847 (0.3)	20,381 (1.3)	19,927 (0.6)	4878 (0.3)	11,748 (0.8)	22,700 (0.7)	10,738 (0.7)	22,354 (1.5)	18,785 (0.6)	4047 (0.3)
missing data	40,487 (2.6)	49,301 (1.6)	23,142 (1.5)	36,070 (2.4)	52,796 (1.7)	24,064 (1.6)	27,112 (1.8)	58,723 (1.9)	27,095 (1.8)	38,004 (2.5)	55,743 (1.8)	19,183 (1.2)
psychological pressure
no	1,123,090 (73.2)	2,578,440 (83.9)	1,371,746 (89.4)	1,128,952 (73.6)	2,549,083 (83.0)	1,395,241 (90.9)	1,218,434 (79.4)	2,564,372 (83.5)	1,290,470 (84.1)	1,151,143 (75.0)	2,564,045 (83.5)	1,358,088 (88.5)
yes	388,944 (25.3)	471,975 (15.4)	157,934 (10.3)	386,202 (25.2)	498,059 (16.2)	134,592 (8.8)	308,487 (20.1)	483,533 (15.7)	226,833 (14.8)	362,333 (23.6)	486,079 (15.8)	170,441 (11.1)
missing data	22,771 (1.5)	22,022 (0.7)	5127 (0.3)	19,651 (1.3)	25,297 (0.8)	4972 (0.3)	7884 (0.5)	24,534 (0.8)	17,502 (1.1)	21,329 (1.4)	22,315 (0.7)	6276 (0.4)
folic acid supplementation
standardized use	679,091 (44.2)	1,349,841 (43.9)	717,989 (46.8)	625,622 (40.8)	1,330,920 (43.3)	790,379 (51.5)	637,804 (41.6)	1,363,821 (44.4)	745,296 (48.6)	686,673 (44.7)	1,322,721 (43.1)	737,527 (48.1)
irregular or no use	823,071 (53.6)	1,680,897 (54.7)	794,763 (51.8)	879,460 (57.3)	1,696,017 (55.2)	723,254 (47.1)	875,764 (57.1)	1,656,529 (53.9)	766,438 (49.9)	817,870 (53.3)	1,701,482 (55.4)	779,379 (50.8)
missing data	32,643 (2.1)	41,699 (1.4)	22,055 (1.4)	29,723 (1.9)	45,502 (1.5)	21,172 (1.4)	21,237 (1.4)	52,089 (1.7)	23,071 (1.5)	30,262 (2.0)	48,236 (1.6)	17,899 (1.2)
season of conception
spring	396,424 (25.8)	876,395 (28.5)	559,657 (36.5)	411,607 (26.8)	867,941 (28.2)	552,928 (36.0)	415,060 (27.0)	914,272 (29.8)	503,144 (32.8)	410,763 (26.8)	910,815 (29.6)	510,898 (33.3)
summer	451,731 (29.4)	808,525 (26.3)	228,206 (14.9)	432,196 (28.2)	761,325 (24.8)	294,941 (19.2)	465,217 (30.3)	783,639 (25.5)	239,606 (15.6)	450,610 (29.4)	771,777 (25.1)	266,075 (17.3)
autumn	387,807 (25.3)	652,912 (21.3)	193,652 (12.6)	365,133 (23.8)	662,743 (21.6)	206,495 (13.5)	370,327 (24.1)	644,521 (21.0)	219,523 (14.3)	377,347 (24.6)	630,584 (20.5)	226,440 (14.8)
winter	298,843 (19.5)	734,605 (23.9)	553,292 (36.0)	325,869 (21.2)	780,430 (25.4)	480,441 (31.3)	284,201 (18.5)	730,007 (23.8)	572,532 (37.3)	296,085 (19.3)	759,263 (24.7)	531,392 (34.6)
year of birth
2014	28,956 (1.9)	118,700 (3.9)	194,489 (12.7)	60,204 (3.9)	160,825 (5.2)	121,116 (7.9)	55,396 (3.6)	120,342 (3.9)	166,407 (10.8)	20,886 (1.4)	117,077 (3.8)	204,182 (13.3)
2015	227,535 (14.8)	659,651 (21.5)	466,169 (30.4)	282,503 (18.4)	678,700 (22.1)	392,152 (25.6)	367,455 (23.9)	643,082 (20.9)	342,818 (22.3)	192,554 (12.5)	668,636 (21.8)	492,165 (32.1)
2016	323,816 (21.1)	669,357 (21.8)	419,331 (27.3)	328,082 (21.4)	663,328 (21.6)	421,094 (27.4)	382,361 (24.9)	688,987 (22.4)	341,156 (22.2)	267,105 (17.4)	699,343 (22.8)	446,056 (29.1)
2017	431,749 (28.1)	755,907 (24.6)	300,495 (19.6)	423,582 (27.6)	732,163 (23.8)	332,406 (21.7)	406,292 (26.5)	719,277 (23.4)	362,582 (23.6)	412,027 (26.8)	804,741 (26.2)	271,383 (17.7)
2018	316,391 (20.6)	553,126 (18.0)	106,484 (6.9)	272,938 (17.8)	510,921 (16.6)	192,142 (12.5)	208,724 (13.6)	538,442 (17.5)	228,835 (14.9)	348,194 (22.7)	533,089 (17.4)	94,718 (6.2)
2019	187,971 (12.2)	287,909 (9.4)	44,505 (2.9)	153,634 (10.0)	294,205 (9.6)	72,546 (4.7)	104,496 (6.8)	330,535 (10.8)	85,354 (5.6)	264,156 (17.2)	232,336 (7.6)	23,893 (1.6)
2020	18,387 (1.2)	27,787 (0.9)	3334 (0.2)	13,862 (0.9)	32,297 (1.1)	3349 (0.2)	10,081 (0.7)	31,774 (1.0)	7653 (0.5)	29,883 (1.9)	17,217 (0.6)	2408 (0.2)
neonate’s sex
male	803,110 (52.3)	1,595,022 (51.9)	791,418 (51.6)	802,195 (52.3)	1,594,718 (51.9)	792,637 (51.6)	801,746 (52.2)	1,591,777 (51.8)	796,027 (51.9)	801,512 (52.2)	1,594,502 (51.9)	793,536 (51.7)
female	731,695 (47.7)	1,477,415 (48.1)	743,389 (48.4)	732,610 (47.7)	1,477,721 (48.1)	742,168 (48.4)	733,059 (47.8)	1,480,662 (48.2)	738,778 (48.1)	733,293 (47.8)	1,477,937 (48.1)	741,269 (48.3)
mode of delivery
vaginal	1,161,378 (75.7)	2,150,004 (70.0)	1,111,041 (72.4)	1,164,838 (75.9)	2,162,877 (70.4)	1,094,708 (71.3)	1,190,187 (77.5)	2,207,589 (71.9)	1,024,647 (66.8)	1,109,548 (72.3)	2,248,903 (73.2)	1,063,972 (69.3)
cesarean	373,159 (24.3)	922,222 (30.0)	423,734 (27.6)	369,714 (24.1)	909,335 (29.6)	440,066 (28.7)	344,469 (22.4)	864,627 (28.1)	510,019 (33.2)	424,964 (27.7)	823,340 (26.8)	470,811 (30.7)
other	268 (0.0)	211 (0.0)	32 (0.0)	253 (0.0)	227 (0.0)	31 (0.0)	149 (0.0)	223 (0.0)	139 (0.0)	293 (0.0)	196 (0.0)	22 (0.0)
geographical region
eastern	326,780 (21.3)	902,334 (29.4)	329,283 (21.5)	295,504 (19.3)	911,380 (29.7)	351,513 (22.9)	195,203 (12.7)	887,662 (28.9)	475,532 (31.0)	387,131 (25.2)	753,263 (24.5)	418,003 (27.2)
northeast	49,485 (3.2)	24,213 (0.8)	823 (0.1)	47,650 (3.1)	25,417 (0.8)	1454 (0.1)	33,590 (2.2)	36,996 (1.2)	3935 (0.3)	36,700 (2.4)	30,437 (1.0)	7384 (0.5)
northern	42,082 (2.7)	111,317 (3.6)	217,240 (14.2)	32,285 (2.1)	117,656 (3.8)	220,698 (14.4)	33,208 (2.2)	79,391 (2.6)	258,040 (16.8)	44,296 (2.9)	92,739 (3.0)	233,604 (15.2)
northwest	215,869 (14.1)	164,869 (5.4)	10,139 (0.7)	125,170 (8.2)	203,468 (6.6)	62,239 (4.1)	186,199 (12.1)	164,024 (5.3)	40,654 (2.6)	140,650 (9.2)	183,847 (6.0)	66,380 (4.3)
southeast	172,482 (11.2)	281,165 (9.2)	24,341 (1.6)	221,764 (14.4)	237,277 (7.7)	18,947 (1.2)	247,743 (16.1)	183,310 (6.0)	46,935 (3.1)	172,929 (11.3)	267,705 (8.7)	37,354 (2.4)
southern	687,902 (44.8)	260,552 (8.5)	943 (0.1)	736,110 (48.0)	213,144 (6.9)	143 (0.0)	502,291 (32.7)	310,783 (10.1)	136,323 (8.9)	549,303 (35.8)	381,838 (12.4)	18,256 (1.2)
central	40,205 (2.6)	1,327,987 (43.2)	952,038 (62.0)	76,322 (5.0)	1,364,097 (44.4)	879,811 (57.3)	336,571 (21.9)	1,410,273 (45.9)	573,386 (37.4)	203,796 (13.3)	1,362,610 (44.3)	753,824 (49.1)

^aThe mean and standard deviation are shown for continuous variables; the number and percentage are shown for categorical variables.

^bAbbreviations: PM_2.5, particulate matter with aerodynamic diameters ≤2.5 μm; PM₁₀, particulate matter with aerodynamic diameters ≤10 μm; NO₂, nitrogen dioxide; SO₂, sulfur dioxide; SD, standard deviation; TSH, thyroid-stimulating hormone; MoM, multiples of the median; BMI, body mass index; SD, standard deviation; IQR, interquartile range.

The median concentrations of PM_2.5, PM₁₀, NO₂, and SO₂ during preconception through pregnancy were 54.7 μg/m³ (IQR = 33.1 μg/m³), 78.5 μg/m³ (IQR = 48.5 μg/m³), 30.2 μg/m³ (IQR = 22.1 μg/m³), and 21.6 μg/m³ (IQR = 16.2 μg/m³), respectively (Table S2). Participants living in the Beijing–Tianjin–Hebei and Henan–Shandong regions were exposed to relatively high levels of air pollution during the study period (Figure S3). As shown in Table S3, we observed a strong correlation between PM_2.5 and PM₁₀ (r = 0.89). Both PM_2.5 and PM₁₀ were positively and moderately correlated with NO₂ and SO₂ (the Spearman correlation coefficient ranged from 0.64 to 0.73).

Associations between Air Pollution Exposure and PTB

Table shows the HRs of PTB and its subtypes associated with air pollution during different exposure windows. We observed that exposure to PM_2.5, PM₁₀, NO₂, and SO₂ from preconception through pregnancy was associated with increased risks of PTB and its subtypes. Per IQR increase in PM_2.5, PM₁₀, NO₂, and SO₂ during preconception through pregnancy was associated with increased risks of PTB, with HRs of 1.25 (95% CI: 1.24, 1.25), 1.13 (95% CI: 1.13, 1.14), 1.10 (95% CI: 1.09, 1.10), and 1.24 (95% CI: 1.24, 1.25), respectively. Significant associations were also found for moderate PTB, very PTB, and extremely PTB, and the HRs ranged from 1.18 (95% CI: 1.14–1.21) to 1.74 (95% CI: 1.66–1.83). The largest estimates were observed for extremely PTB across different exposure windows and air pollutants. As for exposure to PM_2.5 and PM₁₀, the stronger associations were observed during each trimester of pregnancy, when compared with the preconception period of exposure. In contrast, as for NO₂ and SO₂, preconception exposure showed higher effect estimates than those derived from the second and third trimester exposure. Similar associations were also observed when we fitted the model with the categories of air pollutant exposure (Table S4). The HRs of PTB and its subtypes in groups with high air pollutant exposure were significantly higher than those with low or intermediate air pollutant exposure.

2. Hazard Ratios of PTB and Its Subtypes Associated with Each IQR Increase in Pollutant Concentrations during Different Exposure Windows in China, 2014–2020 (N = 6,142,049) .

		hazard ratios (95% CI)
air pollutant	period	PTB	moderate PTB	very PTB	extremely PTB
PM2.5	preconception	1.22 (1.21, 1.22)	1.21 (1.20, 1.21)	1.36 (1.34, 1.38)	1.43 (1.40, 1.45)
	first trimester	1.49 (1.48, 1.49)	1.40 (1.39, 1.41)	1.81 (1.78, 1.84)	1.96 (1.93, 2.00)
	second trimester	1.43 (1.42, 1.44)	1.37 (1.36, 1.38)	1.67 (1.65, 1.69)	1.61 (1.58, 1.64)
	third trimester	1.34 (1.34, 1.35)	1.31 (1.30, 1.32)	1.54 (1.52, 1.56)	1.54 (1.51, 1.57)
	preconception through pregnancy	1.25 (1.24, 1.25)	1.23 (1.22, 1.24)	1.41 (1.39, 1.43)	1.49 (1.46, 1.52)
PM10	preconception	1.12 (1.11, 1.12)	1.10 (1.10, 1.11)	1.24 (1.23, 1.26)	1.31 (1.29, 1.33)
	first trimester	1.29 (1.28, 1.29)	1.22 (1.22, 1.23)	1.52 (1.5, 1.54)	1.67 (1.64, 1.70)
	second trimester	1.27 (1.26, 1.27)	1.23 (1.22, 1.23)	1.44 (1.42, 1.46)	1.41 (1.39, 1.44)
	third trimester	1.15 (1.14, 1.16)	1.14 (1.14, 1.15)	1.25 (1.23, 1.27)	1.25 (1.22, 1.27)
	preconception through pregnancy	1.13 (1.13, 1.14)	1.12 (1.11, 1.12)	1.27 (1.25, 1.29)	1.34 (1.32, 1.37)
NO2	preconception	1.09 (1.09, 1.10)	1.10 (1.09, 1.11)	1.12 (1.10, 1.13)	1.12 (1.10, 1.14)
	first trimester	1.12 (1.11, 1.12)	1.11 (1.10, 1.11)	1.18 (1.17, 1.20)	1.21 (1.19, 1.23)
	second trimester	1.19 (1.18, 1.20)	1.15 (1.15, 1.16)	1.32 (1.30, 1.34)	1.26 (1.24, 1.28)
	third trimester	1.05 (1.05, 1.06)	1.05 (1.05, 1.06)	1.06 (1.05, 1.08)	1.04 (1.03, 1.06)
	preconception through pregnancy	1.10 (1.09, 1.10)	1.10 (1.10, 1.11)	1.13 (1.11, 1.14)	1.13 (1.11, 1.15)
SO2	preconception	1.23 (1.23, 1.23)	1.20 (1.19, 1.20)	1.30 (1.29, 1.31)	1.38 (1.37, 1.39)
	first trimester	1.26 (1.26, 1.27)	1.21 (1.21, 1.22)	1.37 (1.35, 1.38)	1.48 (1.47, 1.50)
	second trimester	1.18 (1.17, 1.18)	1.15 (1.15, 1.16)	1.24 (1.23, 1.25)	1.29 (1.27, 1.30)
	third trimester	1.22 (1.22, 1.22)	1.19 (1.19, 1.20)	1.28 (1.27, 1.30)	1.38 (1.37, 1.40)
	preconception through pregnancy	1.24 (1.24, 1.25)	1.21 (1.20, 1.21)	1.32 (1.31, 1.33)	1.41 (1.40, 1.42)

^aThe corresponding IQR for PM_2.5, PM₁₀, NO₂, and SO₂ was 33.1 μg/m³, 48.5 μg/m³, 22.1 μg/m³, and 16.2 μg/m³, respectively. Cox frailty models were adjusted for maternal age, preconceptional BMI, education, occupation, ethnicity, folic acid supplementation, geographic region, season of conception, year of birth, AT, and RH. Abbreviations: N, number; PTB, preterm birth; IQR, interquartile range; CI, confidence interval; PM_2.5, particulate matter with aerodynamic diameters ≤2.5 μm; PM₁₀, particulate matter with aerodynamic diameters ≤10 μm; NO₂, nitrogen dioxide; SO₂, sulfur dioxide; AT, ambient temperature; RH, relative humidity; BMI, body mass index.

Associations between TSH Abnormalities and PTB

We observed that both subnormal (HR: 1.23 [95% CI: 1.22, 1.25]) and supranormal TSH levels (HR: 1.08 [95% CI: 1.06, 1.09]) were associated with higher risks of PTB compared to the reference group (Table S5). For different subtypes of PTB, subnormal TSH levels were associated with increased risks for very PTB and extremely PTB, whereas supranormal TSH levels showed no significant associations.

Interactions between Air Pollution and TSH Abnormalities

The additive interactive effects of air pollutant exposure and preconceptional TSH abnormalities on the risk of PTB and its subtypes are shown in Figures and S4–S6. Compared with participants with low air pollutant exposure and normal TSH levels, those with intermediate or high air pollutants exposure and abnormal TSH levels had higher HRs of PTB and its subtypes. Specifically, the HRs of PTB in the high pollution and subnormal TSH level group (PM_2.5: 2.27 [95% CI: 2.20, 2.34], PM₁₀: 1.88 [95% CI: 1.85, 1.92], NO₂: 1.50 [95% CI: 1.46, 1.55], SO₂: 2.10 [95% CI: 2.04, 2.15]), as well as high pollution and supranormal TSH levels (PM_2.5: 2.21 [95% CI: 2.14, 2.29], PM₁₀: 1.74 [95% CI: 1.70, 1.78], NO₂: 1.52 [95% CI: 1.47, 1.57], SO₂: 2.04 [95% CI: 1.98, 2.10]) group were all >1. In the additive scale, we observed significant interactions (RERI > 0, AP > 0, and SI > 1; all p < 0.05) between PM_2.5, PM₁₀, NO₂, and SO₂, with the abnormal TSH levels on the risk of PTB and its subtypes (Tables and S6–S8). For example, for high PM_2.5 exposure with subnormal TSH levels, the RERI, AP, and SI were 0.35 (95% CI: 0.27, 0.42), 0.15 (95% CI: 0.12, 0.18), and 1.38 (95% CI: 1.29, 1.46), respectively, indicating synergistic effects on PTB risk. Similar synergistic effects were observed between air pollutants and supranormal TSH levels, with AP ranging from 0.01 (95% CI: −0.03, 0.05) to 0.13 (95% CI: 0.09, 0.17).

1.

Risk of PTB according to PM_2.5 (A), PM₁₀ (B), NO₂ (C), and SO₂ (D) over 12 weeks preconception through pregnancy and preconception TSH categories in China, 2014–2020. Cox frailty models were adjusted for maternal age, preconceptional body mass index, education, occupation, ethnicity, folic acid supplementation, geographic region, season of conception, year of birth, ambient temperature, and relative humidity. Definitions of pollution level: low (<25th percentile), intermediate (25th–75th percentile), and high levels (>75th percentile). The corresponding 25th percentile for PM_2.5, PM₁₀, NO₂, and SO₂ was 36.1  μg/m³, 52.2 μg/m³, 18.5 μg/m³, and 11.4 μg/m³, respectively. The corresponding 75th percentile for PM_2.5, PM₁₀, NO₂, and SO₂ was 69.3 μg/m³, 100.7 μg/m³, 40.6 μg/m³, and 27.5 μg/m³, respectively. Definitions of TSH level: subnormal (<0.37 mIU/L), normal (0.37–4.87 mIU/L), and supranormal levels (≥4.88 mIU/L). Abbreviations: PTB, preterm birth; CI, confidence interval; PM_2.5, particulate matter with aerodynamic diameters ≤2.5 μm; PM₁₀, particulate matter with aerodynamic diameters ≤10 μm; NO₂, nitrogen dioxide; SO₂, sulfur dioxide; TSH, thyroid-stimulating hormone.

3. Additive Interactive Effects of Preconceptional TSH Abnormalities and Air Pollution over 12 weeks Preconception through Pregnancy on PTB Risk in China, 2014–2020 (N = 6,142,049) .

	subnormal TSH			supranormal TSH
pollution level	RERI (95% CI)	AP (95% CI)	SI (95% CI)	RERI (95% CI)	AP (95% CI)	SI (95% CI)
PM2.5
intermediate	0.13 (0.09, 0.17)	0.07 (0.04, 0.09)	1.15 (1.10, 1.21)	0.03 (−0.02, 0.09)	0.02 (−0.01, 0.04)	1.03 (0.97, 1.10)
high	0.35 (0.27, 0.42)	0.15 (0.12, 0.18)	1.38 (1.29, 1.46)	0.27 (0.19, 0.36)	0.12 (0.09, 0.16)	1.29 (1.20, 1.39)
PM10
intermediate	0.28 (0.21, 0.35)	0.15 (0.12, 0.19)	1.53 (1.40, 1.67)	0.23 (0.16, 0.31)	0.13 (0.09, 0.17)	1.43 (1.28, 1.58)
high	0.19 (0.15, 0.23)	0.10 (0.08, 0.12)	1.27 (1.21, 1.34)	0.03 (−0.02, 0.08)	0.02 (−0.01, 0.04)	1.04 (0.97, 1.12)
NO2
intermediate	0.12 (0.09, 0.16)	0.09 (0.06, 0.11)	1.42 (1.28, 1.58)	0.02 (−0.02, 0.07)	0.02 (−0.02, 0.05)	1.09 (0.93, 1.27)
high	0.003 (−0.05, 0.05)	0.002 (−0.03, 0.03)	1.01 (0.91, 1.11)	0.02 (−0.04, 0.08)	0.01 (−0.03, 0.05)	1.04 (0.93, 1.17)
SO2
intermediate	0.05 (0.01, 0.10)	0.03 (0.01, 0.06)	1.08 (1.01, 1.15)	0.06 (0.01, 0.10)	0.04 (0.01, 0.07)	1.15 (1.02, 1.30)
high	–0.09 (−0.15, −0.02)	–0.04 (−0.07, −0.01)	0.93 (0.87, 0.98)	0.14 (0.07, 0.21)	0.07 (0.04, 0.10)	1.16 (1.08, 1.24)

^aCox frailty models were adjusted for maternal age, preconceptional body mass index, education, occupation, ethnicity, folic acid supplementation, geographic region, season of conception, year of birth, ambient temperature, and relative humidity. Abbreviations: PTB, preterm birth; RERI, relative excess risk due to interaction; AP, attributable proportion due to interaction; SI, synergy index; CI, confidence interval; PM_2.5, particulate matter with aerodynamic diameters ≤2.5 μm; PM₁₀, particulate matter with aerodynamic diameters ≤10 μm; NO₂, nitrogen dioxide; SO₂, sulfur dioxide; TSH, thyroid stimulating hormone. To estimate the RERI, AP, and SI, the low-pollution category and the normal TSH groups were the reference categories.

^bDefinitions of pollution level: low (<25th percentile), intermediate (25th–75th percentile), and high levels (>75th percentile). The corresponding 25th percentile for PM_2.5, PM₁₀, NO₂, and SO₂ was 36.1 μg/m³, 522 μg/m³, 185 μg/m³, and 11.4 μg/m³, respectively. The corresponding 75th percentile for PM_2.5, PM₁₀, NO₂, and SO₂ was 69.3 μg/m³, 1007 μg/m³, 406 μg/m³, and 27.5 μg/m³, respectively.

^cDefinitions of TSH level: subnormal (<0.37 mIU/L), normal (0.37–4.87 mIU/L), and supranormal levels (≥4.88 mIU/L).

PTB Burden Attributable to Air Pollution and TSH Abnormalities

The PAF and number of PTB due to TSH abnormalities and air pollution exposure from preconception to delivery are presented in Figure and Table S9. We found that 14.78–33.75% of all PTB could be attributed to TSH abnormalities and air pollution exposure. Specifically, we found that high levels of PM_2.5 exposure were associated with an estimated 2207 (95% CI: 2151, 2262) to 60,903 (60,146, 61,651) additional PTB cases, accounting for 0.38% (95% CI: 0.37%, 0.38%) to 10.35% (95% CI: 10.23%, 10.48%) of all PTBs. For other air pollutants, the PAFs were 24.90% for PM₁₀, 14.55% for NO₂, and 25.77% for SO₂, each attributable to exposure levels above the reference levels.

2.

Population-attributable fractions (a) and number of preterm births (b) due to TSH abnormalities and air pollution exposure over 12 weeks preconception through pregnancy. Definitions of pollution level: low (<25th percentile), intermediate (25th–75th percentile), and high levels (>75th percentile). The corresponding 25th percentile for PM_2.5, PM₁₀, NO₂, and SO₂ was 36.1 μg/m³, 52.2 μg/m³, 18.5 μg/m³, and 11.4 μg/m³, respectively. The corresponding 75th percentile for PM_2.5, PM₁₀, NO₂, and SO₂ was 69.3 μg/m³, 100.7 μg/m³, 40.6 μg/m³, and 27.5 μg/m³, respectively. Definitions of TSH level: subnormal (<0.37 mIU/L), normal (0.37–4.87 mIU/L), and supranormal levels (≥4.88 mIU/L). Abbreviations: PM_2.5, particulate matter with aerodynamic diameters ≤2.5 μm; PM₁₀, particulate matter with aerodynamic diameters ≤10 μm; NO₂, nitrogen dioxide; SO₂, sulfur dioxide; TSH, thyroid-stimulating hormone.

Sensitivity Analyses

Sensitivity analyses largely supported the robustness of our findings (Tables S10–S12). The similar association estimates were observed in two-pollutant models. Further adjustment for the mode of delivery and iodine intake area in our main models did not change our results. The associations remained robust in analyses excluding participants with gestational week ≥42 and analyses using imputed data sets. Results using the official TSH reference range (subnormal level: <0.75 mIU/L; supranormal level: >5.60 mIU/L) were not different from our results. The Fine–Gray subdistribution hazard models yielded effect estimates for TSH–PTB associations that were comparable to, though marginally stronger than, those from the Cox regression analysis. Consistent with the main results, the results of weekly air pollution exposure showed that exposure to air pollution during preconception and middle to late gestational weeks was associated with increased risks of PTB and its subtypes (Figure S7). Finally, the results of restricted cubic spline regression showed a J-shaped dose–response relationship between preconceptional TSH or TSH MoM levels and PTB (Figure S8).

Discussion

To our knowledge, this is the first nationwide study to investigate the interaction between air pollution and preconceptional TSH levels on PTB. In this study, we found that air pollution exposure spanning from 12 weeks preconception to delivery was associated with increased risks of PTB and its subtypes. Of note, we observed positive additive interaction between air pollution and preconceptional TSH abnormalities on the risk of PTB (RERI > 0, AP > 0, and SI > 1). Higher risks of PTB were observed among participants with high air pollution exposure and TSH abnormalities. These findings may have important implications for air pollution control and management of maternal thyroid during preconception.

Although evidence on the association of maternal exposure to ambient PM and PTB has been growing, few studies have focused on exposure during the preconception period. ,, A cohort study among 4928 pregnant women in Guangzhou, China, reported positive but nonsignificant associations between PTB and exposure to high levels (the fourth quartile) of PM_2.5 and PM₁ during the 12 weeks before pregnancy, while a negative association was found with PM₁₀. However, significant increases in the risk of PTB were found with PM_2.5 and PM₁ exposure during the 12th to 20th gestational weeks. Likewise, a US study of 233,502 electronic medical records reported positive associations for PTB with both PM_2.5 and PM₁₀ during the first 7 weeks and the last 6 weeks of pregnancy. Conversely, exposures during the 12 week preconception period showed negative associations. A retrospective study in six California urban counties found that exposure to PM_2.5 during each trimester was associated with increased risks of PTB, whereas no significant association was observed during the 12 weeks before pregnancy. However, in the present study, we observed overall significant associations of PM_2.5 and PM₁₀ exposure during both preconception and each trimester. These associations were consistent in additional sensitivity analyses using distributed lag models. The 21st to 36th weeks of gestation were identified as a potential susceptible window for PM₁₀ exposure. The inconsistency in these findings across studies might be due, at least in part, to differences in population characteristics, local air pollution levels, sources or chemical components of the PM mixture, and the methods used to assess pollutant concentrations. For instance, the average levels of PM_2.5 (13.5 μg/m³) and PM₁₀ (22.3 μg/m³) during the 12 weeks before pregnancy in the two US studies were much lower than those in this study (55.1 μg/m³ for PM_2.5 and 79.2 μg/m³ for PM₁₀). Despite reporting relatively high air pollution levels (32.1 μg/m³ for PM_2.5 and 48.7 μg/m³ for PM₁₀), the aforementioned study in China is limited by its single-center design and restricted temporal scope (2016–2017), which reduces its generalizability to larger temporal and spatial scales. In addition, although PM is conventionally measured by total mass, PM_2.5 originates from various natural or anthropogenic sources and exhibits distinct physical and chemical characteristics that may contribute to varying toxic effects. A recent study indicated that energy coal combustion, agriculture, and road traffic are the primary sources of PM_2.5 in the eastern US, while wildfires are the dominant source in the western US. In China, industry is the primary emission source of PM_2.5 in the northern, southern, and eastern regions, while the residential source dominates in the northeastern, central, and southwestern regions. The heterogeneity in PM sources or components across regions may partly explain the inconsistency in the estimates observed across studies, even though our study and these previous studies did not directly examine the effects of PM species. The current study may have expanded our understanding of the association between maternal PM_2.5 and PM₁₀ exposure and PTB across a wider range of exposure levels, especially for the preconception window. The effects of preconception air pollution exposure on PTB remain to be further explored, as existing population-based studies have primarily focused on exposure during pregnancy. A better understanding of the impact of air pollution exposure prior to pregnancy is essential for establishing comprehensive preconception healthcare. Herein, our findings suggest that exposure to ambient PM before and during pregnancy could significantly increase the risk of PTB. Further studies investigating the effects of specific PM_2.5 constituents on PTB may help to clarify the toxic effects of PM_2.5 on fetal development.

In this study, we observed significant increases in the risk of PTB and its subtypes associated with exposure to gaseous air pollutants (i.e., NO₂ and SO₂) during preconception and each trimester of pregnancy. Sensitivity analyses using two-pollutant models showed that the effects of NO₂ and SO₂ were independent of PM. A study among women with asthma found that exposure to NO _x during preconception and pregnancy was associated with total PTB and early PTB, but SO₂ was not associated with PTB. A recent study among pregnant women with type I diabetes showed no significant association of PTB with NO _x and SO₂ exposure, both during preconception and any trimester of pregnancy. To our knowledge, no studies have examined the effects of NO₂ and SO₂ exposure during the preconception period on the risk of PTB in the general population of pregnant women. NO₂ is directly produced by various anthropogenic activities (e.g., traffic) and may serve as a crucial precursor in the formation of ground-level O₃ and secondary PM. Growing evidence indicates that inhaling NO₂ can lead to rapid increases in circulating nitrite levels. SO₂, an indicator of pollutants from regional coal-fired power plants, can be converted into sulfurous acid in the lungs, which then generates bisulfite and sulfite derivatives that are absorbed into the circulation system or other body fluids. Collectively, our findings further support that maternal exposure to NO₂ and SO₂ during preconception and pregnancy is associated with higher risks of PTB. However, our exposure estimates and those from previous studies were based on participants’ residential addresses, which do not accurately represent individual exposure and could introduce bias. More population-based studies are warranted in the future to use more precise exposure assessment methods, such as personal monitoring, spatiotemporal modeling that incorporates activity patterns, and improved time-solved exposure assessment.

Our findings support that maternal exposure to air pollution during the critical period before conception may be associated with shorter gestation. In speculation, the association between air pollution and PTB may depend not only on acute effects but also on the buildup of accumulation effects of pollutants or corresponding metabolites over a longer period, in which case the risk of PTB may increase significantly when high-level exposure occurs during the preconception period. Three months before conception is an important time for preantral follicle maturation and ovulation. Preconception exposure to air pollution has been considered in association with adverse pregnancy outcomes via oxidative stress, inflammation, DNA methylation alterations in gametes, and compromised placental growth and function. ,, Maternal preconceptional exposure to air pollution may affect oocyte quality, reduce embryo viability and development, and such adverse processes are converging risk factors for PTB. , Evidence suggests that air pollutants may also influence ovarian and uterine function and disrupt normal fertility and implantation, , which likely contributes to the etiology of PTB. In addition, paternal preconceptional exposure to air pollution may alter sperm epigenetic modifications, reduce semen quality, and thus affect embryonic development. , A better understanding of the impact of air pollution exposure prior to pregnancy is essential for establishing comprehensive preconception healthcare.

The adverse effects of thyroid dysfunction during pregnancy have been affirmed in previous studies. , Even subclinical hypothyroidism (defined as increased TSH with normal thyroid hormone) has been found to increase the risk of multiple adverse pregnancy outcomes, including PTB. , Several recent studies have reported that maternal TSH abnormalities before pregnancy are associated with elevated risks of PTB. , Our recent study among 5.8 million Han Chinese women found a significant association between preconception TSH levels and a 13–23% elevated risk of PTB, even when TSH levels were greater than 2.50 mIU/L, yet still within the clinical reference range. These findings were further confirmed in this study, in which we observed that both subnormal and supranormal TSH levels during preconception were associated with increased risks of PTB at the national level. Although systematic differences may exist across counties, significant associations were consistently observed in sensitivity analyses using the recommended office reference range and TSH MoM value as the dependent variable. More importantly, we found that preconceptional TSH abnormalities and high air pollution exposure synergistically increased the risk of PTB. In this study, up to 15% of PTB risk could be attributed to the additive interactions and SI > 1, implying that the interactive effect of TSH abnormalities and air pollution exceeded the sum of their individual effects. Further, we estimated that air pollution and TSH abnormalities accounted for up to 33.75% of PTB cases. Specifically, exposure to PM_2.5 was responsible for an attributable fraction of 24.9%. This finding aligns with previous estimates, which suggest that 28% to 40% of the PTB burden in China is attributable to PM_2.5 exposure. , Collectively, the common potential pathways of air pollution exposure and TSH abnormalities, including inflammatory, endocrine disorders, and compromised placental function, make the synergistic effects biologically plausible. , Previous studies proposed that air pollution exposure and thyroid dysfunction may cause intrauterine inflammation and oxidative stress, which may be risk factors of PTB. , Furthermore, air pollution exposure and TSH abnormalities may disrupt the endocrine and stimulate the uterine contraction, which leads to an early delivery. , In addition, air pollution and thyroid dysfunction may impair placental function, affecting fetal growth and development, potentially inducing the onset of labor. , From a public health perspective, the observed synergistic interactions between TSH abnormalities and air pollution could help identify high-risk populations, which has potential guiding importance for the primary prevention of PTB. Further studies are warranted to confirm our findings and explore the causal relationship.

Previous epidemiological studies have examined the mediation effects of maternal or fetal thyroid function on the associations between maternal exposure to air pollution and birth outcomes, as well as offspring health. , A birth cohort among 499 mother–child pairs in Limburg, Belgium, found that maternal exposure to PM_2.5 during the third trimester was associated with decreased fetal free thyroxine (FT4) levels, potentially contributing to reduced birth weight. Another prospective birth cohort study among 433 pregnant women in Nanjing, China, suggested that maternal FT4 may mediate the relationship between PM_2.5 and its components’ exposure during the first trimester and birth weight. In a cohort study among 1329 mothers in Wuhan, China, preconceptional exposure to PM_2.5 was associated with lower maternal FT4 levels and higher FT4/FT3 ratios during early pregnancy, which contributes to impaired neurodevelopment in offspring. In the present study, the exposure window of air pollution was estimated from 3 months prior to conception to delivery. However, maternal TSH levels in our study population were measured only during the preconception examination stage, and 29.2% of participants became pregnant more than 3 months after health examination, which limited our ability to examine the mediation effects of maternal TSH. Future studies are needed to explore whether maternal preconceptional thyroid function could mediate the association between air pollution and PTB.

Our findings emphasize the importance of continuing education for medical professionals and the public and the necessity for increased awareness about the potential hazard of TSH abnormalities and air pollution in childbearing women. Medical guidelines agree that suspected thyroid dysfunction should be confirmed and managed as early as possible in pregnant women. However, thyroid dysfunction before pregnancy should not be ignored. Given the relatively high prevalence of maternal thyroid dysfunction worldwide, the high levels of air pollution, especially in low- and middle-income countries, and the considerable individual and societal burdens of PTB, policymakers should prioritize improving air quality and managing thyroid status. Form a clinical standpoint, obstetricians and endocrinologists should take note of the emerging evidence and adopt a more active role in communicating risks to childbearing women or patients and integrating environmental risk factors such as air pollution into clinical management.

This study has several strengths. First, in addition to utilizing a prospective study design to investigate the associations between exposures and outcomes of interest, incorporating a large sample size in the analysis could greatly enhance the robustness of our findings. Our participants were distributed across China, with large spatial variations in air pollution exposure, which allowed us to assess the association between air pollution and PTB, especially in heavily polluted regions. Second, we investigated and quantified the interactive effects between preconception TSH levels and air pollution exposure from preconception through pregnancy, which highlights the importance and potential benefits of preconception healthcare. However, we acknowledge that this study is subject to several limitations. First, in China, air pollution is primarily driven by coal energy use and motor vehicle emissions, which vary across regions due to urbanization and industrialization. Although air pollution exposure estimates applied in this study were derived from prediction models that incorporate local emissions and land-use information, the relatively low spatial resolution of air pollutant data may fail to capture spatial variation in exposure estimates to produce potential exposure misclassifications, which could underestimate the impact of air pollution. Second, we estimated air pollution exposure based on residential addresses and lacked detailed information on activity patterns and indoor exposure, which may lead to inevitable exposure misclassification. Third, thyroid function status was evaluated solely on the basis of TSH levels, while information about thyroid autoantibodies and thyroid hormones such as FT3 and FT4 was missing. This is an important limitation, as changes in TSH may reflect alterations in pituitary response or sensitivity to thyroid hormone rather than differences in thyroid hormone availability. In addition, the extent of thyroid dysfunction and the status of thyroid autoimmunity have different associations with PTB. The influence of maternal TSH levels on the risk of PTB observed in this study may have been obscured by thyroid hormones and autoimmunity. However, TSH has served as the gold standard of the first-line thyroid function test for most clinical situations including pregnancy. Moreover, performing detailed thyroid function tests for all participants in such a large-scale project may be impractical and cost-prohibitive. Fourth, maternal thyroid physiology and TSH levels were altered during pregnancy, but TSH levels were measured only once before pregnancy. Therefore, classifying solely based on preconceptional TSH levels may have been inadequate to accurately reflect the association between TSH and PTB. The inclusion of participants with thyroid dysfunction who may receive treatment during gestation could potentially lead to an underestimation of the associations. However, based on the significantly higher risk of PTB in participants with preconception TSH abnormalities, we speculate that thyroid status was not well managed in this cohort. Previous studies have also reported that 72.8–96% of general adults with TSH abnormalities persisted in abnormal TSH levels during follow-up. − Fifth, iodine serves as an important micronutrient for the synthesis of thyroid hormones, and previous studies reported that iodine deficiency was relevant to an increased risk of PTB; however, iodine status at the individual level was lacking in this study, which may bias the observed associations. Nonetheless, China has implemented the universal salt iodization policy, and a recent national survey reports that the majority of areas are iodine-sufficient. Additionally, our sensitivity analysis controlling for area-level iodine intake shows consistent results with the main analyses. Sixth, although many confounding factors have been adjusted in our analysis, we cannot fully exclude potential residual confounding due to unknown and unmeasured factors. For example, gestational weight gain and maternal disease status during pregnancy (e.g., gestational hypertension and gestational diabetes) are relevant to the risk of PTB but were not considered in this study; this may have underestimated the effect estimates. Seventh, previous studies have indicated that air pollution exposure and maternal thyroid dysfunction may cause early pregnancy loss or stillbirth in low gestational age fetuses, ,,, which may lead to potential downward bias or even inverse associations between air pollution or thyroid dysfunction and PTB. Thus, the observed associations between air pollution exposure and TSH abnormalities with PTB in our study may have been underestimated. To compensate for this limitation, we incorporated pregnancy loss as a competing risk in Fine–Gray subdistribution hazard models, and the results were consistent with our main findings. Lastly, given the different etiologies of spontaneous and medically indicated PTB subtypes, it is necessary to distinguish between these groups to evaluate associations. However, we could not estimate the effects of air pollution exposure and TSH abnormalities on spontaneous and medically indicated PTB due to a lack of relevant information, which limits our ability to elucidate distinct pathogenic mechanisms.

Conclusions

In this nationwide cohort study, we found that exposure to ambient PM_2.5, PM₁₀, NO₂, and SO₂ spanning preconception through pregnancy was associated with increased risks of PTB and its subtypes. More importantly, we observed synergistic effects between air pollutant exposure and preconceptional TSH abnormalities on PTB. These findings highlight the importance of implementing rigorous air pollution control policies and managing preconception thyroid health.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/envhealth.5c00444.

• Additional information on the method, descriptive statistics, and distribution of air pollutants; risk of PTB subtypes due to interactive effects; and results of sensitivity analyses (PDF)

††.

Y.Y., X.H., and Hongbing Xu contributed equally to this work. Ying Yang, Xinghou He, and Hongbing Xu designed research, performed research, analyzed data, and wrote the manuscript. Jianbin Wu, Bin Zhang, Xinyi Lv, Long Wang, Mengyao Li, Chuanyu Zhao, Xuyang Shan, Yuan He, Yan Fang, and Yuanyuan Wang performed data collection and cleaning. Huiying Xu, Jihong Xu, Xiaoming Song, Ya Zhang, Hongguang Zhang, Qinghong Zhang, and Zifa Wang managed study databases. Xu Ma and Wei Huang acquired funding, conceived the study, and revised and proofed the manuscript. All authors have read the final version of the manuscript and approved the submission.

The authors declare no competing financial interest.