We obtained data from the National Free Pre-Pregnancy Checkups Project (NFPCP), which aimed to offer free prepregnancy health examination and counseling to couples of reproductive ages who intended to become pregnant within the following 6 months. After prepregnancy examination, trained local health professionals followed up these couples through face-to-face or telephone interview to check on pregnancy status every 3 months for up to 1 year. If pregnant, health personnel would follow up with these participants on their lifestyle in the early stage of pregnancy and pregnancy outcomes after delivery. If pregnancy was not achieved within that year, follow-up was discontinued. Detailed information about how this project was designed, organized, and implemented can be found in previous studies [29-31].

The exclusion criteria for this cohort study are illustrated in Figure 1. Initially, we included 915,931 women from the NFPCP in Guangdong, China, who gave birth from January 1, 2014, to December 31, 2019, and had complete pregnancy outcomes. Then, we excluded 32,014 cases of fetal death, stillbirth, induced labor, or abortion; 9122 cases with multiple births; 159,778 cases with missing or extreme thyrotropin values (TSH>500 mIU/L); 43,047 cases with missing residential addresses; 448 cases of women with missing age data or women who were not of reproductive age (<20 or >49 years); 6863 cases with a gestational age of less than 20 weeks or greater than 42 weeks; 122 cases that lacked data on the infant’s sex; and 2853 cases of women who gave birth to babies with extreme birth weights (<500 g or >5000 g). Then, women who conceived 42 weeks before this study ended (December 31, 2019; n=28,164) were excluded, with the aim of avoiding fixed cohort bias [32]. In addition, 4 cases with duplicate information were excluded. Eventually, this study included 633,516 women who delivered living singletons.

PTB, defined as delivery at a gestational age of earlier than 37 weeks [33], was the primary outcome. In addition, we further divided PTB into 2 categories: early PTB (EPTB; 20‐33 wk of gestation) and late PTB (LPTB; 34‐36 wk of gestation) [34]. We determined the gestational age of the mother by taking into account the first day of her last menstrual period along with the delivery date.

In our study, serum TSH levels were assessed during the preconception period. Blood samples were collected from participants at the preconception examination, which occurred within 6 months before pregnancy. The serum TSH concentration was measured using an electrochemiluminescence immunoassay and a TSH detection kit.

Due to the lack of authoritative guidelines on the reference range of TSH among prepregnancy women, a population-based reference range of TSH was established from the current cohort. Based on previous studies [15,16], the reference population of 311,089 women was selected according to the following criteria: no previous adverse pregnancy outcomes; no history of thyroid diseases, anemia, high blood pressure, or diabetes mellitus; not taking oral contraceptives in the past; not consuming alcohol or cigarettes; having a normal BMI (18.5‐24 kg/m²); giving birth to healthy infants. The 2.5th and 97.5th percentiles for TSH levels were 0.10 mIU/L and 4.11 mIU/L, respectively. The reference range of TSH for women who planned to become pregnant in this cohort was 0.10‐4.11 mIU/L.

According to the reference range of TSH, we divided all participants into 2 groups: the normal TSH group (TSH 0.10‐4.11 mIU/L) and the abnormal TSH group (TSH <0.10 or TSH >4.11 mIU/L).

We collected daily (24 h) concentration data of PM_2.5 (μg/m³) from air pollution monitoring stations located within Guangdong province from the China National Environmental Monitoring Centre. Daily concentrations of SO₂ (μg/m³), NO₂ (μg/m³), and O₃ (μg/m³) were also obtained from the same monitoring stations for adjustment of concurrent exposure to gaseous pollutants.

We geocoded the residential addresses of all pregnant women enrolled in this study to longitude and latitude coordinates using the Gaode Maps open platform. Then, the air pollution exposure of each participant was estimated by using the inverse distance weighting (IDW) method, which has been extensively used in epidemiological research [1-3,21,25,27,35]. Previous studies [36,37] have reported that the health effects of environmental exposure on fetal development varied by trimesters of pregnancy. Therefore, we estimated the exposure to air pollutants for all pregnant women during the first trimester (0‐13 wk of gestation), the second trimester (14‐26 wk of gestation), and the third trimester (27 wk of gestation to delivery), as well as the entire pregnancy. To investigate more refined critical exposure windows during pregnancy, we estimated individual exposure to air pollutants during each gestational week. The time frame of exposure for pregnant women during the entire pregnancy was shown in Multimedia Appendix 1.

Furthermore, to account for potential meteorological confounding factors, we obtained daily mean ambient temperature (℃) and relative humidity (%) from meteorological monitoring stations in Guangdong province via the China Meteorological Data Sharing System. A similar approach for meteorological factors such as air pollution exposure was adopted.

Numbers (percentages) were used to express participants’ baseline characteristics according to TSH levels. Using Pearson correlation, we examined the correlation between air pollutants and meteorological factors in Guangdong from 2014 to 2019.

To evaluate the effect of PM_2.5 exposure throughout the whole pregnancy as well as during each trimester on PTB and PTB subtypes, Cox proportion hazard models were constructed by treating PTB as the outcome and gestational week as the time scale. The effect of prenatal PM_2.5 exposure on PTB and its subtypes was evaluated separately in women with normal and abnormal preconception TSH levels. The effects were expressed as the hazard ratios (HR) for each 10-μg/m³ rise in PM_2.5 concentration during different pregnancy periods. According to previous studies [38], we adjusted for maternal age (<25, 25‐34, or ≥35 years), delivery mode (vaginal delivery or cesarean delivery), prepregnancy BMI (<18.5, 18.5‐24, 24‐28, or >28 kg/m²), maternal smoking (yes or no), maternal drinking (yes or no), the season of delivery (spring, summer, autumn, or winter), and infant’s sex (male or female). In addition, based on previous studies [38,39], natural splines with df of 6 and 3 were used to adjust the nonlinearity of ambient temperature and relative humidity, respectively. Moreover, we built distributed lag models incorporating Cox proportional hazards models with adjustments for the aforementioned covariates to identify more refined critical susceptible exposure windows [27,40]. This model could explore the exposure-response association and lag-response association simultaneously based on a “cross-basis” function. In the distributed lag models of this study, the exposure-response association was assumed to vary smoothly across gestational weeks. The lag distribution of PM_2.5 was modeled as natural cubic splines with optimal df of 4 based on the minimum of the Akaike Information Criterion by varying df from 3 to 10. Additionally, the maximum lag range was set at week 36, since term birth was censored at week 37. The analysis was performed for EPTB and LPTB as we did for overall PTB, except that the exposure period was confined to gestational weeks 1 to 33 for early PTB.

Several sensitivity analyses were conducted. First, we changed the df for mean temperature from 5 to 7 and the df for relative humidity from 2 to 4. Second, all participants were categorized into the abnormal TSH and normal TSH groups based on the reference range of TSH (0.1‐4 mIU/L) for mothers during early pregnancy recommended by the American Thyroid Association [41], as some researchers [42] suggested that pregnant women in early pregnancy and nonpregnant women have similar reference ranges for TSH. Third, we excluded participants whose baseline characteristics were missing and then conducted the analysis. Last, considering the potential confounding effect of other air pollutants (NO₂, SO₂, or O₃), 2-pollutant models were constructed to explore the relationship between atmospheric pollutants and PTB and PTB subtypes.

Statistical analysis was carried out using R (version 4.2.1; R Foundation for Statistical Computing). All statistical tests were 2-sided, and all statistics were considered significant when the P value was lower than .05.

This study was approved by the Medical Ethics Committee of the Guangdong Provincial Reproductive Science Institute (ID of the ethics approval: 202216). This study was in line with the Helsinki ethical guidelines. Written informed consent was obtained from all participants. Before recruitment, each participant provided written informed consent. This study used anonymized data in order to protect the privacy of participants, and no individually identifiable information was available.

We investigated the associations between PM_2.5 exposure and PTB in women with abnormal and normal preconception TSH levels. This study is the first to examine the modulating effect of preconception TSH levels on the association between PM_2.5 exposure and PTB. According to our study, PM_2.5 exposure throughout the entire pregnancy was linked to elevated risks of PTB in women with abnormal and normal preconception TSH levels. Our findings would provide important population-based evidence and a scientific basis for preconception care to further promote maternal and child health.

There is increasing research on the relationship between PM_2.5 exposure and PTB. Among women with abnormal and normal preconception TSH levels, our study observed a positive association between PM_2.5 and PTB, similar to previous studies [8,43,44]. For instance, Laurent et al [44] found that for every increase in the IQR of PM_2.5 throughout the pregnancy, the risks of PTB increased by 13.3%. However, no significant association has been found in some studies [44-47]. Disagreement among studies may be due to differences in the study design, exposure levels to air pollutants, particulate matter components, as well as the sample size.

Additionally, in women with abnormal preconception TSH levels, stronger associations between PM_2.5 exposure and PTB were found compared to women with normal preconception TSH levels. A previous study indicated that thyroid hormone might play a role in inflammation related to PTB [48]. In addition, some researchers found that PM_2.5 exposure could contribute to inflammation [49,50]. We hypothesize that the exacerbated risk of PTB observed in women with thyroid dysfunction might be attributed to a higher vulnerability to the inflammatory and oxidative stress effects of PM_2.5. The interrelationship between thyroid disorder and increased susceptibility to environmental pollutants could be explained by several mechanisms. First, thyroid dysfunction may impair the body’s antioxidant defenses, rendering individuals more susceptible to oxidative damage induced by PM_2.5 [51]. Second, altered thyroid hormone levels can disrupt the immune system balance, potentially exacerbating the inflammatory response to air pollution [52]. Additionally, maternal thyroid disorder may increase the risk of gestational diabetes and hypertension, which are risk factors for PTB [53,54]. However, the underlying biological mechanisms of these associations remain unclear due to the lack of toxicological evidence. Further studies are warranted to explore how TSH levels modulate the effects of PM_2.5 exposure on PTB.

Furthermore, our study also identified the sensitive exposure windows. We found that the sensitive windows might be the 3rd-13th and 28th-35th gestational weeks for women with abnormal preconception TSH levels. We also found that the sensitive windows might be the 1st-13th and the 21st-35th gestational weeks for women with normal preconception TSH levels. It appears that early pregnancy and late pregnancy might be the susceptible exposure windows. A previous study [55] suggested that PM_2.5 exposure in late pregnancy may stimulate the release of cytokines that promote inflammation, thereby triggering PTB. Additionally, some studies have suggested [55,56] that early pregnancy might be the most sensitive window. However, there has been no consistent conclusion regarding the susceptible exposure window. Therefore, further studies are warranted to confirm our results.

Only a very small number of studies have explored the relationship between PM_2.5 exposure and specific PTB subtypes. Our study found a stronger association between PM_2.5 exposure during the entire pregnancy and LPTB than that between such PM_2.5 exposure and EPTB in women with abnormal and normal preconception TSH levels. We speculate that the effects of air pollution may be “masked” by other risk factors such as intrauterine infections and nutritional deficiencies in infants born early preterm.

Our study results have important public health implications. Previous studies have noted that PTB could place significant burdens on families, health systems, and socioeconomics. To mitigate the risks of PTB, women with abnormal preconception TSH levels should take additional protective measures against air pollution, as they were identified as the susceptible population in this study. For women with abnormal preconception TSH levels, we recommend enhancing the education on reproductive knowledge and improving the accessibility of health care services.

Our study has several strengths. First of all, this study has a relatively large sample size, enabling sufficient statistical power to assess the modulating effect of TSH on the impact of PM_2.5 exposure on PTB. Second, we explored more precise susceptible windows, which could help us to develop specific clinical and public health interventions to reduce the risks of PTB, as well as inform research on potential etiological mechanisms regarding the relationship between PM_2.5 and PTB. Third, we investigated the relationships between PM_2.5 exposure and PTB subtypes as well.

However, several limitations should be noted. First, we did not collect the TSH levels during pregnancy. Therefore, classification based on preconception TSH alone may not be adequate to accurately reflect the modulating effects of serum TSH levels on the association between PM_2.5 and PTB. Second, couples who participated in the NFPCP were those who intended to become pregnant, which may have led to selection bias with regard to those with unplanned pregnancies. More specifically, participants who plan to become pregnant will be less exposed to other risk factors compared to those with unplanned pregnancies owing to them taking protective measures or developing a healthy lifestyle. Third, air pollution exposure for all participants was estimated based on the residential addresses using the IDW method, a common practice in environmental health research [21,25,27,35]. This method assumes spatial continuity, where PM_2.5 concentrations at unmeasured locations are interpolated based on nearby monitoring station data, adhering to the principle that environmental measurements in closer proximity are more similar than those farther apart. While this approach requires assumptions about spatial distribution and participant mobility, the IDW method was used for its applicability in large-scale epidemiological studies where direct, individual-level exposure measurement is challenging. More advanced exposure assessment methods, such as land use regression or random forest, were not conducted in this study because we did not have research data such as land use information, which is often used in these exposure assessment methods. Future studies should consider more accurate exposure assessment methods. In addition, we did not consider the mobility of participants due to unavailable information on residential mobility. Therefore, exposure misclassification was possible in this study. Fourth, due to the lack of relevant information, some confounding factors, such as thyroid-related medications, were not adjusted in the model.

In summary, PM_2.5 exposure during pregnancy increased PTB risks for women with abnormal and normal preconception TSH levels. Additionally, stronger associations were found among women with abnormal preconception TSH levels. Moreover, women in early or late pregnancy appear to be more vulnerable to harmful effects from air pollution. Additional protective measures against air pollution, such as wearing masks and using air purifiers, should be taken, especially for pregnant women with abnormal preconception TSH levels.