The study participants were pregnant women who were part of the Peking University Retrospective Birth Cohort for whom data were available in the hospital information system of Tongzhou for the period between January 2013 and December 2017. All the participants received antenatal examinations and gave birth at the Tongzhou Maternal and Child Health Care Hospital, which is the maternal health center for Tongzhou district, Beijing. The sociodemographic and health details of the participants were recorded at the antenatal examinations, including maternal age (<35 years vs ≥35 years), employment status (employed vs unemployed), parity (nulliparous vs multiparous), prepregnancy weight, height, maternal education level (high school or below, junior college, university or above), maternal ethnicity (Han vs non-Han), first day of last menstrual period, and delivery date. Prepregnancy BMI was calculated by dividing weight (in kilograms) by height (in meters squared) and categorized according to the WHO standards for Asian women: underweight (BMI <18.5 kg/m²), normal weight (BMI 18.5-23 kg/m²), overweight (BMI 23-27.5 kg/m²), or obese (BMI >27.5 kg/m²) [20]. The study data were extracted from the electronic medical records of Tongzhou Maternal and Child Health Care Hospital and were routinely reviewed by in-house professional data engineers every week.

The inclusion criteria of this study were (1) a single gestation, (2) an available residential address, (3) no history of previous HDPs, and (4) a delivery date before the 42nd gestational week. A total of 43,894 pregnant women met the inclusion criteria. We excluded 21,324 pregnant women who met the following exclusion criteria: (1) preexisting conditions, including metabolic syndrome; hemolysis, elevated liver enzymes, slow platelets; or hypertension before the HDP diagnosis; (2) missing data for delivery date; (3) an age at delivery less than 18 years; and (4) missing values for maternal age, ethnicity, education level, employment status, or prepregnancy BMI. Finally, 22,570 women were included in this study. The flowchart in Multimedia Appendix 1 shows the details of participant selection.

The de novo HDP group was defined as patients who experienced GH or PE during the current pregnancy. The normotensive group was defined as those free of any diagnosis of GH or PE and without a history of hypertensive disorders. Information on the history of hypertensive disorders was obtained from initial antenatal examination records. GH or PE diagnoses were made by obstetricians in accord with the Chinese Clinical Practice Guidelines, which are consistent with the guidelines of developed countries [21]. International Classification of Diseases–10 (ICD-10) codes were used to define GH and PE (the ICD codes for GH and PE are described in our previous study [22]). Medical records containing information related to GH and PE were extracted for double confirmation of the diagnosis and to obtain the week of disease diagnosis.

The pregnant women’s residential addresses were collected at the initial antenatal examination. Each residential address was transformed into longitude and latitude values to determine the concentration of PM_2.5 or PM₁ in the center of the nearest 1 km × 1 km area on a grid. Daily PM concentrations were calculated at a spatial resolution of 1 km × 1 km in Beijing during the study period using satellite remote sensing, meteorology, and land-use information. Details of the calculations and validation of PM daily concentration measurements have been reported previously [23]. Data for daily temperature and relative humidity during the study period were obtained from the National Oceanic and Atmospheric Administration [24]. Average daily values for PM, temperature, and relative humidity from preconception (12 before gestation) to delivery (42 gestational weeks) were used to assess atmospheric exposure in our analysis.

The study was approved by the Institutional Review Board of Peking University Health Science Center (IRB00001052-21023). All the study participants provided written consent. The study data were anonymized to protect the privacy of the study participants.

The Student t test (2-tailed) and chi-square test were applied to assess differences in the distribution of characteristics between the subjects with HDPs and those who were normotensive. The relationships among PM_2.5, PM₁, temperature and relative humidity were assessed with the Pearson correlation test.

We followed the weekly exposure to PM₁ and PM_2.5 from the 12th week before gestation to the 42nd gestational week [25]. Cox proportional hazard models with a DLNM were used to estimate the association between the weekly PM_2.5 and PM₁ exposures and the hazard of de novo HDPs. The basis of the Cox proportional hazard model is the proportional hazards assumption, which means that all the study participants had the same hazard function. Our study complied with this assumption, which means that the ratio of the hazards for any 2 individuals was constant over time [26]. A DLNM was adopted to assess the effects of weekly PM exposures on hazards for HDPs by the “cross basic” function (cb) by constructing a 2D cross matrix that included exposure doses and time to control for the lag effect of PM exposure and the nonlinear relationship of exposure–reaction. A natural spline with 3 degrees of freedom was used to further accommodate the nonlinear effects of PM exposure, temperature, and relative humidity. The maximum number of lag weeks was set at 54 (from the 12th week before gestation to the 42nd gestational week). The lag effect of PM exposure was evaluated by the model as follows: ln(h(t,X)/h0(t)) = β1*Zt + β2*ns(temperature,3) + β3*ns(relative humidity,3) + β4*covariables. “ln(h(t,X)/h0(t))” indicates the hazard of de novo HDPs at the specific lag week (t), and the filed event is de novo HDP, noted as X. Zt represented the cross matrix of PM at specific lag weeks. β2 and β3 are the regression coefficients for temperature and relative humidity with 3 degrees of freedom in the natural spline function. β4 is the coefficient for the covariables, including maternal age, employment status, prepregnancy BMI, maternal education level, parity, maternal ethnicity, and conceptional year.

We also performed several sensitivity analyses to test the consistency of the main results. First, we further used natural splines with 4 and 5 degrees of freedom in the cross matrix to assess the hazards of de novo HDPs. Second, considering the clinical differences between GH and PE, we repeated our analyses excluding GH patients to explore the independent association between PM and PE. Meanwhile, stratified analyses were conducted across strata for differences in employment status, parity, and maternal education level.

All statistical analyses were performed with R (version 4.0.0; R Foundation for Statistical Computing). A 2-sided P value <.05 was considered a significant difference.

There were 1520 participants with de novo HDPs and 21,050 normotensive participants in this study. Table 1 shows the characteristics of the participants. The mean maternal age of the pregnancies was 29.1 (SD 4.1) years. There were 3802/22,570 (16.8%) unemployed women, and 21,222/22,570 (94%) women of Chinese Han ethnicity. A higher percentage of de novo HDPs was observed in the follow sociodemographic strata: low education level, primiparity, and prepregnancy overweight or obesity.

The distribution of PM_2.5, PM₁, temperature, and relative humidity is summarized in Multimedia Appendix 2. The average concentration of PM_2.5 was 74.2 (SD 53.2) μg/m³, ranging from 8.5 to 398 μg/m³. The concentration of PM₁ ranged from 16.0 to 75.1 μg/m³. Mean temperature and relative humidity were 13.5 °C (SD 11.0 °C) and 54.7% (SD 18.2%). Figure 1 presents the daily fluctuation in PM_2.5 and PM₁ concentration from January 2013 to December 2017, revealing seasonal fluctuations in PM_2.5 and PM₁ exposure. Meanwhile, PM_2.5 and PM₁ were strongly correlated (r=0.80). The Pearson correlations among PM_2.5, PM₁, temperature, and relative humidity are depicted in Multimedia Appendix 3.

Figure 2 shows that a 10 μg/m³ increase in PM₁ and PM_2.5 was associated with de novo HDPs in 2D and 3D plots. The hazard of de novo HDPs was associated with PM₁ exposure from the 5th week before gestation to the 6th gestational week, and the strongest effect of PM₁ exposure was observed in the 2nd gestational week (HR 1.024, 95% CI 1.007-1.042; Figure 2A). Figure 2B depicts the effects of PM₁ exposure on de novo HDPs with different exposure levels and different lag weeks, with a significant single peak by lag week from preconception to the first trimester (from lag 36 weeks to lag 47 weeks). For PM_2.5, the significant lag window was from the 1st week before gestation to the 6th gestational week and the maximum lag effect of PM_2.5 was found in the 6th gestational week (HR 1.007, 95% CI 1.000-1.015, Figure 2C). Similarly, we also observed a significant single peak from lag 36 weeks to lag 43 weeks when we assessed the association between PM_2.5 exposure and de novo HDPs for different exposure doses and lag weeks (Figure 2D).

We conducted two sensitivity analyses to assess the consistency of the sensitivity windows for PM₁ and PM_2.5 for de novo HDP hazard. First, using natural splines with 4 and 5 degrees of freedom for the lag constraint, we observed similar results, that is, the significant lag windows for PM₁ and PM_2.5 spanned the preconception period and the first trimester (Multimedia Appendix 4). Furthermore, we excluded patients with GH to explore the independent association between PM and PE. Multimedia Appendix 5 shows a result consistent with our primary analyses, that is, that the lag windows for PM₁ and PM_2.5 spanned the preconception period and the first trimester.

In the stratified analyses, we estimated the effects of PM on de novo HDPs across differences in employment status, parity, and maternal education level. In terms of PM₁, a significant sensitivity window for de novo HDPs was observed for employed, primiparous, and low educational status women (Figure 3). Similar findings are shown in Figure 4: women who were employed, primiparous, or had a low education level were more sensitive to PM_2.5 exposure across the preconception period and the first trimester.

This study found that the significant lag windows for PM₁ and PM_2.5 were from the 5th week before gestation to the 6th gestational week and the 1st week before gestation to the 6th gestational week, respectively, in terms of the de novo HDP hazard. The findings remained robust in sensitivity analyses, which reexamined the associations using different degrees of freedom and after excluding patients with GH. Stratified analyses indicated that pregnant women who were employed, primiparous, or had a lower education level were more sensitive to PM₁ and PM_2.5 exposure in the prepregnancy period and the first trimester.

In recent years, numerous studies have evaluated the association between air pollution and cardiovascular health during pregnancy. However, there has been no study to assess the adverse effect of preconception air pollution on gestational health in terms of de novo HDPs. We adopted a time-serial study design to explore the weekly effects of PM₁ and PM_2.5 from the preconception period to delivery. To our knowledge, this is the first study to focus on the detrimental association between PM during both the preconception and gestational periods and de novo HDPs. During the study period, the average exposure to PM_2.5 was 74.2 μg/m³, which is more than 2-fold higher than the recommended maximum level of 35 μg/m³ set by the National Ambient Air Quality Standards organization [27]. Moreover, exposure to PM showed seasonal fluctuations, peaking in winter and spring. Considering that our participants were mainly northern Chinese living in Beijing, the capital of China, the high concentration of PM can be attributed to central heating and a dense population.

In this study, the prevalence of de novo HDPs was 6.7% (1520/22,570). However, a lower prevalence (5%) of HDPs was reported in a recent study conducted in Shanghai [13]. In comparison with the Shanghai study, the pregnant subjects in our study were exposed to a higher concentration of PM pollution. Since both Beijing and Shanghai have populations of over 24 million and similar demographic compositions, the prevalence of HDPs is unlikely to have been influenced by differences in industrialization or urbanization between these two megacities. The difference in HDP prevalence might be partially explained by the higher PM exposure in the north due to the climate difference and use of central heating.

In our time-serial study, we observed that the women were sensitive to PM₁ and PM_2.5 during the preconception period and the first trimester. Our PM exposure measurements were at the week scale, which is more precise than previous studies that have used the trimester scale. Previous studies have made inconsistent findings on the association between fine PM and HDPs. Among 4 US studies, one in California [28] and one in Florida [14] observed that PM_2.5 exposure was associated with HDPs during pregnancy. However, studies conducted in Rhode Island [29] and New York [30] failed to observe significant associations. There are some potential explanations for these inconsistent results. First, the latitudinal difference may have influenced the temperature and humidity in the different regions. A recent national cohort study indicated that temperature was positively associated with the risk of HDPs during pregnancy [31]. Second, exposure measurement bias may have resulted in a false-negative association. Third, ethnicity differences can affect the association between PM and HDPs. A national cross-sectional study from 2007 to 2018 showed that there was a racial difference in the prevalence of HDPs in the United States [32]. Our study participants mainly lived in Beijing, and 94% were members of the Chinese Han population. Meanwhile, we described our method for high-precision air pollution measurement in a previous study [23]. Our results are consistent with those of the Shanghai study, which had a similar ethnicity and urbanization profile. Thus, climate, measurement, and ethnicity differences were well controlled in our study.

Furthermore, we used a DLNM model, which allowed an examination of the lag effect of PM on HDPs at the week scale, rather than the previously used trimester scale, to improve the precision of the search for the most sensitive gestational week. We also considered potential bias in the model parameters by testing the consistency of our results with different degrees of freedom. Based on the robust results from our sensitivity analyses, we consider that bias from model parameters was not likely. Additionally, our second sensitivity analysis excluding the patients with GE further showed that exposure to fine PM during the preconception period and first trimester was independently associated with the risk of PE. PE is one of the leading causes of adverse pregnancy outcomes, accounting for 10% to 15% of maternal deaths and 15% to 20% of preterm births [33,34]. Early identification of PE risk in clinical practice remains a difficult problem [35]. Our environmental evidence for an association between fine PM exposure and PE might indicate the need for preconception PE prevention to consider environmental hazards.

Considering the strong link with maternal characteristics and gestational blood pressure [36], we further investigated the effects of PM₁ and PM_2.5 on HDPs across different maternal characteristics, including employment status, parity, and education level. Among employed women, exposure to PM was significantly associated with the risk of HDPs. PM exposure also elevated the risk of HDPs in women who were primiparous and had a low education level, which indicates that women with these maternal characteristics have a higher risk of HDPs with exposure to PM.

There are several strengths of this study. First, to our knowledge, this was the first to adopt the Cox proportion model with a DLNM to estimate the week-scale sensitivity window to PM for HDPs in a Chinese population. HDPs are a major cause of adverse maternal and prenatal outcomes [37]. However, the identification of risk factors for HDPs has been limited due to complex confounders. Previous studies primarily explored the association between PM and HDPs in a specific trimester [38,39]. Our study analyzed the specific gestational weeks and improved the precision of the sensitivity window. Second, this was the first study to cover the effect of air pollution, starting from the preconception period, on de novo HDPs. Our findings reconfirmed the necessity of focusing on air pollution before pregnancy in terms of de novo HDPs. Third, the large sample size of this study made it possible to determine the effects of PM on subtypes of HDP in stratified analyses.

Some limitations must be noted. First, all the participants were residents of Beijing. The lag windows for PM₁ and PM_2.5 should be tested in different regions to increase generalizability. Second, PM is a mixture of types of air pollution. The effects of specific PM components should be further explored in future studies.

Exposure to PM₁ and PM_2.5 was associated with the risk of de novo HDPs. The significant lag windows for PM exposure were between the preconception period and the first trimester. Women who were employed, had a lower education level, or were primiparous were more vulnerable to the effects of PM exposure on the risk of HDPs; more attention should be paid to these groups for the early prevention of de novo HDPs.