The preliminary screening phase encompassed pregnant women who underwent antenatal examinations at Jinhua Maternal and Child Health Hospital, located in Zhejiang Province, China, between January 2017 and December 2019. A total of 6,090 pregnant women underwent antenatal checkups during the data collection period. Demographic information (e.g., maternal age, pre-pregnancy weight and height, educational attainment, family economy, nationality, and family address), lifestyle information (e.g., folic acid intake), and health status information (including medical and medication histories) were collected. Pregnant women were excluded if they resided outside Jinhua City, Zhejiang Province (156 cases), had missing data on their delivery date (869 cases), were carrying twin or multiple pregnancies (77 cases), had missing data on PROM (118 cases), had a gestational age of less than 20 or more than 47 weeks (3 cases), or experienced stillbirths (3 cases). Finally, 4,864 participants were included in the study (Supplementary Figure S1).

PROM is defined as spontaneous rupture of fetal membranes prior to delivery, irrespective of gestational age. In our study, PROM was defined after delivery. The gestation period in this study was from the last menstrual period to the date of labor.

To assess exposure, PM_2.5 samples were collected continuously for 7 days each month, from 10:00 AM to 9:00 AM the following day, using quartz fiber filters. Sampling was conducted using a medium-volume air sampler (model MSV6) manufactured by LECKEL in Germany, operating at a flow rate of 38.3 liters per minute. The metals bound to PM_2.5, such as aluminum (Al), arsenic (As), beryllium (Be), cadmium (Cd), plumbum (Pb), manganese (Mn), chromium (Cr), hydargyrum (Hg), nickel (Ni), selenium (Se), and thallium (Tl), were analyzed using inductively coupled plasma-mass spectrometry (ICP-MS) equipment from Thermo Fisher Scientific (16). The mean value of PM_2.5-bound metals for each month for each pregnant woman was calculated on the basis of the data collected on metals.

In our study, we categorized each participant’s exposure into four distinct periods: (1) the entire gestational period, encompassing the entire duration of pregnancy; (2) the first trimester, spanning from the beginning of pregnancy (week 1) to the end of the 12th week; (3) the second trimester, covering the period from the 13th to the 27th week of pregnancy; and (4) the third trimester, extending from the 28th week until the moment of delivery.

Descriptive analysis of demographic characteristics and exposures of the entire population. The metal concentration is converted to a standard normal. The relevance between each metal is estimated via Spearman correlation.

We use the Cox regression to evaluate the risk ratio of PM_2.5-bound metals to PROM using the number of gestation days as the time scale and use the Schoenfeld residual method to test the proportional hazards hypothesis. And we applied the False Discovery Rate (FDR) to consider the potential for type I error due to multiple comparisons. Covariates, including PM_2.5 (continuous variable), maternal age (continuous variable), pre-pregnancy body mass index (BMI) (continuous variable), nationality (Han nationality, others, unknown), parity (nulliparous, multiparous, unknown), family economy (high, middle, low, unknown), educational attainment (junior high school or below, senior high school, college or above, unknown), sex (male, female, unknown) and season of conception (spring, summer, autumn, winter), were selected on the basis of existing studies (16–18). For missing values of covariates, we interpolated continuous covariates using the mean value and directly categorized the missing categorical variables into a new class. Moreover, the E value was used to evaluate the potential impact of unmeasured confounders on our study’s outcomes. The E value is defined as the smallest strength of the correlation in a hazard ratio (HR) scale, which depends on the measured covariate, whereas unmeasured confounding factors require an explanation of their association in terms of both treatment and outcome. The E value is concerned with the magnitude of the association of a confounder that may produce the same confounding bias as the observed association of the treatment outcome. We define the occurrence of PROM with a gestational age less than 37 weeks as preterm PROM, and that with a gestational age of 37 weeks or more as term PROM. Stratified analyses of term PROM and preterm PROM were performed to explore the effects of PM_2.5-bound metals on PROM at different gestational ages. To explore the exposure-response relationship between PM_2.5-bound metals and PROM, we plotted restricted cubic spline curves with 3 knots at the 10th, 50th and 90th percentiles.

We used the group weight quantile sum model (GWQS), quantile g-computation (Q-gcomp), and Bayesian kernel machine regression (BKMR) to assess the joint effect of PM_2.5-bound metals on the PROM. GWQS may consider multiple groups of exposures to estimate as many associations with outcomes as possible for different magnitudes and directions (19). In our study, we assessed the overall effect of PM_2.5-bound metals on PROM by dividing them into positive and negative groups on the basis of the results of Cox regression. Like WQS regression, Q-gcomp is characterized by simplicity of interpretation and ease of calculation, but it does not assume directional homogeneity (20). BKMR makes use of kernel regression, leveraging the Bayesian algorithm, to address high-dimensional parameter spaces as well as nonlinear and nonadditive data based on the Bayesian algorithm, thus providing flexibility in modeling the combined effects of the mixtures (21). BKMR was implemented through the “bkmr” package in R. Comparison of the three models can be found in Supplementary Table S1.

In the sensitivity analyses, (a) we tested for correlations via the accelerated failure time model. (b) We used Cox proportional risk models stratified by several basic characteristics to identify susceptible populations, including maternal age (≤24, 25–29, ≥30), infant’s sex (male, female), pre-pregnancy body mass index (BMI) (<18.5, 18.5–23.9, ≥24), parity (nulliparous, multiparous), educational attainment (junior high school or below, senior high school, college or above). (c) We stratify by some basic characteristics on the basis of grouped WQS to examine the robustness of the GWQS results. All the analyses were conducted with R version 4.1.3.

Table 1 summarizes the characteristics of the study population. A total of 4,864 participants with 194 (4.0%) PROM cases were included. The mean age and pre-pregnancy BMI of the mothers were 29.39 and 21.41, respectively. Most of the study subjects had a median family economy (95.5%). More than 50% of mothers had a university education or above (58.8%). The vast majority of mothers were Han Chinese (96.3%).

Table 2 describes the distribution of PM_2.5-bound metal exposure levels among the study population during pregnancy. The average concentrations of Al, As, Be, Cd, Pb, Mn, Cr, Hg, Ni, Se, and Tl throughout pregnancy were 27.74 ng/m³, 3.16 ng/m³, 0.21 ng/m³, 15.43 ng/m³, 29.63 ng/m³, 15.67 ng/m³, 3.24 ng/m³, 2.75 ng/m³, 3.80 ng/m³, 6.66 ng/m³ and 0.71 ng/m³, respectively. Supplementary Figure S2 shows Spearman correlation coefficients between PM_2.5-bound metal exposure throughout pregnancy. Al and Be, Cd, Pb, Ni, Se, As and Pb, Mn, Cr, Be and Cd, Pb, Cd and Pb, Ni, Se, Pb and Mn, Mn and Cr, Ni and Se were strongly correlated.

The associations between individual PM_2.5-bound metals and preterm rupture of membranes (PROM), along with their corresponding E values for each trimester and the entire pregnancy, are presented in Table 3. The results of proportional hazards hypothesis test by the Schoenfeld residual method showed that all models met the equal-proportionality assumption (p > 0.05) (Supplementary Figure S3). We can observe that eleven PM_2.5-bound metals were associated with the PROM, but this association varies according to the exposure window. During the entire pregnancy, Al, Cd, Pb, Cr, Ni, Se, and Tl increased PROM risk (HRs ranging from 1.40 to 1.87). Be decreased risk (HR = 0.79). In the first trimester, Al, As, and Pb increased risk (HRs from 1.49 to 2.06). In the second trimester, Cd, Ni, and Se elevated risk (HRs from 1.36 to 1.44). In the third trimester, Pb, Cr, and Tl increased risk (HRs from 2.04 to 4.12), while Mn and Hg decreased risk (HRs of 0.47 and 0.63). E values were in the range of 1.00 ~ 7.71, which indicated that an unmeasured confounder beyond those included in our models would need to have a hazard ratio of ≥7.71 relative to the outcome to completely diminish the observed effects in our study.

The results in Table 4 show most PM_2.5-bound metals were more strongly associated with preterm PROM than term PROM. Pb, Se and Tl were significantly linked to preterm PROM, with HRs of 2.77 (95% CI: 1.02, 7.54), 2.94 (95% CI: 1.07, 8.06) and 2.35 (95% CI: 1.11, 4.99), respectively.

The exposure-response relationship between individual PM_2.5-bound metals and PROM was fitted by a restricted cubic spline function with a knot of 3 knots. As, Mn and Hg showed a linear relationship with PROM (p > 0.05), with a monotonically increasing trend for Tl. Al, Cd, Pb, Ni, and Se showed an approximately U-shaped relationship with PROM, with the risk of PROM decreasing and then increasing as the concentration of pollutants increased. As Mn and Hg exhibited an inverted U-shaped relationship, the risk of PROM increased and then decreased with increasing pollutant concentration, whereas Be and Cr tended to increase within a certain range (Figure 1).

The mixed analysis results based on the GWQS model, Q-gcomp and BKMR are shown in Figure 2. Throughout the gestational period, positive and negative groups were generated according to the results of Cox regression. In the GWQS model, a mixture of metals in the positive direction increased the risk of PROM by 3% (95% CI: 2, 4%), and the positive weights associated with PROM were driven primarily by Pb, followed by Cr, Ni, Tl, As, Al, Se, and Cd (Figure 2A). In the negative direction, the PROM risk increased by −1% (95% CI: −2, 0%), and Be is the most heavily weighted in the negative direction, followed by Hg and Mn (Figure 2B). The results of Q-gcomp show that for every quarter increase in all the metal mixtures, the incidence of PROM increased by 5% (95% CI: 2, 8%), where the positive direction was driven mainly by Pb and the negative direction was driven mainly by Al (Figure 2C). The result of BKMR reveals that Ni had the largest PIP (1.000), followed by Al (0.954) (Figure 2D). Figure 2E illustrates the nonlinear relationships for each mixture metal. The joint effect of PM_2.5-bound metals on PROM was greater at the 60th percentile and above, and this effect was statistically significant (Figure 2F).

Supplementary Table S1 and Supplementary Figures S4, S5 show the results of the sensitivity analyses. Supplementary Table S1 shows the association between PROM and individual PM_2.5-bound metals throughout gestation on the basis of accelerated failure time models. The results reveal that maternal exposure to Al, Cd, Pb, Cr, Ni, Se, and Tl throughout gestation; to Al, As, Pb and Tl in early pregnancy; to Cd, Ni, and Se in mid-pregnancy; and to Pb, Cr and Tl in late pregnancy increased the incidence of PROM.

Supplementary Figure S4 shows the results of the stratified analyses based on Cox proportional risk regression. Exposure to Se, Ni, and Tl during pregnancy and mothers aged <30 years were associated with increased PROM risk. In addition, exposure to Cr and Mn increased the susceptibility to develop PROM in pregnant women aged ≤24 years and ≥30 years. For Al, Be, Pb and Cd, the susceptibility period was 25–29 years. There were differences in susceptibility to different metal exposures according to the infant’s sex. Compared with males, females were more susceptible to most metals. Pregnant women with pre-pregnancy BMIs between 18.5 and 23 kg/m² were susceptible to the occurrence of PROM. Exposure to Al, Cr, Se, Tl, and Ni during pregnancy was associated with a higher PROM risk in primiparous women. The risk of PROM was greater when the mother’s educational level was college and above. Supplementary Figure S5 shows the results of the stratified analyses based on GWQS. The results show that the GWQS results were almost robust.