Exploration of the preterm birth risk-related heat event thresholds for pregnant women: a population-based cohort study in China

Summary

Background

Heat events increase the risk of preterm birth (PTB), and identifying the risk-related event thresholds contributes to developing early warning system for pregnant women and guiding their public health response. However, the event thresholds that cause the risk remain unclear. We aimed to investigate the effects of heat events defined with different intensities and durations on PTB throughout pregnancy, and to determine thresholds for the high-risk heat events.

Methods

Using a population-based birth cohort data, we included 210,798 singleton live births in eight provinces in China during 2014–2018. Daily meteorological variables and inverse distance weighted methods were used to estimate exposures at a resolution of 1 km × 1 km. A series of cut off temperature intensities (50th–97.5th percentiles, or 18 °C–35 °C) and durations (at least 1, 2, 3, 4 or 5 consecutive days) were used to define the heat events. Cox regression models were used to estimate the effects of heat events on PTB in various gestational weeks during the entire pregnancy, and event thresholds were determined by calculating population attributable fractions.

Findings

The hazard ratios of heat event exposure on PTB ranged from 1.07 (95% CI: 1.00, 1.13) to 1.43 (1.15, 1.77). Adverse effects of heat event exposure were prominently detected in gestational week 1–4, week 21–32 and the four weeks before delivery. The heat event thresholds were determined to be daily maximum temperature at the 90th percentile of the distribution or 30 °C lasting for at least one day. If pregnant women were able to avoid the heat exposures from the early warning systems triggered by these thresholds, approximately 15% or 17% of the number of total PTB cases could have been avoided.

Interpretation

Exposure to heat event can increase the risk of PTB when thermal event exceeds a specific intensity and duration threshold, particularly in the first four gestational weeks, and between week 21 and the last four weeks. This study provides compelling evidence for the development of heat-health early warning systems for pregnant women that could substantially mitigate the risk of PTB.

Funding

National Key R&D Program of China (No. 2018YFA0606200), National Natural Science Foundation of China (No. 42175183), Sanming Project of Medicine in Shenzhen (No. SZSM202111001).

Research in context.

Evidence before this study

We searched PubMed for studies published until December 2022, using the key terms “preterm birth”, “adverse birth outcomes”, “climate change”, “heat event”, “heatwave” and “threshold”. Previous epidemiologic studies have reported that maternal exposure to heat event may increase the risk of adverse birth outcomes, among which, preterm birth (PTB) remains a matter of great public health concern. Several studies indicated that heat exposure lasting for a few days could exacerbate the risk of PTB, and the risk may increase with increased intensity or duration of heat, especially in the week before delivery. Heat-related burden on PTB can be mitigated by effective heat action plans which include early warning systems and emergency public health measures. Pregnant women are more vulnerable to hot weather than the general population, but public health measures do not specifically target pregnant women, mainly because the event thresholds that cause the risk is not clear. Therefore, identifying PTB-risk related heat events thresholds that would maximize the benefits from a heat health early warning system constitutes a research priority, especially in the context of climate change.

Added value of this study

This study identifies heat event thresholds based on the impacts of maternal exposures to heat events variously defined on the risk of PTB in China. In this large-scale and population-based prospective cohort study, we provide an innovative approach for determining health-related heat event thresholds for pregnant women. In addition to evaluating relative risks of heat exposures on PTB in different gestational weeks and detecting susceptible windows, the approach takes early warning efficiency into account by calculating attributable risks. Hence, the early warnings triggered by the thresholds could not only remind pregnant women when and to what extent to take heat protection measures, but also produce the maximal health benefits.

Implications of all the available evidence

In the context of climate change, extreme heat will become more frequent and intense, further threaten maternal and child health and cause intergenerational effects. Establishing heat early warning system for pregnant women and guiding them to take timely and effective heat protective measures will greatly mitigate the heat-related risk. Our findings provide strong evidence of increased risks of preterm birth due to heat exposures exceeding obvious intensity and duration threshold, and provide a scientific basis for identifying risk-related threshold for the development of heat-health early warning systems that target pregnant women and mitigate the risk.

Introduction

Climate change is unequivocal and results in a significant increase in exposures to heat events and subsequent adverse health impacts.^1,2 Due to physiological and anatomical changes during pregnancy, pregnant women may be more susceptible to heat events.³ In recent years, the heat impacts on adverse birth outcomes have gradually become a big concern.⁴ Among them, preterm birth (PTB) remains a major global public health problem, and imposes a substantial burden on newborns and children as well as their lifelong health.⁵ There are about 15 million PTBs in 2014 worldwide, while China ranks second with 1.17 million PTBs (6.9% of all births).⁶ Previous studies suggested that maternal heat exposures could lead to dehydration or increased secretion of prostaglandin F2α and oxytocin, further reducing placental blood flow and shifting fetal metabolic pathways from anabolic to catabolic, which may ultimately induce preterm labor.^7,8 Much epidemiological evidence has extensively demonstrated that heat exposures are detrimental for PTB.⁴ Furthermore, several latest studies have shown heat exposure lasting for several consecutive days, which was usually defined as a heatwave event, can significantly exacerbate the PTB risk.9, 10, 11 In the context of climate change, the frequency and intensity of heat event are expected to increase, and the temperature increase rate in China is higher than that of the global average level in the same period, which will accordingly add to the existing PTB burden.

Heat-related burden on PTB can be mitigated by effective heat action plans which include early warning systems and emergency public health measures.^12,13 An important consideration when developing early warning system is the identification of an appropriate heat event thresholds of intensity and duration, which refers to an extended period of time with unusually hot weather conditions that potentially begins to harm human health.¹⁴ In most countries, pregnant women and adverse birth outcomes such as PTB were not typically considered to be a high-risk group or outcome in their design of heat early warnings.¹⁵ Most early warning systems have been triggered by thresholds based on a statistical-meteorological criterion,¹⁶ while a few have taken the association between heat event and the risk of mortality or morbidity into account.17, 18, 19, 20, 21, 22, 23, 24 The above thresholds were generally determined based on the optimal model principle, which usually indicated that the model best interpreted the relative risk of heat related health outcomes with the fewest free parameters.

Taking the effectiveness of early warning systems into account, it is desirable to identify heat event thresholds that activate early warnings based on the health benefits optimization from such warnings.²⁵ Attributable risk is an informative indicator for quantifying the health benefit from early warning systems that are triggered by thresholds,²⁵ which may help provide timely early warning information for pregnant women to minimize the health hazards. Besides, as a pregnancy can span months and the vulnerability to heat events may vary over the duration of the pregnancy, thresholds may be more complex and differ by exposure windows.⁹ Thus, it is crucial to identify thresholds based on critical exposure windows to provide more accurate information for responses to heat events and mitigate adverse health impacts.

In this study, we conducted a population-based birth cohort in China to estimate the risks of PTB associated with various definitions of heat event in every four gestational weeks, and to identify critical exposure windows. Besides, we estimated attributable risks and further determined PTB risk-based heat event thresholds.

Methods

Study population

Since 2013, the National Center for Women and Children's Health, Chinese Center for Disease Control and Prevention has launched the National Maternal and Newborn Health Monitoring Project to prospectively monitor and improve the health of mothers and infants in 16 monitoring counties in eight provinces (Liaoning, Hebei, Hubei, Hunan, Fujian, Guangdong, Sichuan and Yunnan) of China (Supplementary Figure S1).²⁶ Pregnant women establishing the prenatal care handbook in the monitoring areas were prospectively monitored throughout their pregnancy from March 6, 2013, at the first antenatal care, until the delivery date of December 31, 2018. A total of 271,720 pairs of “pregnant women and baby” were monitored.

Since the monitoring project was in the process of continuous improvements at the initial stages of the study period, we established a population-based cohort including 222,447 pairs of mothers and their singleton live births from March 11, 2014 to December 31, 2018 when the database of the project was relatively complete. According to the exclusion criteria, after excluding stillbirths (n = 314), multiple births (n = 6803), births with abnormal gestational age (n = 123) and maternal age (n = 4409), 210,798 singleton live births remained in the study sample. More details about the study population and the information we collected during the following up have been described in our previous work.²⁶

The study was approved by the Ethical Committees from the National Center for Women and Children's Health (NCWCH) (No. FY2015-007) and the School of Public Health, Sun Yat-sen University (No. 2021-014). Informed consent was provided by all participants at enrollment in the cohort.

Outcome

Preterm birth (PTB) was defined as delivery before 37 completed weeks of gestation. Gestational age was objectively determined for all births by combining ultrasound examination and mother-reported last menstrual period (LMP). When available, ultrasound estimates were used; otherwise, the LMP was used. In this study, there were 202,470 (96%) births determining gestational age by ultrasonography, 8328 (4%) births by LMP.

Exposure assessment

Meteorological and air pollution data were derived from the China Meteorological Data Service Center (http://data.cma.cn/) and China National Environmental Monitoring Center (http://www.cnemc.cn/), respectively. Inverse Distance Weighting (IDW) was applied to interpolate the exposure of daily temperature, relative humidity and air pollution at a resolution of 1 km × 1 km for each woman by linking the home address (Supplementary Methods).²⁶

We used the daily maximum temperature as heat event indicator, and higher temperature thresholds corresponding to the 50th–97.5th percentiles of temperature during the study period and lasting for at least 1–5 consecutive days were used to define heat events (Supplementary Methods). Besides, we also used a range of temperature cut-off points for absolute temperature (18 °C–35 °C) in conjunction with the above durations to define events (Supplementary Methods).

Heat event exposure was assigned as a dichotomous variable (yes/no) in the week before delivery as in previous studies.¹⁰ Then the exposure in every four weeks from the conception to the delivery was assigned in the same way.⁹ As about half of PTBs occurring before gestational week 33–36, we evaluated the exposure in the four weeks before delivery. Therefore, the exposure windows were gestational week 1–4, 5–8, 9–12, 13–16, 17–20, 21–24, 25–28, 29–32 and four weeks before delivery.

Statistical analyses

Analyses strategies

In line with previous studies, we first defined heat event based on temperature distributions in the warm season (May–October), and included mothers giving birth in the warm season only to evaluate the effects in the week before delivery.^10,11 However, we could not identify event thresholds, given that the event with the lowest intensity and the shortest duration (50th-D1) was associated with the largest adverse effect. And a decreasing effect was observed when the intensity or duration of heat event continued to increase. Then, we created heat event definitions based on temperature distributions for the full year and evaluated the effects in mothers giving births in the full year. A pattern of higher effects with increased intensity or duration was detected, suggesting that it may be feasible to identify thresholds in this way. Further, we evaluated effects of heat events defined with different intensities (the 50th–97.5th percentiles of temperature, or 18 °C–35 °C) and durations (at least 1–5 consecutive days) in gestational week 1–32 and four weeks before delivery and explored critical exposure windows and thresholds.

Heat event impacts

We followed up the pregnant women from the date of conception to the end of delivery and recorded the gestational age at delivery. A time-to-event framework was used for the main analysis, and a Cox regression was applied to evaluate the effects of heat events in various gestational weeks on the risk of PTB. An important assumption of a typical Cox regression is that the hazard ratios for a given variable should be independent of time. However, in this study, environmental exposures such as relative humidity and air pollutants are time-varying, and their hazard ratios are no longer proportional. Thus, an extended Cox regression with time-varying variables was used in the study, which allows nonproportional hazards. In the model, we included time to PTB as the outcome and gestational week as the temporal unit. Besides, term births were censored at week 37 and the maternal residence was included as a random effect to account for spatial clustering of PTB. Exposure was assigned using the week before delivery, gestational week 1–4, 5–8, 9–12, 13–16, 17–20, 21–24, 25–28, 29–32 and four weeks before delivery. For each heat event exposure in each exposure window, we fitted the Cox regression model separately. The effect of each heat event definition is reported as hazard ratios (HRs) and 95% confidence intervals (95% CIs).

Covariates included in the model were selected a priori as potential confounders and well documented risk factors of PTB. The following time independent covariates were adjusted for: maternal age, maternal education (≤6, 7–9, 10–12, or >12 years), pre-pregnancy body mass index (<18.5, 18.5–23.9, >23.9), parity (primiparous, multiparous), season of conception (spring, summer, fall, winter) and infant sex (male, female). The Adequacy of Prenatal Care Utilization Index (inadequate, intermediate, adequate, adequate plus) was also included in the model, which was the ratio of actual prenatal care visits to recommended prenatal care visits calculated by the recommendations of the Guidelines for Pre-pregnancy and Pregnancy Health care (2018) in China.^27,28 Besides, time-varying variables were adjusted for in the model, including exposures to relative humidity, air pollutants, average gestational temperature, heat or cold event.⁹ Due to the high correlations between PM_2.5 and other air pollutants except for O₃, we only included PM_2.5 and O₃ in the entire pregnancy in the model to avoid collinearity. Additionally, when we explored the effects of the heat events in one week or four weeks before delivery, we controlled for average gestational temperature except for the last week or the last four weeks as a natural cubic spline with three degrees of freedom.¹⁰

Pregnant women may experience both heat event and cold event during the pregnancy. To explore heat event effects in a certain exposure window of each four gestational week, we excluded births delivered before that window and controlled for heat or cold effects in other exposure windows.⁹ Exposure to heat event was included in the models as a binary variable. Cold event exposure was defined as a daily maximum temperature lower than the corresponding percentile of temperature distribution (e.g., if a heat event was defined as a daily maximum temperature above the 90th of the distribution, then cold was defined as a daily maximum temperature lower than the 10th of the distribution) and included in models as a binary variable.

Heat events thresholds

Based on the estimated risks of PTB associated with heat events defined by various relative temperature thresholds (50th–97.5th percentiles) and durations (at least 1–5 consecutive days) in gestational week 1–4, 5–8, 9–12, 13–16, 17–20, 21–24, 25–28, 29–32 and four weeks before delivery, we defined the high-risk threshold as the 80th percentile of risk distributions. Any HR above the threshold was identified as a high risk, otherwise it was a low risk. Then, we calculated the proportion of heat event definitions with a high PTB risk to all heat event definitions in each four gestational weeks, and further identified critical exposure windows.

For each critical window, we identified all possible heat event thresholds, that were the definitions with the lowest intensity or duration and meanwhile corresponded to high risks of PTB. Then we calculated the corresponding population attributable fraction (PAF), which was an index of attributable risk. The Population Attributable Fraction (PAF) is the proportional reduction of health outcomes in population that would occur if exposure to a risk factor was reduced to an alternative ideal exposure scenario.²⁹ For a binary exposure in a cohort study, the formula of PAF is as follows:

$$\text{PAF}=\frac{{\mathrm{P}}_{\mathrm{e}}\times (\text{RR}-1)}{{\mathrm{P}}_{\mathrm{e}}\times \left(\text{RR}-1\right)+1}$$

where, P_e indicates the proportion of the population exposed to risk factors, which is the proportion of the pregnant women exposed to heat events. RR is a relative risk. For a cohort study, the HR can be substituted for the RR. Therefore, the HRs were used to calculate PAFs.²⁹

We selected an appropriate threshold with the maximal PAF for each critical window. Further, for each selected threshold, we calculated sum of PAF in all critical windows. In order to optimize the health benefits, the threshold corresponding to the maximal sum of PAF in all critical windows was determined as the appropriate early warning system threshold.²⁵ It indicates the maximal reduction in PTB that would occur if pregnant women avoided heat event exposures after an early warning system was activated.

Sensitivity analyses

Sensitivity analyses were conducted. Based on the hazard ratios distribution of the effects of heat events on preterm birth with various definitions in nine exposure windows, we divided hazard ratios above the 70th percentile of the distribution into high risk, and the others into low risk; or above the 90th percentile of the distribution into high risk, and the others into low risk. Then we determined the heat event threshold using the method above. Besides, we still regarded hazard ratios above the 80th percentile of the distribution as high risk, and the others as low risk. Then we used the optimal model principle to identify the heat event threshold based on the minimum Akaike Information Criterion (AIC).

This study is reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guideline (STROBE-checklist). The R software (version 3.4.2) with the “survival” package (R Foundation for Statistical Computing, Vienna, Austria) was used to fit all models.

Role of the funding source

The funding source had no role in study design, data collection, data analysis, data interpretation, or manuscript preparation. The corresponding authors had full access to all data in the study and had final responsibility for the decision to submit for publication.

Results

Supplementary Table S1 shows the summary characteristics of the study population. There were 210,798 singleton live births, including 8587 (4.1%) PTBs, during the study period. Mothers with older ages and higher pre-pregnancy BMI were more likely to have PTBs. Compared with term births, PTBs were observed more frequently among multiparous mothers, male babies, mothers with education of college and above, and conception in summer. The proportion of women with an adequate pregnancy health care was 32.4% among PTBs, which was higher than that among term births (19.4%).

Summary statistics and distributions of temperatures for all study sites are shown in Supplementary Tables S2 and S3. Average daily maximum temperature varied from 15.8 °C (±12.7 °C) to 26.4 °C (±7.1 °C) in the study areas in the full year and from 25.2 °C (±3.1 °C) to 31.6 °C (±3.4 °C) in the warm season, respectively. When temperature varied from the 50th to 97.5th percentile of the temperature distributions, average absolute temperature ranged from 23.1 °C to 34.3 °C, and from 28.7 °C to 34.9 °C in the areas, respectively.

Supplementary Figure S2 shows annual heat event days with event defined by different intensities and durations. When heat event varied from 90th-D2 to 97.5th-D5, the proportion of heatwave days in the full year ranged from 9.5% to 1.4%. The proportions of PTBs in pregnant women exposed and unexposed to heat events during their first four gestational weeks showed that the proportion in exposure group increased when the intensity or duration of heat event increased (Supplementary Table S4). However, the highest proportion was not related to exposure to the most intense or longest heat event.

We first explored the effects of heat event exposure in the week before delivery using common definitions and adjusting for potential covariates as previous studies did (Fig. 1). Adverse effects were found for all cities combined and also for several individual cities. Larger effects were detected for mothers giving birth in the warm season compared to all births in the whole study period. For instance, when defined heat event as 90th-D2, the most commonly used definition in previous studies, the corresponding absolute temperatures were 33.3 °C and 32.1 °C for warm season and full year, and the HRs were 1.19 (95% CI: 1.09, 1.29) and 1.09 (95% CI: 1.01, 1.16), respectively. Besides, the risk of PTB decreased as the intensity of heat event increased in the warm season. However, the risk tended to increase with increasing intensity or duration in the full year.

Fig. 1.

Hazard ratios of preterm birth associated with heat events in the week before delivery.

Abbreviation: nth-DN, heat event defined as maximum temperatures above the nth percentile of each city-specific temperature distribution lasting for at least N days. The averaged absolute temperatures corresponding to the 85th, 90th, 95th percentile of city-specific temperature distribution in all study areas in warm season were 32.58 °C, 33.31 °C, 34.23 °C, respectively, and in full year were 30.93 °C, 32.06 °C, 33.34 °C, respectively.

Then we explored the effects in the week before delivery with more heat event definitions, and we found more substantial differences in the patterns of the effects between warm season and full year (Fig. 2). For births in the warm months, we didn't capture a heat event threshold because the event with the lowest intensity and the shortest duration (50th-D1, 28.7 °C-D1) was associated with the largest adverse effect. However, for births over the whole study period, there appeared to be a heat event threshold for the effect on PTB, because the effects increased with increasing intensity or duration of event.

Fig. 2.

Preterm birth risk map of heat events with various event definitions in the week before delivery.

Abbreviation: HR, Hazard ratios. Based on the hazard ratios distribution of the effects of heat events on preterm birth with various definitions, we divided hazard ratios into low risk, medium low risk, medium-high risk and high risk for each 25% interval of the distribution.

We further evaluated the effects of heat events with various intensities and durations during the entire pregnancy and drew the risk map (Fig. 3). The risks of PTB associated with exposures to heat events across the pregnancy varied with different events definitions, and the high PTB risk threshold corresponding to the 80th percentile of the risk distribution was at an HR of 1.056 (95% CI: 1.011, 1.143). The susceptibility to heat event also differed with various exposure windows. Pregnant women appeared to be susceptible to heat event in gestational week 1–4, and then again from mid-pregnancy (gestational week 21–24) to late pregnancy (gestational week 29–32), and also in the four weeks before delivery. Depending on the above critical windows and event definition used, HRs ranged from 1.07 (95% CI: 1.00, 1.13) to 1.43 (1.15, 1.77).

Fig. 3.

Preterm birth risk map of heat events with different heat event definitions in each gestational month.

Abbreviation: Pre-4 weeks, four weeks before delivery. HR, Hazard ratios. Based on the hazard ratios distribution of the effects of heat events on preterm birth with various definitions, we divided hazard ratios above the 80th percentile of the distribution into high risk, and the others into low risk.

Based on the above critical windows, we identified possible heat event thresholds, which were the heat events defined with the lowest intensity or duration and meanwhile with significant high risks of PTB (HR > 1.056) (Table 1). The possible intensity of thresholds ranged from 70th (28 °C) to 97.5th (34 °C) percentile of the temperature distributions, and the duration of thresholds varied from one to five days. Corresponding to these thresholds, PAFs varied from 1.44 (95% CI: 0.08, 2.85) to 6.47 (95% CI: 1.94, 11.17). Then the thresholds with the maximal PAF were selected for each critical window, and they ranged from 80th-D5 to 97.5th-D2. Further, the threshold corresponding to the maximal sum of PAF in all critical windows was determined as the appropriate early warning threshold, which was 90th-D1. The sum of PAF was 15.2% (95% CI: 5.0%, 25.6%), indicating that approximately 15.2% of the number of total PTB cases could be prevented if pregnant women received early warnings and took measures to avoid exposure to the heat event.

Table 1.

Preterm birth risk-related relative heat events thresholds.

Susceptible windows	Possible thresholdsa	HRb (95% CI)	PAF (%) (95% CI)	Thresholdsc for windows
Week 1–4	90th-D1	1.07 (1.01, 1.15)	2.60 (0.18, 5.00)	80th-D5
	85th-D3	1.07 (1.00, 1.13)	2.08 (0.06, 4.13)
	80th-D5	1.11 (1.04, 1.19)	3.25 (1.11, 5.44)
Week 21–24	97.5th-D2	1.29 (1.19, 1.39)	2.73 (1.82, 3.69)	97.5th-D2
	95th-D3	1.19 (1.07, 1.32)	1.90 (0.74, 3.15)
	92.5th-D4	1.19 (1.01, 1.41)	2.24 (0.17, 4.58)
Week 25–28	90th-D1	1.21 (1.06, 1.39)	6.47 (1.94, 11.17)	90th-D1
	85th-D4	1.22 (1.10, 1.34)	5.03 (2.43, 7.73)
	80th-D5	1.15 (1.03, 1.28)	3.79 (0.69, 7.01)
Week 29–32	85th-D1	1.15 (1.04, 1.26)	5.44 (1.60, 9.31)	85th-D1
	80th-D2	1.14 (1.06, 1.24)	5.37 (2.15, 8.69)
	75th-D3	1.13 (1.03, 1.23)	4.74 (1.18, 8.38)
	70th-D5	1.13 (1.02, 1.25)	4.82 (0.81, 8.88)
Pre-4 weeks	92.5th-D3	1.08 (1.00, 1.15)	1.44 (0.08, 2.85)	80th-D5
	90th-D4	1.10 (1.02, 1.17)	1.79 (0.42, 3.23)
	80th-D5	1.12 (1.04, 1.20)	3.23 (1.15, 5.35)

Abbreviation: nth-DN, heat event defined as maximum temperatures above the nth percentile of each city-specific temperature distribution lasting for at least N days. Pre-4 weeks, four weeks before delivery.

HR, Hazard ratio. PAF, Population attributable fraction.

^aAll heat event definitions with the lowest intensity or duration and meanwhile corresponded to high risks of PTB.

^bThe reference group used for the HR estimates was participants who did not experience heat event exposure in the specific exposure window. HR is an expression of the hazard or chance of preterm birth occurring in the heat event exposure group as a ratio of the hazard of the preterm birth occurring in the non-exposure group. For example, when heatwave defined as 95th-D3, HR for heatwave exposure in gestational week 21–24 was 1.19, indicating that heatwave exposure increased the risk of PTB by 19% compared to those did not experience heatwave exposure.

^cThe selected threshold with the maximal PAF for each critical window.

The results on heat event thresholds and critical exposure windows for heat defined by absolute temperatures are presented in Supplementary Table S5 and Supplementary Figure S5. The critical windows were the same with those detected by relative temperatures. The thresholds varied from 28 °C-D1 to 35 °C-D3 in all critical windows. The heat event warning threshold was 30 °C-D1, indicating that approximately 16.6% (95% CI: 6.0%, 27.2%) of PTBs would be avoided if women were not exposed to the heat event.

In the sensitivity analyses, when we defined the high-risk threshold as the 70th or 90th percentile of the risk distribution, the risk threshold was 1.015 (95% CI: 0.921, 1.146) and 1.150 (95% CI: 1.024, 1.314), respectively. Both critical exposure windows of heat events on PTB were still gestational week 1–4, 21–32 and four weeks before delivery (Supplementary Tables S6 and S7). And the heat event warning thresholds were the same with the main results as 90th-D1 (Supplementary Figures S3 and S4). Besides, the threshold of heat event determined by the minimum AIC was 97.5th-D2 (Supplementary Table S8). Both intensity and duration were higher than the threshold (90th-D1) based on the maximum PAF.

Additionally, we removed all covariates from the model and evaluated the effects in the week before delivery. Compared with the adjusted estimate in the main result, the crude HR with only heat event exposure in the model increased slightly, and HRs were 1.09 (95% CI: 1.01, 1.16) and 1.10 (95% CI: 1.04, 1.16), respectively (Supplementary Table S9). Besides, when we respectively excluded covariates about gestational temperature exposure across the entire pregnancy except for the last week before delivery, maternal residence, season of conception, or the APNCU Index from the adjusted model, the magnitude of HR changed slightly, indicating that these covariates might be confounders.

Discussion

This study firstly examined the impacts of maternal exposures to heat events with various definitions on PTBs, and determined the event thresholds based on critical exposure windows and optimal health benefits in China. The risks of PTB increased with exposures to heat events in gestational week 1–4, 21–32, and the four weeks before delivery, which suggests that these are critical window periods for the heat related PTB. The heat event threshold was determined as daily maximum temperature at the 90th percentile of the city-specific distribution that lasts for at least one day or 30 °C for one day, and approximately 15% or 17% of the number of total PTB cases could be avoided if women avoided the heat exposures after activated early warning systems triggered by these thresholds.

Identification of appropriate health-related heat event thresholds is crucial for the development of heat-health early warning systems and public response measures. Several studies have identified heat event thresholds by examining mortality or morbidity risk due to event. For instance, Yang et al. and Gao et al. evaluated the risk of heat-related mortality risk in several major Chinese cities, and determined the threshold as 92.5% of the daily maximum temperature distribution lasting for at least three days, or from 30 °C for at least two days to 34.5 °C for at least three days, using the Akaike Information Criterion (AIC).^17,19 Xu et al. compared generalized cross validations of the models, and found that heat event lasting for two days were more detrimental, and that the 97th percentile of the daily mean temperature distribution might be a threshold of heat-related morbidity in Brisbane, Australia.²⁴

The above studies mainly relied on relative risk to estimate the health impacts, and determined thresholds based on optimization statistical models by estimating the relative goodness of fit across the models. However, a heat event threshold that best mitigates its health impacts on a population may not correspond to the highest relative risk. For example, although an extreme heat event may impose a greater relative risk, the attributable risk may be lower because they occur less frequently, and a smaller proportion of the population is exposed to the heat event. Therefore, the identification of heat event thresholds should be based on the attributable risks, which consider the proportion of the population exposed to event.²⁵

We have provided an innovative approach to identify heat event thresholds in this study based on the attributable risks, i.e., the population attributable fraction (PAF). It was calculated taking into account not only the relative risk of PTB but also the proportion of pregnant women exposed to the heat event. The threshold was determined as the daily maximum temperature at 90th percentile of distribution or 30 °C lasting for at least one day, lower than the current definition of heatwave recommended by the China Meteorological Administration (35 °C, at least 3 consecutive days). The threshold determined in our study may play an important role in better alerting pregnant women when and to what extent they should take proper precautions, as well as in informing policy makers who aim to optimize health benefits from early warning systems.²⁵

Similarly, a study in the United States estimated both population attributable fraction and relative risk of hospitalization for heat-related illness and dehydration to identify heat event threshold. It found that absolute risk was more appropriate to identify the threshold to activate early warning system and ultimately promoted greater alleviation of heat impacts.²⁵ Studies on the identification of heat event threshold based on the attributable risk remain limited, more related explorations are needed in future.

In China, heat-health early warning systems have been established in Shanghai, Harbin, Chongqing, Shenzhen, Nanjing and other pilot cities. However, the thresholds that activate the systems were mainly identified based on the heat events impacts on mortality or morbidity, and the population was only classified by gender or age.^30,31 The systems failed to identify pregnant women and other vulnerable groups, which led to inadequate interventions and reduced effectiveness.³⁰ Future studies need to collect more extensive data on health and socioeconomics before and after the early warning, and further assess the accuracy and stability of early warning.^31,32 By clarifying the effect of early warning, the system can be continuously optimized, and precise health risk intervention measures can be formulated.

Given that a pregnancy lasts for 9–10 months, it is important to account for critical exposure windows when determining heat event threshold. However, previous studies on heat event and PTB have predominantly focused on the exposure on the day of delivery to the week prior to delivery.^10,11,33 For example, cohort studies from China¹⁰ and the United States¹¹ evaluated the effects in the week before delivery, and found the risks increased with increased intensity and duration of heat events. Only a recent study from Australia attempted to explore the effects of heat events in each gestational month. Nonetheless, specific months of increased susceptibility were not identified, with HRs varying from 1.08 (95% CI: 1.00, 1.18) to 1.53 (1.41, 1.68).⁹

We regarded every four gestational weeks as a gestational month,⁹ and detected that gestational week 1–4, 21–32 and the four weeks before delivery were critical exposure windows. The HRs ranged from 1.07 (95% CI: 1.00, 1.13) to 1.43 (1.15, 1.77). These identified critical windows may provide more accurate and effective information for developing heat-health early warning systems and provide guidance to pregnant women by giving timely and precise interventions from a public health perspective. Besides, critical windows also shed light on designing experimental studies to explore the underlying mechanisms.

The mechanism of which heat event influences PTB is not completely understood. Previous studies suggested exposure to heat event could induce different types of PTB through multiple biological pathways. Maternal exposure to heat event could lead to preeclampsia³⁴ by damaging vascular endothelial cells and affecting the regulation of systemic vascular tone,³⁵ and further induce medically induced PTB. It could also restrict fetal growth by affecting nutrition and oxygen delivery from the placenta to the fetus,³⁶ and ultimately lead to medically induced PTB. Besides, heat event exposure may also play an important role in triggering spontaneous PTB by elevating the secretions of oxytocin³⁷ or stimulating rupture of the membranes.³⁸

There is evidence for potential mechanisms by which heat event affects PTB in different stages of pregnancy. Gestational week 1–4 is the period of the formation and implantation of fertilized eggs.39, 40, 41 Exposure to heat event in this period or even in three months before pregnancy may influence the quality of fertilized eggs.^42,43 Gestational week 21–27 belongs to the second trimester, when baby is in rapid-growth. An animal study in ewes found that heat event exposure in mid-pregnancy could affect the development of the placental vessels, and restrict placental weight by reducing mitotic activity.⁴⁴ Besides, heat event could affect fetal growth by restricting uterine blood flow and cellular growth, or even permanently limit the capacity of fetus to develop normally.⁴⁴ The above changes may ultimately contribute to early delivery.

Gestational week 28 to the four weeks before delivery belongs to the late pregnancy, when pregnant women need to deal with emotional symptoms and new physical symptoms,⁴⁵ and therefore may increase susceptibility to adverse environment. Exposure to heat event in late pregnancy has been extensively shown to be associated with an immediate elevated risk of delivery both in animals and in humans.^10,46 One possible mechanism is that heat event leads to an increase in the secretion of oxytocin and prostaglandin F2α (PGF2α), which play an important role in regulating the onset of delivery.⁷ Alternatively, heat event might lead to PTBs via cardiovascular stress.⁴⁷

The risk of PTB decreased with increased intensity or duration of heat event in the week before delivery in the warm season. The heat acclimatization may explain this pattern. The average temperature corresponding to the 50th percentile of temperature distributions in the warm season was about 29 °C, which may be sufficiently stressful to activate adaptation mechanisms.⁴⁸ In hot weather, humans usually adapt to heat better by physiological changes such as increasing total body water, expanding plasma volume, better sustaining and/or elevating stroke volume, reducing heart rate, improving ventricular filling and myocardial efficiency, enhancing skin blood flow as well as sweating responses.⁴⁸ Besides, pregnant women may also change their behaviors such as staying at home or turning on air conditioning to mitigate heat impacts.

Several limitations of this study should be noted. First, although we used city-specific heat events to assign individual exposure, it might not accurately reflect the real individual exposure due to lack of information on the maternal activity patterns, residence mobility, work address and its mobility, as well as use of air conditioning. Further, information on the presence of cervicovaginal or intrauterine infections, specific gestational complications, physical activity, dietary intake, employment, income levels and spouse was not available, thus we could not account for these potentially important modifying factors in the analyses.

In conclusion, this study provides evidence of increased risks of preterm birth due to maternal heat event exposures exceeding obvious intensity and duration threshold. Additionally, the first four gestational weeks, and week 21 to the last four weeks are critical windows for the impacts of heat events on preterm birth. In the context of climate change, our novel findings provide guidance for policymakers and service-providers to deal with the adverse birth outcomes from heat events, contribute to the development and implementation of heat-health early warning systems to target pregnant women and protect the health of newborns.

Contributors

All authors contributed to the research article and approved the final version. Cunrui Huang initiated this study and contributed to the methodological design; Aiqun Huang, Chunying Zhang, Jiangli Di and Meng Ren collected and managed the data; Huiqi Chen provided technical guidance on exposure measurement; Meng Ren analyzed the data and drafted the manuscript; John S. Ji, Aiqun Huang, Jiangli Di and Wannian Liang provided expert review of the manuscript.

Data sharing statement

The individual, de-identified participant data that underline the results reported in this article (text, tables, figures, and appendices) are available on reasonable request from the corresponding authors under certain conditions (with the consent of all participating centers and with a signed data access agreement).

Declaration of interests

The authors declare no competing interests.