Ambient Temperature and the Risk of Preterm Birth in Guangzhou, China (2001–2011)

1Preterm birth, defined as birth < 37 completed weeks of gestation, is a complex syndrome and remains a major global public health problem. It is estimated that approximately 15 million babies were born preterm in 2010 worldwide, accounting for 11% of all live births. Preterm birth is associated with a variety of adverse outcomes, including infant mortality, acute morbidity, long-term neurocognitive disorders, poor growth, respiratory illnesses, behavioral problems, and other complications. Although a number of risk factors have been explored and relevant interventions have been developed, the incidence of preterm birth has been increasing during the past two decades in most countries, indicating that further research is needed to elucidate its etiology.

2Seasonal weather changes have extensive effects on human health outcomes, especially for overall mortality, infectious diseases, and cardiovascular diseases. As a specific population, pregnant women might be susceptible to effects of weather changes. Compared with elderly people, however, pregnant women have received less attention. Recently, there is an emerging interest in examining the effects of environmental temperature on the risk of preterm birth. Several time-series studies observed that maternal short-term exposure to high temperature was associated with an increased risk of preterm birth. On the other hand, negative results were also reported in a few studies. Furthermore, a recent cohort study in Sweden reported that extremely cold temperature was associated with an increased risk of preterm birth. In the context of global climate change and high medical cost for preterm birth, it is necessary to clarify the effects of environmental temperature on the risk of preterm birth.

3The most biologically relevant time period for effects of temperature on preterm birth is not known. Previous studies focused mainly on the short-term exposure, which was defined as a few days or weeks before delivery. Because long-term exposure to temperature might also be relevant for preterm birth, we used a Cox proportional hazards model with gestational age as the time axis to evaluate the relationship between short- and long-term temperature exposures and the risk of preterm birth from 2001 to 2011 in Guangzhou, a subtropical city in China. Considering time-dependent exposures, the Cox proportional hazards model has been shown to be useful for evaluating exposure windows for preterm birth.

4Animal studies have suggested that high temperature exposure could lead to dehydration, reduce uterine blood flow, and increase the secretion of proinflammatory cytokines and oxytocin, thereby inducing the uterine contraction and onset of labor. Previous studies also proposed that cold temperature could lead to elevated blood viscosity and vascular constriction and increase exposure to risk factors of preterm birth, such as passive smoking and infectious agents. These studies suggested that temperature might affect the timing for initiation of spontaneous labor. In the present study, we aimed to examine the effects of temperature on the timing for natural labor. Similar to previous studies, we excluded cesarean section deliveries because they would not reflect physiological mechanisms of temperature on spontaneous birth timing.

5Guangzhou is the largest metropolis in Southern China with a population of 12.9 million (Figure 1). It has a typical subtropical climate with mild winters and hot summers. Data on all singleton births between 1 January 2001 and 31 December 2011 were collected from the Guangzhou Perinatal Health Care and Delivery Surveillance System (GPHCDSS), which was initiated in 2000 and covered > 97% of deliveries in Guangzhou. In the GPHCDSS, information on births is collected from hospitals via computer network and is used to issue birth certificates. The information is confirmed by the chief of midwives and the chief of physicians in hospitals after data entry. When the birth certificate is issued, the Department of Medical Administration and parents also confirm the birth information. Data obtained from GPHCDSS included date of birth, gestational age at birth, birth weight, infant sex, maternal age, educational level, registered residence, and parity. Registered residence represents the permanent residence in the system of household registration (Hukou system) in China, and generally categorized as rural and urban residence. Urban residents find it easier to get access to health services, medical insurance, education, and other social welfare. The date of conception was calculated by subtracting the gestational age at birth from the date of birth.

6All singleton vaginal births at the gestational age of 20–43 weeks were included in the present study. Because the data were collected by the birth date, short pregnancies from the earliest possible conception dates that terminated before the start date of the study (1 January 2001) were not included in the analysis; similarly, long pregnancies from the latest possible conception dates that terminated after the end date of the study (31 December 2011) were also not included in the analysis. Thus, women included at the beginning of the study tended to have longer pregnancies, whereas those included at the end of the study tended to have shorter pregnancies. This phenomenon could lead to artificial seasonal patterns in gestation length and distort the association between time-dependent exposures (e.g., temperature) and outcomes. To avoid this potential “fixed cohort bias,” we excluded those pregnancies conceived earlier than 20 weeks before the study’s start date and those conceived later than 43 weeks before the study’s end date to ensure that all pregnancies are entirely observable within the study period.

7This study was approved by the institutional ethical committee board of Guangzhou Women and Children’s Medical Center. Data used in this study were anonymous, and no individual identifiable information was available for the analysis.

8Preterm birth was defined as birth < 37 completed weeks of gestation. Gestational age at birth was determined according to an ultrasound examination in the first or second trimester. If the ultrasound examination was unavailable, the last menstrual period (LMP) was used to calculate the gestational age. Preterm birth was also subdivided into three groups: 20–31, 32–34, and 35–36 weeks of gestation.

9The daily meteorological data during the study period, including mean temperature (degrees Celsius) and relative humidity (percent), were collected from the Guangzhou Meteorological Bureau. The meteorological data were recorded at a meteorological station in Guangzhou. Daily values for temperature and relative humidity were calculated by averaging four measurements taken at 6-hr intervals over each 24-hr period. Daily air pollution data from 2006–2010 were obtained from Environmental Monitoring Center, including particulate matter with an aerodynamic diameter ≤ 10 μm (PM10; micrograms per cubic meter), sulfur dioxide (SO2; micrograms per cubic meter), and nitrogen dioxide (NO2; micrograms per cubic meter). Air pollution data were measured at a fixed environmental monitoring station located in the center of Guangzhou. The monitoring station providing meteorological and air pollution data follows the quality assurance and quality control programs that were mandated by the Chinese government. These data on temperature and air pollution have been widely used to explore their effects on mortality in Guangzhou.

10We used the estimated date of conception to determine the start and end date of each gestational week for each pregnancy. We then derived the weekly average temperature during each completed gestational week by averaging the daily mean temperature on the last day of the corresponding gestational week and the 6 previous days. We used the same approach to estimate weekly average values for relative humidity, PM10, SO2, and NO2 for each completed gestational week of pregnancy.

11We used a Cox proportional hazards model with gestational age as the time axis and preterm birth as the outcome. Term births were censored at the end of the 36th week of gestation because they were no longer at risk of being born preterm. We modeled weekly average temperature as a time-varying independent variable to estimate the hazard of preterm birth during each completed gestational week t in association with temperature during four exposure windows: a) 1-week (the weekly mean temperature during week t); b) 4-week (weekly mean temperature for each gestational week from week t – 3 through t; the 4 weeks are t – 3, t – 2, t – 1, and t); c) late pregnancy (weekly mean temperature for each gestational week from week 20 through week t) and d) cumulative (weekly mean temperature for each gestational week from week 1 through week t).

12We modeled time-varying weekly temperatures during each time window as a cubic spline with 3 degrees of freedom (df) to allow a nonlinear relationship with preterm birth, and plotted the resulting hazard ratios (HRs) and 95% confidence intervals (CIs) to show the relationship over the entire temperature distribution. In addition, we calculated specific HRs relative to the local median temperature (24.4°C) for locally extreme cold (defined as the 1st percentile of temperature, 7.6°C) and moderate cold (the 5th percentile, 11.2°C), and for locally extreme heat (the 99th percentile, 31.9°C) and moderate heat (the 95th percentile, 30.7°C). In addition to estimating associations with all preterm births, we performed stratified analyses to estimate associations with three preterm birth subgroups (births during gestational weeks 20–31, 32–34, and 35–36).

13Maternal age was categorized into three groups (< 20, 20–34, and > 34 years) and included in the model as a stratified variable, because we observed that both women aged < 20 and > 34 years had higher risk of preterm birth. The following sociodemographic variables were adjusted for: maternal education level (high school or below vs. college, undergraduate, master or above), parity (primiparity vs. multiparity), and baby’s sex (male vs. female). To control for long-term trend and seasonality, we included the year of conception and the month of conception as indicator variables. Similar to the temperature, the relative humidity was included as a time-dependent variable at each exposure window using a spline with 3 df.

14Sensitivity analyses were performed to evaluate the robustness of results. We included the month of conception as a spline with 3 to 6 df. For pregnancies that took place during 2006–2010, we performed separate models adjusted for weekly average concentrations of PM10, SO2, and NO2 as time-varying covariates, using time windows corresponding to those used for temperature. However, because air pollution exposures might be causal intermediates between temperature and preterm birth, adjustment may block part of the total effects of temperature. Finally, we used a quasi-Poisson regression with distributed lag nonlinear model (DLNM) as an alternative approach to estimate the cumulative effects of low and high temperatures on preterm birth for single-day lags (lag 0, representing temperature on the same day of preterm birth, up to 27-day lag) and during three cumulative lag periods (lag 0–6, 0–13, and 0–27 days, where lag 0–6 represents the temperature on the day of preterm birth and the previous 6 days, and so forth). The model used for analysis was

15Log[E(Yt)] = α + βTempt, l + NS(RHt, 5) + NS(Time, 5/year) + γDOWt + δLog(Zt),

16where t represents the day of the observation; Yt is the observed number of preterm births on day t; Tempt,l is a matrix representing the two-dimensional relationship of mean temperature and lag days; l denotes the maximum lag days; NS(RHt, 5) is a natural cubic spline with 5 df for relative humidity; NS(Time, 5/year) is a natural cubic spline with 5 df/year for long-term trend and seasonality; and DOWt is the day of the week (a categorical variable with Sunday as the reference). Zt is a daily expected count of preterm births on day t:

17where Zjt is the number of pregnancies at risk during gestational week j (ranging from 20 through 36 weeks) on day t, and Wj is the conditional probability of birth during gestational week j, as proposed by. The quasi-Akaike information criterion (QAIC) was used to choose the df for temperature and lags, and the final model used a natural cubic spline with 4 df to estimate the nonlinear effect of temperature, and 4 df to estimate the lag effects. We plotted relative risks (RRs) across the entire temperature distribution for single-day lags, and estimated specific RRs for extremely low and high temperatures relative to the median (24.4°C) for the three cumulative lags.

18All statistical analyses were performed using SAS statistical software version 9.3 (SAS Institute Inc., Cary, NC) and the R statistical environment (version 2.15) with the smoothHR and DLNM packages.

19A total of 1,420,348 singleton births at the gestational age of 20–43 weeks were recorded during the study period. After exclusion of 582,202 cesarean section births (with preterm birth rate of 5.2%), 838,146 singleton vaginal births were included in this study. Among them, 47,209 (5.6%) were preterm births, with a daily average of 11.8 preterm births. The demographic characteristics of the study population are shown in Table 1. Compared with the mothers of term babies, those of preterm babies tended to be primiparous and have higher percentage of age < 20 and > 34 years. Male infants were more likely to be born preterm. Preterm babies had a slightly higher percentage of being born in winter (December–February) (Table 1). The distributions of educational level and registered residence were similar between the preterm birth and term birth groups (Table 1).

20The median daily mean temperature and relative humidity were 24.4°C and 74% during the study period. The Spearman’s correlations showed a weak correlation between mean temperature and relative humidity (rs = 0.098). Figure 2 shows the 11-year time series of temperature and preterm birth rate, which was calculated by dividing daily counts of preterm birth by daily births. The daily mean temperature was lowest in December and January and peaked in July and August (Figure 2). The daily rate of preterm births appeared to be relatively high in January to March, although the yearly variations were not completely consistent over the study period (Figure 2).

21Figure 3 shows adjusted HRs for weekly mean temperatures, relative to 24.4°C, during different exposure windows. Associations were U-shaped, with significant associations between preterm birth risk and cold temperatures during all four time windows, and significant associations with high temperatures during the 4-week, late pregnancy (≥ 20 weeks), and cumulative (entire pregnancy) time windows, but not the 1-week time window. Associations with extremely and moderately low and high temperatures were strongest for temperature during the 4-week time window, and HRs for low temperatures were larger than corresponding HRs for high temperatures during all time windows (Table 2). For example, relative to 24.4°C, exposures to extreme cold (7.6°C) and extreme heat (31.9°C) during the 4-week time window were associated with 17.9% (95% CI: 10.2, 26.2%) and 10.0% (95% CI: 2.9, 17.6%) higher risks of preterm birth, respectively.

22Figure 4 presents the stratified relationship of preterm birth subgroups with the mean temperature. Generally, the shapes of stratified relationships were similar to those in the overall analysis. Associations with higher temperatures were stronger for preterm births during weeks 20–31 (n = 4,988 preterm births) and 32–34 (n = 11,323) compared with preterm births during weeks 35–36 (n = 30,898), though estimates for later preterm births were more precise, at least in part due to larger numbers of observations (Figure 4). Specifically, the HRs (95% CIs) of extremely high temperature for preterm births during weeks 20–31 and 32–34 were 1.45 (95% CI: 1.15, 1.83) and 1.29 (95% CI: 1.08, 1.54) at the 4-week time window, respectively, and the HR during weeks 35–36 was 1.08 (95% CI: 0.97, 1.21).

23In the sensitivity analyses, we included the month as a spline with 3–6 df, and the effect estimates for temperature remained similar (see Table S1). HRs for temperature and preterm among pregnancies during 2006–2010 (n = 450,230, including 25,329 preterm births) after adjustment for weekly average PM10, SO2, and NO2 concentrations during time windows corresponding to those for temperature were similar to those for overall study population, although HRs were slightly attenuated and less precise after adjustment. However, these differences may be attributable at least partly to the use of a different population (2006–2010 vs. 2001–2011), rather than to effects of adjustment for air pollutants per se (see Figure S1).

24Associations between preterm and temperature also tended to be U-shaped when estimated using a quasi-Poisson regression with DLNM (see Figure S2). Associations with extreme cold and extreme heat both were strongest for temperature on the same day of preterm birth (lag 0) and null after 4- and 6-day lags, respectively. The association with extreme cold was strongest for the longest cumulative lag (lag 0–27), whereas HRs for extreme heat were > 1 at lag 0–6 but < 1 for longer cumulative lags (see Table S2).

25In the present study, both low and high temperatures during pregnancy were associated with an increased risk of preterm birth. Extreme cold temperatures were associated with the risk of preterm birth during all time windows examined (1-week, 4-week, late pregnancy, and entire pregnancy), and extreme heat was associated with preterm birth risk during all but the shortest time window of exposure. Furthermore, extreme heat was more strongly associated with preterm births during 20–31 and 32–34 weeks of gestation than with preterm births during weeks 35–36.

26To our knowledge, this study may be the first one to investigate the effects of ambient temperature on the risk of preterm birth in developing countries. Air conditioning or central heating is less used in developing countries than in developed countries; thus the population may be more susceptible to climate change, and the harmful effects of extreme temperature on health may be more pronounced. Using the time-to-event model, we also extended earlier work by examining different temperature exposure windows, including both short- and long-term exposures. Moreover, our study has a larger sample size compared with previous studies, which enabled us to evaluate expected subtle effects of temperature.

27The prevalence of preterm birth among singleton in the study area (5.6% for vaginal delivery, 5.2% for cesarean section delivery) was similar to those of previous reports in Guangzhou (5.1% to ~ 6.4% from 2002 to 2012), Korea (5.3% in 2008), and Japan (5.7% in 2012), which were lower than that of the United States (9.7% in 2013). The rate of preterm birth in China is lower than that of some Western countries. This may be explained by several factors, including genetic predisposition (black women tend to have higher rates of preterm birth than other ethnic groups), early prenatal identification and termination of pregnancies with significant fetal anomalies that might otherwise result in a preterm birth, and lower prevalence of risk factors, such as maternal advanced age and active smoking. Furthermore, we did not observe significant differences in maternal educational level and status of registered residence between preterm birth and term birth subjects. This might be explained partly by implementation of several public health programs to promote maternal and children’s health in Guangzhou, which improved the equality for access to antenatal care service among different social economic statuses.

28Few studies have explored the effects of ambient temperature on the rate of preterm birth with inconsistent results. Two large studies, conducted in London, United Kingdom, and in two German states did not find any association between temperature and the risk of preterm birth. Nevertheless, a study in Rome, Italy, reported that a 1°C increase in maximum apparent temperature at lag 0–2 resulted in an increase of 1.9% in daily preterm birth in warm season. In addition, a study in California showed that every 5.6°C increase in apparent temperature at lag 0–6 increased preterm births by 8.6%. In the present study, we observed a U-shaped relationship between temperature and preterm birth over the whole year. This finding was in accordance with a recent study in Sweden, which also found that both cold and warm temperature extremes were associated with an increased risk of preterm birth. Furthermore, we observed that an association of extreme cold with preterm birth risk was stronger than that with extreme heat, although the low temperature is mild in our study region. This result might be explained by the fact that the population in Guangzhou, a subtropical city, is less accustomed to cold weather, and thus more susceptible to cold stress. In addition, central heating is rarely used in subtropical areas such as Guangzhou, which may lead to greater adverse effects of extremely cold weather on health.

29There were several potential reasons for differences between our findings and those of previous studies. First, the range of climatic conditions, socioeconomic level, and population susceptibility may vary across areas. In addition, we speculate that the study design, statistical methods, exposure windows, and sample size are important factors. For example, previous studies have examined nonlinear relationships using logistic regression, Cox proportional hazards model, and Poisson generalized additive model, whereas the delayed effects have been estimated using single-day lags and moving average lags, distributed lag model, or distributed lag nonlinear models. In the present study, estimates from both the Cox proportional hazards model and the quasi-Poisson regression with DLNM indicated U-shaped associations between preterm and mean temperature, though associations based on the quasi-Poisson regression with DLNM were strongest for recent exposures (single-day lags < 1 week), whereas high temperatures during the 1-week time window were not significantly associated with preterm based on the Cox proportional hazards model. This indicated that different statistical methods used might result in different estimates, even in the same study population. Compared with other statistical methods, the Cox proportional hazards model has several advantages. For example, considering time-varying exposures, the time-to-event analysis could effectively evaluate both long- and short-term exposure windows. Furthermore, the time-dependent survival analysis could also eliminate potential confounding, which is caused by seasonal variation in the gestational age distribution of pregnancies at risk (< 37 weeks) and the increasing probability of birth with higher gestational age, by comparing fetuses at the same gestational age. Therefore, the Cox proportional hazards model has been recommended for future studies of time-dependent exposures on preterm birth.

30Biological mechanisms that might contribute to the association between ambient temperature and preterm birth are unclear. Previous studies proposed several hypotheses. High temperature exposure causes dehydration and reduces uterine blood flow in pregnant women, which can induce the uterine contraction and onset of labor. In addition, animal studies showed that heat exposure may increase the secretion of proinflammatory cytokines and oxytocin to induce labor. Furthermore, high temperature may increase blood viscosity and cholesterol levels, which may be risk factors of preterm birth. In our study population, extreme cold (7.6°C) also was associated with preterm birth. There are several explanations for the association with low temperature. Previous studies showed that cold temperature could increase exposure to passive smoking and infectious agents, and result in occurrence of preeclampsia and pregnancy-induced hypertension, all of which are potential risk factors for preterm birth. In addition, cold temperature could also lead to increased blood viscosity and vascular constriction, and thus might be associated with an increased risk of preterm birth. Taken together, there are many potential pathways through which temperature can affect the risk of preterm birth. Further studies are needed to elucidate the underlying mechanisms.

31Several limitations should be mentioned. First, we used data from one meteorological station to define the exposure for all women in Guangzhou. This could cause exposure misclassification. Second, we were unable to distinguish spontaneous preterm birth from iatrogenic preterm birth, although we restricted the study population to those with vaginal delivery. Exposure to temperature might have different effects on these subtypes of preterm birth. We have no data on the percentages of identification method of gestational age (LMP, clinical estimate, ultrasound) and their changes during the study period, which might influence the incidence of early preterm birth and late preterm birth. In Guangzhou, it is a routine practice to confirm gestational age using ultrasound examination in the first or second trimester. However, some clinicians in poor rural areas might still be using the LMP-based or clinical estimation–based methods (estimated as < 10% of all births). We also did not have information on the history of preterm delivery and pregnancy complications, all of which are potential risk factors for preterm birth. Finally, we are unable to distinguish the consequences of the temperature itself with other factors associated with changes in weather, such as nutrition, activity, and infectious exposures.

32Exposure to both low and high temperatures during pregnancy was associated with an increased risk of preterm birth in our study population. These findings might have important implications in preventing preterm birth in Guangzhou as well as other areas with similar variations in weather conditions.