Data on all singleton births in Brisbane from 1 January 2000 to 31 December 2010 were collected from the Data Collections Unit (DCU) of the Queensland Health Statistics Centre, which records antenatal, intrapartum, and postpartum data for all live births and stillbirths of at least 20 weeks of gestation, and/or at least 400 g in weight, born in Queensland. The data set we collected included the following variables: date of birth, gestational age in weeks, gender of baby, weight of baby, onset of labour (spontaneous, induced, and caesarean), mother's residential area (postcode), maternal age group, marital status, indigenous status, and parity. We only included spontaneous live births, which accounted for 56% of all births in this study, in order to gain a more accurate estimation of gestational effects attributed to heatwave, as most induced and caesarean births were medically indicated. In addition, to focus on heatwave exposure, as in another study, only spontaneous births that occurred in warm seasons (1 November–31 March) were used in our final analysis.
As the earliest conception date of pregnant women in the birth cohort was 13 March 1999, we collected environmental data for 1999–2010 to obtain information on environmental exposure for the whole pregnancy period. Data on meteorological factors, including daily maximum temperature, relative humidity, and ambient barometric pressure, from eight monitoring stations in Brisbane during 1999–2010 were obtained from the Australian Bureau of Meteorology. We used nine definitions of heatwave in this study through combining a series of cut-off percentiles and different durations: daily maximum temperature exceeding the 90th, 95th, and 98th percentiles of daily maximum temperature distribution of the study period for at least 2, 3, or 4 consecutive days (Table 1). We estimated the effects of heatwave exposure of pregnant women during their last gestational weeks before delivery.
Table 1. Summary of heatwave events during summer seasons in the period 1999–2010 using different heatwave definitions (HWDs) in Brisbane, Australia
|HWDs||Cut-off percentile and temperature (°C)||Duration (days)||Number of heatwave events|
We also acquired data on ambient air pollutants, including particulate matter with a diameter <10 μm (PM10), ozone (O3), carbon monoxide (CO), and nitrogen dioxide (NO2) from five stations in Brisbane, as many previous investigations have found that ambient air pollution could affect birth outcomes.[24-26] The air pollution data were provided by the Queensland Department of Environment and Resource Management. Weekly average levels of meteorological factors and ambient air pollution exposures were calculated by using the original daily data.
In addition, we also acquired data on the Social Economic Index for Areas (SEIFA) from the Australian Bureau of Statistics (ABS) to estimate maternal socio-economic status. SEIFA is a product developed by the ABS that ranks areas in Australia according to relative socio-economic advantage and disadvantage. The index is based on information from the 5-yearly census, and consists of four subindexes. In this study we used SEIFA data released in 2001 and 2006 to represent the maternal economic status of the births that occurred during 2000–2005 and 2006–2010, respectively. SEIFA data was linked to the birth records by year and area postcode.
Survival analysis was used in this study to explore the influence of heatwave on spontaneous preterm birth. Several studies have used the Cox proportional hazards (PHs) regression model to investigate the birth effects of environmental factors such as temperature and air pollution.[13, 24] The authors in these studies believed that survival analysis combined the advantages of cohort and time-series studies, as this model allows for the simultaneous examination of the impact of both subject-specific (e.g. individual behaviour risk factors) and time-related factors (e.g. air pollution, humidity, and other meteorological factors). Furthermore, the power of the survival analysis is increased compared with a case-crossover approach because all the subjects are examined.[24, 25, 27]
Preterm birth was defined as a live birth occurring at <37 completed weeks of gestation, according to the definition of the World Health Organization. We used the Cox PHs regression model (Cox regression), which is widely used in survival analysis, with time-dependent covariates to estimate the acute effects of heatwave. Cox regression has been used in epidemiological studies to analyse time-to-event data, with censoring. The standard Cox regression model assumes a constant hazard ratio (HR) over time: in other words, the effects on survival of covariates should be time independent. However, in our study, we must consider that values of covariates such as heatwave exposure at different times of the whole pregnancy period were not fixed, and so we fitted the Cox regression model with time-dependent covariates as follows:
where h0(t) is the baseline hazard function, being the hazard function for individuals with all explanatory variables equal to zero; XI refers to the values of time-independent variables such as the baby's gender, mother's age, and parity; Xi(t) are the values of the time-dependent variables (meteorological factors and air pollutants); and βI and βd are vectors of model parameters for time-independent and time-dependent covariates.
Heatwave exposure in this study was assigned as a binary value (yes/no), which indicated whether they experienced at least one heatwave event in the last gestational weeks before delivery.
Preterm birth is the outcome of interest in this study. We divided the whole pregnancy period into gestational weeks and assumed that within each of these, time-dependent variables such as weekly air pollution levels could be fixed. Consequently, the birth states of each pregnant woman were represented by a series of intervals of 1-week duration. Within each interval, a censoring variable was created that was 0 if a preterm birth did not occur, and 1 if a preterm birth occurred. For example, if one delivery occurred at 36 weeks of gestation, the pregnancy would be recorded as 36 intervals, with the final censoring variable equal to 1; if a delivery occurred at 38 weeks of gestation, the pregnancy would be recorded as 38 intervals with all values of censoring variable equal to 0.
We classified maternal age into three groups, <20, 20–34, and >34 years, and treated age groups as stratified variables in the model, as previous studies have found a nonlinear relationship between maternal age and birth outcomes: i.e. women younger than 20 years or older than 34 years were more likely to suffer from adverse birth outcomes, such as preterm birth, low birthweight, and even stillbirth.[30, 31] Indigenous status (yes/no), marital status (yes/no), parity (primiparity/multiparity), baby's gender (male/female), as well as SEIFA scores were entered into the model to adjust for demographic factors and areal socio-economic status. In addition, we used both single-pollutant and multi-pollutant models to adjust for the confounding effects of air pollution. To control for long-term trends, we added ‘year’ as a factor variable into the model.
We restricted the study to warm seasons so as to control the effects of seasonality of birth; meanwhile, a factor variable ‘Month’ was also included in the model.
The graphical methods, which used cumulative sums of the martingale-based residuals, were performed to check the proportionality assumptions of the Cox models used in our study.[32, 33] The results of the PH assumption on all covariates showed that for heatwave and most of other covariates, except relative humidity, the standardized and the observed score processes fluctuated randomly around zero, and the P values of the Kolmogorov-type supremum tests were larger than 0.05, which indicated that the PH assumptions on most variables were satisfied (Figures S1–S13). However, when we removed relative humidity from the model the estimates of heatwave barely changed. All analyses were conducted using sas 9.2 (SAS Institute Inc., Cary, NC, USA).