Introduction

Heatwaves can result in increased deaths and emergency hospital admissions, especially among vulnerable groups such as elderly people, young children and patients with chronic diseases. [1] Heatwaves are a significant but perhaps under-estimated environmental hazard as ambient thermal conditions have a significant impact on human health. [2]–[4] These impacts are highest during prolonged periods of unusual heat, such as the 15,000 deaths during the 2003 heatwave in France alone from the 1st to 20^th August. [5]

The magnitude of heatwave-related health effects depends on the intensity and duration of high temperatures, and also population acclimatisation and adaptation. [3], [5]–[7] What would be described as a heatwave in a temperate city may not be unusual in a tropical or subtropical city. These differences make it impossible to develop a global definition of a heatwave. [8] Even at a local level there can be disagreement about what constitutes a heat wave. For example, in Brisbane, the local health authority changed its definition in 2006. [9] This definition is used by a heat-warning system that gives advanced warning to the public and emergency services when forecast temperatures are above the heatwave threshold.

We are interested in the development of a heatwave definition and in whether changes in the definition are associated with appreciable changes in health outcomes. In this study, we compared heat-related health impacts using ten different definitions of heatwave in Brisbane, Australia.

Materials and Methods

Brisbane is the capital city of Queensland, located in the south-east corner of the state, and has a sub-tropical climate with the latitude 27°29'S and longitude 153°8'E. It is Australia's third largest city (after Sydney and Melbourne), covering an urban area of 1326.8 km² with a population of 992,176 on 30 June 2006. [10] In this study, we used daily time series data on emergency hospital admissions, air pollution and climatic variables collected for Brisbane city between January 1996 and December 2005. Mortality data from January 1996 to November 2004 (54,318 deaths totally) were also used. The mortality data were obtained up to November 2004 because of the delay in registering deaths by the government as all the data used in this study were acquired in 2006.

Data analysis

Statistical analyses were undertaken using daily data on temperature, relative humidity, deaths and emergency hospital admissions. The heat index was calculated by a formula based on given air temperature and relative humidity. [11] Ten heatwave definitions (HWDs) were compared based on the local climate data (Table 1). The first three definitions have been widely used in the literature. [6], [11], [14] In order to more broadly assess the heat-related health impacts, we also used other seven definitions based on both the intensity and duration of heatwaves. For example, we assessed the health impacts of exposure to 35°C (i.e., top 1% of daily maximum temperature) for 2 or 3 consecutive days (HWDs 1 and 4), and also examined the health risk of exposure to 32.6°C (i.e., top 5% of daily maximum temperature) for up to 5 days (HWDs 8–10).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Heatwave definitions (HWDs) and heatwave days during 1996–2005 in Brisbane, Australia.

https://doi.org/10.1371/journal.pone.0012155.t001

As a preliminary investigation the relationships between maximum temperature and mortality and emergency hospital admissions were explored using non-linear regression. We regressed the average daily number of deaths and EHAs against maximum temperature rounded to whole degrees C. We estimated the risk shape using a non-linear (quadratic) model, to capture the well known U-shaped association between temperature and health. [15]

Case-crossover analyses were used to assess the relationship between heatwaves and health outcomes using the ten HWDs. The case-crossover design is particularly useful here because it controls for trends and seasonal patterns in the dependent and independent variables. [16]–[18] We used the time-stratified case-crossover with a stratum length of 28 days, and additionally matched control days to case days using day of the week. The main independent variable was heatwave (categorised as yes/no). The dependent variable was either the daily number of deaths or the daily number of emergency hospital admissions. We also adjusted for the confounders including humidity and air pollution (ie, PM₁₀, NO₂ and O₃).

Results

Table 2 shows the daily summary statistics for maximum temperature, minimum temperature, relative humidity, mortality and emergency hospital admissions. Separate statistics are shown for the whole year and summer only in Brisbane between 1996 and 2005. The average daily deaths and EHAs varied from 5 to 43 and from 95 to 258, respectively. The maximum temperature was as high as 41.5°C with a mean maximum temperature of 26.3°C (standard deviation: 3.9°C); and the mean minimum temperature was 14.6°C (standard deviation: 5.3°C) for whole year.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summary statistics for daily temperature and health outcomes, 1996–2005 in Brisbane, Australia.

https://doi.org/10.1371/journal.pone.0012155.t002

The average number (represented by the circles in the plot) of daily deaths and EHAs was clearly associated with maximum temperature for the whole year and just in summer between 1996 and 2005 in Brisbane (Figure 1). There is a U-shaped relation between maximum temperature and health outcomes as both the mean daily deaths and EHAs increased in winter and summer. The temperature threshold appears to be around 27°C (Figure 1 C and D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Scatter plots of the relationship between maximum temperature and mortality (A and C) and emergency hospital admissions (B and D) between 1996 and 2005.

Fitted regression line (and 95% confidence interval) from a non-linear (quadratic) regression. EHAs: Emergency Hospital Admissions.

https://doi.org/10.1371/journal.pone.0012155.g001

Table 3 shows the estimated odds ratios (OR) of deaths and emergency hospital admissions during heatwave versus non-heatwave days. During heatwave days, there was a statistically significant increase in the odds ratios of emergency hospital admissions for all ten definitions. The odds ratios of mortality also increased for nine of the ten definitions using an unadjusted model, and eight of ten definitions using an adjusted model. Even a small change in the heatwave definition had an appreciable impact on the estimated morbidity and mortality. For example, the odds ratios of emergency hospital admissions varied from 1.03 to 1.18 for heatwave days compared to non-heatwave days (after adjustment for confounding factors). The results show that there is a strong association between heatwaves and health, but the strength of this association is variable and depends on the definition used.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Odds ratios of emergency hospital admissions and mortality during heatwave periods using the case-crossover, 1996–2005 in Brisbane, Australia.

https://doi.org/10.1371/journal.pone.0012155.t003

Discussion

There has been increasing research interest in assessing the health impact of heatwaves as the heat-related health risk is projected to rise in the coming decades because of climate change. [1] In this study, we found: i) small changes in the definition of heatwaves can lead to considerable differences in the risk assessment for heatwaves; ii) some commonly-used heatwave definitions do not appear to suit Brisbane (e.g., using HWDs 1 to 3); and iii) there was no conclusive evidence about which heatwave definition should be used, even though HWD 5 (i.e., the daily maximum temperature greater than or equal to 37°C—approximately the top 0.5% of temperatures—for at least 2 consecutive days) seems to be superior than other HWDs when the impact of heatwaves on both mortality and EHAs is considered. The intense of heat in HWD 5 seems to be greater than most definitions in Table 1. Figure 1 showed that the temperature threshold at which deaths and emergency hospital admissions began to rise was around 27°C in Brisbane, which is higher than the thresholds observed in most cities from Europe and the United States. [2], [15], [19] A high threshold for Brisbane appears reasonable because it is a subtropical city with a population acclimatised to heat.

Different definitions for heatwave have been used in recent publications. For example, Anderson and Bell reported that, comparing the 99th and 90th percentile temperatures for the community in the United States, heat-related mortality was most associated with a shorter lag (average of same day and previous day), with an average increase of 3.0% in mortality risk (95% posterior interval: 2.4% to 3.6%). [2] The study also found that heat effects were generally larger in the northern USA compared with the south, probably because people in the south are better adapted to hot weather. Hansen et al. defined heatwaves in Adelaide, Australia as three or more consecutive days when the daily maximum temperature reached or exceeded 35°C. [6] Compared with non-heatwave periods, hospital admissions increased by 7.3% during heatwaves. In another study, Knowlton and colleagues used a variety of locally varying definitions of “heatwave” in assessing the impact of the 2006 California heatwave on hospitalisations and emergency department visits. [20] They suggested that a heatwave definition based on a higher temperature threshold would be associated with a greater increased relative risks for hospital admissions compared with emergency department visits. Metzger et al. examined a range of different temperature measures and found that the maximum heat index over the previous three days gave the best estimates of heat-related mortality in New York. Evidently, there is a need to evaluate locally valid HWDs and their impact on health outcomes. [21]

This study has three major strengths: 1) this is the first study to broadly examine heat-related health effects using different kinds of HWDs. The results of the study demonstrate that it is fundamentally important to evaluate how a heatwave is defined when a comparison of heat-related health effects is made; 2) the data used in this study were of good quality, with no missing values; and 3) importantly, we were able to adjust for the possible confounding effects of air pollution and humidity.

This study also has two key weaknesses. Firstly, it focused on only one city. However, the finding that different heatwave definitions lead to variations in the size of the health outcomes is likely to apply to other locations. Secondly, we only considered the effect of heatwaves on total mortality and emergency hospital admissions, but did not examine different causes of deaths or specific types of admissions. This issue will be addressed in further research.

Finding the best definition of a heatwave is a challenging research question. [3], [22] This study used ten different heatwave definitions, including both absolute and relative ones, and demonstrated that even a small change in the definition had a considerable impact on the estimated risk of heatwave. As global warming continues, the frequency, intensity and duration of heatwaves are likely to increase. [1] Thus, a community-based response plan for heatwave is necessary to ensure that prompt and appropriate public health intervention can be implemented to minimise the impact of heatwaves. It is important to identify an appropriate definition of heatwave locally and to understand its local health effects in order to develop appropriate public health intervention strategies to prevent and mitigate the impact of heatwaves. It is also one of the fundamental research issues which need to be addressed when we compare the heat-related health impacts across different regions.

Acknowledgments

We thank the Queensland Treasurer, Queensland Health, Queensland Environmental Protection Agency, Australia Bureau of Meteorology and Australian Bureau of Statistics for providing the relevant data.