Exposure to extreme heat has direct and indirect health effects. Heat stress can directly induce illnesses such as heat edema, rash, cramps, syncope, exhaustion, and heat stroke [1
]. Heat stress can indirectly increase the likelihood of severe adverse health events including cardiovascular mortality [2
], stroke, and renal colic [3
]. There is also increased risk of adverse maternal [4
] and birth outcomes [5
], including gestational diabetes [4
], sudden infant death syndrome [5
], term placental abruption [6
], and early delivery [7
]. Use of antipsychotics, antidepressants, diuretics [8
], and illicit substances such as cocaine [9
] can increase the likelihood of experiencing heat-related illness or experiencing adverse health events because these drugs interfere with natural thermoregulatory mechanisms [1
]. This spectrum of health outcomes is influenced by the risk and protective factors within each population, and the intensity, duration, and timing of the heat event.
Extreme heat is a leading cause of illness and death from weather-related hazards in Canada. For example, there were an estimated 114 excess deaths during a 5-day event in greater Vancouver, British Columbia (BC) in the summer of 2009 [10
]. Similarly, there were an estimated 106 excess deaths during the 3-day event in Montréal, Quebec in the summer of 2010 [11
]. In addition, mean estimates for the Canadian cities of Montréal, Toronto, Ottawa, and Windsor were 121, 120, 41, and 37 excess annual heat-related deaths, respectively, between 1954 to 2000 [12
]. Heat exposure has also been associated with increased ambulance dispatches [13
] and emergency room visits [14
]. For example, a study in Toronto found an increase (95% confidence interval) of 32% (24%, 41%) in ambulance dispatches for heat-related illnesses associated with every 1 °C increase in mean temperature [13
]. Another study in rural southern Ontario found that the average rate of emergency room visits was 11% (7%, 15%) higher during heat events compared with average summer temperatures [14
Maximum temperatures during a heat event are strongly associated with the magnitude of the observed health outcomes. The temperature thresholds at which Canadian populations exhibit increased morbidity and mortality vary by location across the country. For example, mortality curves generated for 21 Canadian cities showed an increase in relative risk above baseline at mean temperatures as low as 20 °C (Calgary, Alberta) and as high as 27 °C (Windsor, Ontario) [15
]. Within BC, a similar study found increased risk over maximum apparent temperatures as low as 14 °C in the north, and as high as 22 °C in the south [16
]. This variability is due to several factors, including: different adaptive capacities, such as access to air conditioners [17
]; population and demographic differences; historical meteorological events, especially those hot enough to drive adaptation activities; and, most importantly, historical climate. To account for all of these factors, heat alert thresholds would be expected to differ by region [18
Environment and Climate Change Canada (ECCC) is responsible for issuing timely weather forecasts, warnings, and alerts across Canada, including heat alerts. Prior to 2015, ECCC issued heat alerts using a single national criterion based on the Canadian humidex, without consideration of different climates or population responses. At present, ECCC is modernizing the national heat alert program to incorporate regional climatology, health evidence, heat event duration, and overnight temperatures. Under this updated program, heat alerts are issued based on forecast high temperatures for two consecutive days and the intervening overnight low (referred to herein as the high-low-high approach). In some jurisdictions, ECCC has partnered with provincial public health agencies and Health Canada to complete the analyses necessary to identify evidence-based high-low-high thresholds. In other regions, thresholds are currently based on guidance from the World Health Organization (WHO) using the 95th percentiles of daily high and low temperatures [19
The BC Centre for Disease Control (BCCDC) collaborated with ECCC, Health Canada, and BC health authorities to establish high-low-high alert thresholds across all regions of BC. As of 2017, the only operational Heat Alert and Response System (HARS) in BC was restricted to the greater Vancouver area [10
]. Within this HARS, a heat alert was issued on dayt
when the average of the dayt
14:00 observed temperature and dayt+1
forecast high temperature equaled or exceeded 34 °C at Abbotsford airport or 29 °C at Vancouver airport [20
]. These thresholds were used from 2011 to 2017. Although no heat alerts were issued during this period, temperatures approached the threshold values on multiple occasions. Whenever this occurred, the BCCDC contextualized the temperature data with real-time data on registered deaths and an evaluation of results from a mortality nowcasting model and shared this information with health authority partners.
At the end of every summer, the BCCDC assessed the operation of the greater Vancouver HARS through retrospective evaluation of forecast temperatures, observed temperatures, and observed mortality to ensure that no important events were missed. This assessment process identified a period of prolonged heat in late June and early July 2015 that affected greater Vancouver but was not identified in real-time. The same episode also affected other regions of BC that did not have established HARS. Previous work has shown that populations in the coastal and northern regions of BC are susceptible to heat impacts despite the temperate climate of these regions [16
], further highlighting the need for more comprehensive HARS in the province.
One of the first steps in establishing HARS is to determine appropriate heat alert thresholds for different regions with varying climates and vulnerabilities, and a number of different approaches have been used worldwide. Synoptic classification systems identify air-mass categories using several meteorological factors and then assess excess mortality within each category [19
]. Other systems model the relationship between mortality and maximum temperatures, minimum temperatures, or apparent temperatures, using single-day values or multi-day averages [22
]. Many systems have different alert levels (i.e., heat warning versus heat emergency) with thresholds for each level reflecting a certain percentage increase in morbidity or mortality, which varies by jurisdiction [19
]. In Canada, the new high-low-high approach being used by ECCC required that we use methods tailored to this approach. Here we describe the process of establishing the high-low-high thresholds for four regions of BC. For each region, we identified thresholds that were (1) reliably associated with increased population mortality, and (2) unlikely to cause warning fatigue. We also discuss how such thresholds should be used for ongoing assessment of health vulnerabilities and climate change adaptation in the context of public health surveillance.
We identified high-low-high heat alert thresholds for four regions in BC as a necessary step in establishing HARS province-wide. We considered multiple factors in selecting the thresholds including minimization of warning fatigue, distributions of forecast and observed temperatures, evidence-based associations with daily mortality, and consistency with neighboring Canadian jurisdictions. We are confident that the high-low-high thresholds for the Southwest and Southeast HAAs will accurately and flexibly capture periods of increased risk of overall mortality. In the Northeast and Northwest HAAs, however, we did not observe any consistent or significant increases in overall mortality during heat events identified using the high-low-high approach. Northern BC is sparsely populated and hot weather is rare, which makes it challenging to detect an acute temperature-mortality effect using these methods. As such, we placed more emphasis on other factors, including the 95th percentiles of daily high and low observed temperatures, and consistency with neighboring jurisdictions. For example, the Northeast threshold of 29-14-29 °C is one degree below the 95th percentile of observed maximum daily temperatures (30 °C) and one degree above the 95th percentile of observed minimum daily temperatures (13 °C), but it is the same as the threshold for the Northern Prairie region of the neighboring province of Alberta.
Identifying the final heat alert thresholds was a more collaborative and iterative process among stakeholders than we can succinctly describe here. In brief, the BCCDC proposed an initial set of thresholds to ECCC based on the reported analyses and results. The thresholds were then modified slightly after an operational review by ECCC, and presented to the five regional health authorities, all of which have some area in at least two of the HAAs (Figure 1
). This led to further modifications, particularly for the Southwest region where the original proposal was for 28-15-28 °C. Because the densely populated greater Vancouver area spans the Southwest and Southeast HAAs and is administered by two different health authorities, we had to ensure that the two sets of thresholds identified similar alert periods in the historic data. Once complete, ECCC took a number of steps to operationalize the finalized thresholds by: (1) preparing standard operating procedures for forecasters; (2) preparing standardized impact and call to action statements that are available to forecasters for heat health messaging; and (3) establishing a communications strategy to ensure timely notification of public health authorities when extreme heat is expected.
Overall, minimization of warning fatigue was the most important consideration in finalizing the heat alert thresholds. Warning fatigue occurs when people become desensitized to warnings after hearing recurring messages about an event that did not materialize, thereby reducing vigilance and preparation in future [27
]. Factors such as trust and credibility, over-warning, false alarms, skepticism, and helplessness all contribute to warning fatigue, which must be managed with carefully designed systems and risk communication strategies [27
]. Too many heat alerts in a particular region could result in warning fatigue, especially given the uncertainty of temperature forecasts. In BC we found that forecasts consistently over-predicted the observed high and low temperatures (Figures S1 and S2
). It follows that there were more heat alerts per year based on forecast temperatures than on observed temperatures (Table 2
, Figure 3
), with the exception of some forecast areas including Vancouver, Victoria, Kamloops, and Penticton. These four areas are densely populated and located in HAAs with significant increases in mortality when temperatures meet or exceed the heat alert thresholds, and as such these results highlight the need for ongoing evaluation of the high-low-high thresholds established here.
The sub-analyses provided valuable insight into the utility of the final heat alert thresholds, particularly in the northern HAAs. First, the analysis on within-season variability suggested that thresholds should remain the same throughout the summer. Second, the analysis by age group found that risk was generally higher in those aged 65–75 years than risk for all ages, even in the Northwest HAA where the relative rate of mortality for all ages was less than 1.0. This is consistent with our prior work in the greater Vancouver area [10
], and continues to suggest that people 65–75 years of age are a higher risk group in BC. Equal risk between age groups in the Northeast HAA may be due to significantly lower life expectancy in this region [28
]. Finally, the analysis on location of death showed clear and consistent patterns across all four HAAs. One hallmark of extreme heat events is an increased number of deaths occurring at home and in the community [26
]. As such, it was important for our high-low-high thresholds to reflect this risk, even in HAAs where the overall association with mortality was largely null. Risk of mortality at home or in the community was clearly and significantly increased in the Northeast, Southwest, and Southeast HAAs, and elevated in the Northwest HAA.
One limitation of this work was its dependence on mortality as the indicator of population health effects from extreme heat. There are studies of other indicators available for other Canadian jurisdictions, including emergency room visits [14
] and ambulance dispatches [13
]; however, we did not have such data available for these analyses. Future work to establish heat alert thresholds could be strengthened by considering the combined effects of multiple outcomes that are known to be associated with heat to ensure that identified thresholds are appropriately predictive of morbidity and mortality. Another limitation involves the large differences in population sizes between the HAAs, which simply reflects the reality of BC and many other places in Canada. While it may have improved the analyses to have smaller HAAs with more evenly sized populations, it would have been too challenging for ECCC to operationalize.
Based on previous experience in the greater Vancouver area, operational heat alert thresholds and HARS provide a framework for routine and ongoing assessment of population response to extreme heat. Best practice requires evaluation of the heat alert thresholds and HARS at the end of every summer to ensure that they remain appropriately protective for the upcoming year [31
]. The thresholds presented here will first be assessed after the summer of 2018 using analyses similar to those we have described. For each HAA, the BCCDC plans to: (1) characterize any heat alerts that occurred during the summer in terms of forecast and observed temperatures and daily mortality; (2) identify false positive (forecast but not observed) and false negative (observed but not forecast) heat alerts; (3) identify anomalies in daily mortality and deaths out of care, and assess the temperatures before and during those anomalies; (4) evaluate whether lower thresholds would capture any important events that were missed; and (5) evaluate whether higher thresholds would reduce false positives and warning fatigue. This type of ongoing and systematic evaluation process is critical to ensure heat alert thresholds and HARS are effective. It can also play an important role in helping public health authorities to track and understand the evolving relationships between temperature and population health in a changing climate.