Temporal trends in human vulnerability to excessive heat

An overview of the manuscripts (table 2) reveals the varied ways in which this topic has been approached across the 71 articles (table 3). The locations of analysis reveal a bias that echoes that of the overall heat-related literature, with a preponderance of studies that assess temporal shifts in heat-related mortality in North America and Europe, with no studies using data from outside of these two regions, Australia, and East Asia, likely reflecting the difficulty in securing long-term, often sensitive data sets. Complicating the fact that the definition of heat will vary from one research project to another is that 41 of the 71 papers evaluate trends at one single spatial unit, ranging from the city to country, along with one paper assessing all US Military bases globally (Carter et al 2005). This leaves only 30 papers for which spatial variability in temporal changes in heat vulnerability can be explicitly defined, although in only 12 of them do the studies span multiple countries or climate zones, and only two explicitly segregate rural areas in their study design.

In terms of how heat itself was evaluated, 16 primarily focus on comparing the health impacts of two or more heat waves within the same location, 11 of which analyzed only two events; six of these 11 directly or indirectly cite the development of heat warning systems as an aspect of the differences observed.

The remaining 55 papers focus upon changes in the overall statistical relationships between high temperature and health outcome; more than half of these subdivide available data into general periods (typically decades) to foster comparison, while eight have divided their data into subsets based on a singular event (implementation of a watch warning system, unprecedented heat event, or in the case of Japan, electricity rationing following the 2011 tsunami); 17 papers examined generalized trends, either calculating a linear or non-linear slope, or evaluating interannual variability through other means. The papers have a median of 29 years' worth of data, ranging from five to 151 and an interquartile range of 18–38 years.

The overarching majority of papers utilize some broad aggregation of daily mortality totals; depending on the study this may include all-cause mortality, all non-accidental mortality, or mortality within certain broad subsets (e.g. cardiovascular, respiratory). Only five papers exclusively examine direct heat-related mortality events, and seven examine heat-related morbidity (either hospitalizations or ambulance call-outs). The papers also vary considerably, including estimating direct absolute differences in mortality, rates of mortality that account for demographic changes, or relative-risks with regard to temperature values. Roughly half of the papers (33) consider either lagged influence of weather on health; these range from assessing cumulative impacts via models such as the distributed lag nonlinear model (DLNM, Gasparrini et al 2010) to simply lagging the mortality response; most lags were one week or less, though several papers extended the impacts out to up to 30 days.

Seven papers (Åström et al 2013a, Carson et al 2006, Coates et al 2014, Åström et al 2016, Barreca et al 2016, Ekamper et al 2009, Petkova et al 2014) assessed trends in heat-related mortality over a period dating back to at least 1900. Given the coarser resolution of mortality data (generally weekly or monthly aggregates), the changes in the relationship over time are studied in a more general fashion, but all seven studies show a clear decrease in human mortality to all temperature extremes; including overall heat-associated mortality rates decreasing 70% in the US (Barreca et al 2016) and 85% in Australia (Coates et al 2014). In these papers, broad changes in the seasonality in mortality are observed, as they span vast changes in societal lifestyle, education, housing, and wealth.

Far more studies avail of shorter periods of study, generally from the 1960s to the present. The overwhelming majority of the research studies (56 out of 64) have concluded that the relative impact of heat events on human health has declined in recent decades. This is true for both heatwave-to-heatwave comparisons as well as long-term epidemiological studies. While most studies conclude that heat-related mortality has decreased, most still show a statistically significant increase in mortality during periods of hot weather; further, most studies have not directly compared the statistical significance of the change in vulnerability over time and thus the results are difficult to assess. In other studies where differences have been tested for significance, changes can be ambiguous. For instance, in Gasparrini et al (2015), at 90 percent (245/272) of the locations studied the point estimate for relative risk (RR) of human mortality at the highest observed temperature is lower in 2006 than in 1993; however, when aggregated to the national level, in only three of the six countries which were studied was the decrease in RR statistically significant.

Several factors must be considered to contextualize these results. First, many of the longer-term studies evaluate heat vulnerability in terms of relative risks to exposure, with few papers actually exploring the overall total change in heat burden, in which changes in the frequency of heat events and as well as human vulnerability are assessed. For example, in Christidis et al (2010), a decrease in relative vulnerability to hot conditions is offset by an increasing number of heat events, resulting in an overall insignificant (but upward) trend in total mortality. This result is echoed in two studies that have examined totals of direct heat-related deaths and found an increase over time (Akihiko et al 2014, Johnson et al 2016), and several studies that have found increases in specific types of mortality (e.g. cardiovascular in Kim et al (2015)), and Li et al (2017) who found a substantial increase in years of life lost due to heat in Ningbo, China, in the two years following a major heat event. In other cases, where ambiguous results have emerged, these have been related to unprecedented conditions, such as de'Donato et al (2015), in which observed increases in vulnerability in Helsinki may be associated with increased maximum temperatures; and Nitschke et al (2011) who in studying the 2009 heat event in Adelaide, Australia notes its lack of similar events. Miron et al (2008) note an increased vulnerability to heat that they associated with a lowering of the temperature at which heat impacts begin at their sites in Spain, which they suggest is associated with an increased elderly population.

Of the studies that have examined heat wave-to-heat wave variability in human impacts, only in the Nitschke et al (2011) study are heat impacts greater with the later event, though this is framed in the context that the later (2009) event was unprecedented for Adelaide, Australia. For the summer of 2015 in the Czech Republic, heat related mortality was similar to that of 1994, despite a baseline decline in heat vulnerability (Urban et al 2017). For the remaining studies, all show general declines; for instance, Palecki et al (2001) and Weisskopf et al (2002) show mortality in the midwestern USA in a 1999 heat wave was far lower than that of the 1995 'Chicago Heat Wave'; Smoyer (1998) shows in St. Louis, USA the 1995 heat event was less impactful than a 1980 event; and Tan et al (2007) show similar decreases between heat events in 1998 and 2003 in Shanghai, China.

Several studies have focused on the 2003 European Heat Wave, and all point to general decreases in impacts over time. In some areas that were more marginally affected, mortality was observed to be lesser than in previous events (e.g. De Castro et al 2011, in Spain, Pascal et al 2013, in Ireland, and Kyselý and Kříž 2008, in the Czech Republic). In core areas, where mortality was much more substantial in 2003, Fouillet et al (2008) compared the event in France to a 2006 event, and estimates that approximately 4400 fewer deaths were observed in 2006 than would have been the case if the impacts were similar to 2003; similarly, Green et al (2016) shows mortality in the UK in a 2013 event was only one fifth as strong as 2003.

Historical heat-mortality research has frequently noted that some places are more vulnerable than others, and several papers have examined the spatial variability in temporal changes in vulnerability, with mixed results. Several studies that have examined multiple cities across diverse climates have shown the biggest decreases in heat mortality across the regions that were traditionally more vulnerable (e.g. Donaldson et al 2003, Barnett 2007, Sheridan and Dixon 2017, Bobb et al 2014, Davis et al 2002, 2003a, 2003b, 2004, Sheridan et al 2009), such as the northeastern, midwestern, and northwestern US. In these cooler climates, higher mortality risk was found for at the beginning of the period, suggesting hotter climates had already adapted to some extent. In contrast, other studies that assess spatial variability see no clear connection between why certain places have greater reductions than others (Miron et al 2008, Linares et al 2015, Michelozzi et al 2006, Guo et al 2012, Wang et al 2016, Kalkstein et al 2011, Schifano et al 2012), or no clear variability (Onozuka and Hagihara 2017). In South Korea, Lee et al (2016) showed heterogeneity in the temperature-mortality relationships when comparing inland and coastal locations. Ragettli et al (2017) show a greater decrease in heat vulnerability in the warmer cities in Switzerland, to which they attribute greater emphasis on heat awareness. de'Donato et al (2015) similarly see a greater decrease in warmer locales, in Europe, with increases in vulnerability in cooler cities, including Helsinki, by virtue of heat events exceeding historical norms. Converse to expectations, Kim et al (2017) show a greater decrease in heat vulnerability in Japanese prefectures where electricity consumption decreased the most following the 2011 Fukushima disaster.

In some cases, economic development is cited in heat vulnerability decreases; Ng et al (2016) showed statistically the relationship between improvement in wealth and decrease in mortality; while Chung et al (2017) show an earlier decrease in vulnerability in Japan than Korea or Taiwan, which is connected into different timing of development.

Possibly due to data limitations, very few studies examine rural communities explicitly. In Ireland, Pascal et al (2013) noted a greater decline in heat vulnerability in rural areas; while in urban Berlin and rural Brandenburg, Germany, Gabriel and Endlicher (2011) found excess mortality during two heatwaves to be higher in urban regions overall, with no clear difference in trend. Todd and Valleron (2015), in examining all of France, found a evidence of acclimatization in rural areas as well as urban areas over time, though these were not specifically partitioned.

The majority of the studies considered mortality health outcomes, likely due to the availability of health data; for the sake of statistical power, most of these studies made their primary outcome all-cause total mortality, in some cases excluding accidents. Of the very long-term studies, differential changes in vulnerability broadly reflect larger societal changes; e.g. in Australia, with decreases in outdoor labor the male heat mortality rate has fallen faster than the female rate (Coates et al 2014), and in the greater decreases that are seen with children mortality rates, with other studies showing slower decreases in the elderly mortality rate (Åström et al 2013b) or even a noticeable uptick (Ekamper et al 2009) indicative of changes in overall age structure. In studies confined to more recent decades, decreases in vulnerability have been most substantial in the oldest age groups (Ruuhela et al 2017, Sheridan and Dixon 2017, Bobb et al 2014, de'Donato et al 2015), although in others there was no consistent age-related differentiation (Onozuka and Hagihara 2015, Ha and Kim 2013).

In terms of causes, when split, some studies see relatively similar declines in respiratory and cardiovascular deaths (e.g. Bobb et al 2014, Li et al 2017, Ng et al 2016); in others the results are more diverse, such as in de'Donato et al (2015) where the relative changes between respiratory and cardiovascular vary across the 9 European cities studied, Kyselý and Plavcová (2012) where cardiovascular mortality shows a steeper drop, in Miron et al (2015) where cardiovascular mortality decreases but respiratory does not, Ha and Kim (2013) which shows a similar drop in cardiovascular mortality but actually shows an increase in non-accidental mortality due to heat over time, or Yang et al (2015) which found no long-term trend associated with cardiovascular-heat effects.

In studies that have examined differences between the sexes, generally results are comparable, although in some cases (e.g. Muthers et al 2010a, Kyselý and Plavcová 2012) females show a greater decrease in heat vulnerability over time, though coming from a higher baseline. In Vienna, Matzarakis et al (2011) showed a slightly higher sensitivity to heat stress for women, but this finding was contingent upon the thermal variable used to assess sensitivity.

Of the studies included in our review, only three (Onozuka and Hagihara 2017, Fechter-Leggett et al 2016, Wang et al 2016) explicitly analyzed morbidity, and four others (Weisskopf et al 2002, Carter et al 2005, Nitschke et al 2011, Sheridan and Lin 2014) considered both mortality and morbidity outcomes. Morbidity data included out-of-hospital cardiac arrests (Onozuka and Hagihara 2017), heat stroke hospitalizations (Wang et al 2016, Carter et al 2005), EMS runs (Weisskopf et al 2002), heat stroke illness (Fechter-Leggett et al 2016). Some studies found reductions in heat-related morbidity outcomes (Fechter-Leggett et al 2016, Weisskopf et al 2002), but others showed increasing trends with regard to heat stroke hospitalizations (Wang et al 2016) and ambulance call outs (Nitschke et al 2011). In Japan, Onozuka and Hagihara (2017) found no temporal variation in out-of-hospital cardiac arrests. Carter et al (2005) found reductions in heat exhaustion cases but a five-hold increase in heat stroke hospitalizations on US military bases. Sheridan and Lin (2014) show a general decrease in heat-related mortality over time in New York City concomitant with an increase in heat-related hospital admissions.

Despite the overall decrease in heat-related impacts, health risks clearly vary across different illness types and demographic confounders. Physiological mechanisms associated with heat-health relationships may help explain some of these differences, and the impact of short-term mortality displacement or harvesting also plays a role in some of the heat-heath relationships (Sheridan and Lin 2014). Such differences reflect the geographical variability in heat-related health outcomes as well as the complex interaction of non-climate factors including air quality, pre-existing health status, and socio-economic status.

Various technological, infrastructural, and societal mechanisms help explain the general decrease in heat-related impacts (Davis et al 2004, Davis et al 2002). For instance, Nordio et al (2015) showed a lower risk of death associated with the increased use of air conditioning. Each 50% increase in air conditioning prevalence lead to a 1.37% decrease in heat effect on mortality. Other studies however, indicated no significant relationship between health outcome and air conditioning (Heo et al 2016, Ng et al 2016, Bobb et al 2014). While air conditioning offers a mitigation opportunity, Smoyer (1998) notes that the cost can serve as an obstacle in reducing adverse heat-health outcomes. The most vulnerable—the poor or those living on fixed incomes—may not be able to afford to operationalize the air conditioning.

The occurrence of extreme heat events is often a spur for implementation of mitigation plans, and where heat mortality has decreased over time, this is often cited as a contributing factor to reduced heat sensitivity (Tan et al 2007, Benmarhnia et al 2016, Pascal et al 2012, Rey et al 2007, Schifano et al 2012, de'Donato et al 2015), although it can be difficult to make specific attribution given the myriad confounding variables (Toloo et al 2013). In response to the extreme heat in France in 2003, extra vigilance was prospectively connected to an overall decline in the total mortality rate (Toulemon and Barbieri 2008, Rey et al 2007). Several studies have confirmed a decrease in heat-related mortality in Italy following implementation of a heat warning system after 2003 (Michelozzi et al 2006, Morabito et al 2012, Schifano et al 2012), similar to Ebi et al (2004) for Philadelphia, USA. Linares et al (2015), however, note that the decrease in mortality in Spain following implementation of a heat warning system in 2004 was inconclusive and varied spatially. Miron et al (2008) draws attention to the need to monitor and review such systems. With increasing elderly populations and in the context of a changing climate, such interventions must account for changing demographics (Hess and Ebi 2016).

Many studies have suggested the decreased heat vulnerability is a function of broader societal changes, such as increased improved living standards (e.g. Tan et al 2007, Åström et al 2013a), healthcare resources and improved economic status (e.g. Kyselý and Kříž 2008, Ng et al 2016), long-term adaptation (e.g. Muthers et al 2010b), and general overall awareness (e.g. Ekamper et al 2009, Gabriel and Endlicher 2011). However, despite the recent declines in heat sensitivity, many of the analyzed studies note a stabilization in recent years (e.g. Miron et al 2008, Sheridan et al 2009, Åström et al 2013b), and thus within the context of the reasons for the decline, questions are raised as to the ultimate level of plausible acclimatization.