Temperature-related mortality: a systematic review and investigation of effect modifiers

We conducted a systematic search using a MEDLINE/PubMed database for population-based studies of exposure to heat or high temperature, cold, heat waves, and cold spells published between 1980 and 2017. We performed four searches: (1) ('heat' or 'temperature' or 'cold' or 'heatwave' or 'heatwaves' or 'heat-wave') AND (mortality or death); (2) modif* AND (effect or effects or impact or impacts) AND ('heat' or 'temperature' or 'cold' or 'heatwave' or 'heatwaves' or 'heat-wave'); (3) ('climate' or 'weather' or 'temperature') AND (mortality or death); and (4) ('climate' or 'weather' or 'temperature') AND (effect or effects or impact or impacts) AND modif*; where * indicates any combination of subsequent letters. The systematic search was conducted with consideration of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines (Moher et al 2015).

We selected studies meeting the following inclusion criteria. Studies had to: (1) be population-based; (2) consider exposure to heat or high temperature, cold, heat waves, or cold spells; (3) explore mortality; (4) examine effect modification; (5) be peer-reviewed; (6) be written in English; and (7) be published from 1980 to 2017, inclusive. Both single-city and multicity studies were included.

Next, we excluded studies by screening of titles and abstracts based on the inclusion criteria. We then reviewed the full texts of remaining articles. Figure 1 provides a flow diagram for the identification and selection of studies. We identified 728 studies for full-text screening. We extracted information of each article's study location, time frame, health outcome (i.e. total, cardiovascular, and respiratory mortality), study population, temperature exposure metric (e.g. mean temperature, maximum temperature), increment of exposure used in effect estimates (e.g. unit increase, comparison of percentiles), lag structure (i.e. delayed effects of temperature on mortality) (e.g. previous day or average of previous days), effect modification studied, and results for main findings and effect modification factors. For results without quantitative results, such as graphical presentation of findings, we contacted the original study authors at least twice. We classified results into two categories: (1) studies investigating associations of single days of high or low ambient temperature with mortality and (2) studies investigating associations of consecutive days of extreme temperatures (i.e. heat waves, cold spells) with mortality. We used results from the key findings presented by study authors, as originally reported.

Zoom In Zoom Out Reset image size

One of our objectives is to investigate effect modification of the association between ambient temperature and mortality. We included all effect modifiers from all eligible studies included in this review. We summarized findings on effect modification based on individual- and community-level characteristics. We identified the following potential individual- and community-level effect modifiers to temperature-mortality associations: (1) individual-level: sex, age, education, place of death (in or out of hospital), occupation, race, body mass index (BMI), marital status, and chronic conditions; and (2) community-level: SES based on income, gross domestic product (GDP), fraction living in poverty, or higher proportion of people living in low SES, population density, latitude, urban/rural, prevalence of heating systems, prevalence of air conditioning (AC), climatic condition, healthcare facilities, housing quality, proportion of green space or vegetation, proximity to water, previous winter mortality, air pollution, and reduced electricity consumption.

Most analyses of heat waves or cold spells compared risk during heat waves or cold spells to risk on non-heat wave or cold spell days, although conditions during heat waves or cold spells can vary. Therefore, some studies have investigated how the health impacts of heat waves or cold spells differ by heat wave or cold spell characteristics. We evaluated findings on effect modification of the heat wave/cold spell-mortality association, based on heat wave/cold spell characteristics as a modification of the exposure for heat waves/cold spells studies (e.g. length of heat waves). For effect modifiers, we summarized overall evidence using the following categories: no evidence, weak evidence, limited/suggestive evidence, and strong evidence. We assigned a category based on the quantity of studies providing consistent evidence compared with conflicting findings for the overall summary of evidence. A similar approach was used in previous studies (Bell et al 2014). We also provided the classification criteria for overall summary of evidence for effect modification (table S1 is available online at stacks.iop.org/ERL/14/073004/mmedia).

We identified age as the most consistent effect modifier of the association between ambient temperature and mortality, with higher risks from heat or cold exposure for the elderly. Vulnerability for the elderly may relate to physiological changes, different activity patterns, housing quality, and social factors. Older persons can have limited thermoregulatory response, different prevalence of comorbidities, live alone, different access to AC or heating systems, and limited access to healthcare facilities and social services (Hajat and Kosatky 2010, Bunker et al 2016). Exposure from heat or cold in the elderly may trigger adverse airway responses, increased inflammatory factors, and increased susceptibility to infection (Eccles 2002, Anderson et al 2013).

We found strong evidence of higher temperature-mortality risks associated with heat exposure for women than for men. This may result from differences in physiology, exposure patterns and occupational exposure between men and women. However, some studies reported no difference in risk between men and women, or a higher estimated effect for men (Goggins et al 2013, Bai et al 2014, Zeng et al 2017). Education level is often used as an indicator of SES, and many studies reported higher mortality risks for those with no or low education (Borrell et al 2006, Kaiser et al 2007, Son et al 2012, Yang et al 2013, Bai et al 2014, Chen et al 2015, Huang et al 2015, Zeng et al 2017). However, some studies also identified no difference or higher risk for those with higher education level (Ma et al 2012, 2015, Wang et al 2014, Ding et al 2016, Isaksen et al 2016). Socioeconomic factors can be correlated with each other and one specific indicator does not fully represent actual SES, which can relate to many different factors such as multiple sources of income, family income, and historical SES (O'Neill et al 2003, Mechanic 2007). Both individual and community SES can play a role in health. Further, other demographic factors can be related to SES. For example, sex may be related with other SES factors. In our previous work in Korea, women study participants were older and less educated (Son et al 2012). Future research considering multiple SES indicators and related factors, and the relationships among these factors, is needed to better understand how they modify the temperature-mortality association.

We also found limited or suggestive evidence of effect modification for other individual-level factors such as place of death, occupation, race, marital status, and chronic conditions. Findings suggest higher risks with out of hospital death (Medina-Ramón et al 2006, Son et al 2011, 2016a Huang et al 2015, Ma et al 2015, Zhang et al 2016), with lower level employment (Anderson and Bell 2009, Yang et al 2012, 2013, Xu et al 2013, Heo et al 2016), in black populations (O'Neill et al 2005, Medina-Ramón et al 2006, Kaiser et al 2007, Madrigano et al 2015a, Lee et al 2016), and for unmarried persons (Stafoggia et al 2006, 2008, Schifano et al 2009, Gronlund et al 2015, Qiu et al 2016) from heat and/or cold exposure.

Higher heat-related mortality risk was associated with higher population density (Medina-Ramón and Schwartz 2007, Ma et al 2015, Madrigano et al 2015b), lower prevalence of heating systems (Curriero et al 2002), poor healthcare facilities (Huang et al 2015), far from the water (Burkart et al 2016), old and poor housing quality (Xu et al 2013), and high air pollution levels (Ren et al 2008, Breitner et al 2014, Li et al 2015). Studies used a range of variables to define community-level SES including indirect variables anticipated to be correlated with SES: multiple indicators, income, GDP, poverty, and proportion of elderly, Black/African-American, or low educated people. Findings suggest higher heat-related mortality risk with low community-level SES (Wichmann et al 2011, Ma et al 2014, Rosenthal et al 2014, Lim et al 2015, Madrigano et al 2015b), in middle- or higher latitude (Curriero et al 2002, Ma et al 2014, Xiao et al 2015, Lim et al 2016), in urban areas (Anderson and Bell 2009, Burkart et al 2014, Urban et al 2014, Ma et al 2015), with lower AC prevalence (O'Neill et al 2005, Medina-Ramón and Schwartz 2007, Anderson and Bell 2009, Basu and Malig 2011, Nordio et al 2015, Chen et al 2016), in communities with warmer climates (Rosenthal et al 2014, Huang et al 2015, Madrigano et al 2015a), in areas with less proportion of green areas or vegetation (Xu et al 2013, Ma et al 2014, Gronlund et al 2015, Madrigano et al 2015a, Burkart et al 2016, Son et al 2016b), and with low previous winter mortality (Rocklöv et al 2009, Stafoggia et al 2009, Ha et al 2011, Qiao et al 2015).

Relatively less evidence was reported for risk from cold or cold spells than risk of exposure from high ambient temperature or heat waves. Some studies reported that effects from cold temperature are higher in regions with warmer climate (Curriero et al 2002, Ma et al 2014). Findings suggest higher cold-related mortality risk in lower latitude (Curriero et al 2002, Ma et al 2014, Xiao et al 2015), and in rural areas (Zhou et al 2014, Urban et al 2014). Although we found weak or limited/suggestive evidence on effect modification by some community-level factors, these results reflect the current scientific evidence, which is related to the lack of a sufficient number of adequate studies. More work investigating the community-level effect modification on the temperature-mortality association is required to fully understand this issue, and our findings give directions for future work.

Our review identified several challenges to comparing and summarizing results across the studies. Many studies used different definitions to define a disease or exposure. Each mortality cause (i.e. total, cardiovascular, and respiratory mortality) were identified through five or more different sets of ICD codes, an issue that has been previously identified (Ji et al 2011). To define heat waves or cold spells, many studies applied various definitions, and there exists no standard definition of a 'hot day' or 'heat wave', etc in scientific research or in policy.

Our findings showed that most studies on the association between ambient temperature and mortality were based on areas in Asia, Europe and North America. Current epidemiologic evidence is very limited in many other regions. There is substantial epidemiologic evidence that ambient temperature is associated with mortality and that the relationship is nonlinear across many regions with different climate and populations (Guo et al 2016, Ryti et al 2016). However, the magnitude of the associations differs by several characteristics such as population structure and region. Spatial variability in effect estimates across studies may relate with several factors such as regional climate, population characteristics (e.g. population's sex or age structure), presence of heat warning systems, and acclimatization to local climate. Thus, more studies are needed worldwide, especially in understudied regions, and should be conducted in more varied climate zones to better understand the association between ambient temperature and mortality outcomes and to allow further investigation of effect modification.

We were unable to quantify the overall effect modification for the mortality risk from exposure to heat, cold, or extreme temperature events across studies. There were insufficient studies with similar approaches to conduct meta-analysis by each category. However, we provided detailed information such as mortality category, location, exposure metric and increment, and effect modifiers studied for each study. This information from our review provides guidance on which effect modifiers and populations show consistent patterns in general although there may exist differences across regions and these findings could help identify critical areas for future study. One of our most important contributions is that findings for effect modifiers varied by region and population, which highlights the importance of local studies. To provide location-specific information for governments to identify target areas and most effectively use resources, future study such as vulnerability mapping is needed regarding which populations and areas are vulnerable to temperature-mortality associations in a given location. Publication bias may exist; studies that did not find statistical significance may be selectively submitted or published. There may be other relevant studies exploring important issues of risk of mortality in relation to metrics of temperature and effect modification that did not include in this review due to our selection criteria.

In our review, we found that some characteristics such as AC, housing quality, pre-existing conditions may associate with increased vulnerability. Disparities in temperature-mortality association could be mediated by various factors (Gronlund 2014). Providing evidence on the behavioral modifiers such as individual's actions or ability to reduce the exposure, physiologic changes or adaptation, which may vary by climate zone could play a critical role for decision makers to develop appropriate interventions and policies. Future research using multiple relevant disciplines and resources may contribute to better understanding of the characteristics of vulnerability, mechanisms, and potential mediators of these relevant characteristics and risk factors.

The effect modifiers identified in this review such as SES include multiple components and may influence the risk of mortality through multiple pathways such as environmental exposure, behavior/lifestyle, and healthcare resources. Previous study evaluated whether areas with higher heat vulnerability experienced higher rates of morbidity and mortality on abnormally hot days. They found that heat vulnerability index may be an indicator of overall health vulnerability, regardless of exposure to heat, although greater vulnerability to heat was observed in some areas (Reid et al 2012). Variability in health risk across studies may result from differences in location characteristics, methodologies used, and population structure over time. Understanding local differences about which variables contribute most in some regions but not in other regions is critical thus future studies could consider local information for assessing vulnerability at local scales.

To our knowledge, this is the largest systematic review of effect modification of the temperature effect on mortality considering heat, cold, and extreme temperature events with a focus on vulnerability. Consistent with our findings, previous reviews reported several factors with higher vulnerability. A study by Hajat and Kosatky (2010) suggested that aging population, higher population density, and lower GDP were associated with higher risk of heat exposure on mortality. Another review found strong evidence of heat-related vulnerability for the elderly and low SES groups (Benmarhnia et al 2015). Our findings provide evidence on which populations are most vulnerable and which factors affect vulnerability for the temperature-mortality association, and identify critical areas for future study. Better understanding about the vulnerable populations for the association between ambient temperature and mortality benefits policy makers in establishing appropriate strategies to reduce the health burden from current day weather extremes. Findings also inform our understanding of the consequences of climate change. Future studies could consider the impacts of temperature on health under a changing climate in relation to vulnerable populations and other relevant effect modifiers, and temporal changes in these factors.