Articles Trends in temperature-related age-specific and sex-specific mortality from cardiovascular diseases in Spain: a national time-series analysis Hicham Achebak, Daniel Devolder, Joan Ballester Summary Background Climate change driven by human activities has increased annual temperatures in Spain by around 1°C Lancet Planet Health 2019; since 1980. However, little is known regarding the extent to which the association between temperature and mortality 3: e297–306 has changed among the most susceptible population groups as a result of the rapidly warming climate. We aimed to Published Online June 20, 2019 assess trends in temperature-related cardiovascular disease mortality in Spain by sex and age, and we investigated the http://dx.doi.org/10.1016/ association between climate warming and changes in the risk of mortality. S2542-5196(19)30090-7 See Comment page e281 Methods We did a country-wide time-series analysis of 48 provinces in mainland Spain and the Balearic Islands between Centre for Demographic Jan 1, 1980, and Dec 31, 2016. We extracted daily cardiovascular disease mortality data disaggregated by sex, age, and Studies, Autonomous province from the Spanish National Institute of Statistics database. We also extracted daily mean temperatures from the University of Barcelona, European Climate Assessment and Dataset project. We applied a quasi-Poisson regression model for each province, Barcelona, Spain (H Achebak MSc, controlling for seasonal and long-term trends, to estimate the temporal changes in the province-specific temperature- D Devolder PhD); and Climate mortality associations with distributed lag non-linear models. We did a multivariate random-effects meta-analysis to and Health Program, Barcelona derive the best linear unbiased prediction of the temperature–mortality association and the minimum mortality Institute for Global Health, Barcelona, Spain (H Achebak, temperature in each province. Heat-attributable and cold-attributable fractions of death were computed by separating the J Ballester PhD) contributions from days with temperatures warmer and colder than the minimum mortality temperature, respectively. Correspondence to: Dr Joan Ballester, Climate and Findings Between 1980 and 2016, 4 576 600 cardiovascular deaths were recorded. For warm temperatures, the increase in Health Program, Barcelona relative risk (RR) of death from cardiovascular diseases was higher for women than men and higher for older individuals Institute for Global Health, (aged ≥90 years) than younger individuals (aged 60–74 years), whereas for cold temperatures, RRs were higher for men 08003 Barcelona, Spain joan.ballester@isglobal.org than women, with no clear pattern by age group. The heat-attributable fraction of cardiovascular deaths was higher for women in all age groups, and the cold-attributable fraction was larger in men. The heat-attributable fraction increased with age for both sexes, whereas the cold-attributable fraction increased with age for men and decreased for women. Overall minimum mortality temperature increased from 19·5°C between 1980 and 1994 to 20·2°C between 2002 and 2016, which is similar in magnitude to, and occurred in parallel with, the observed mean increase of 0·77°C that occurred in Spain between these two time periods. In general, between 1980 and 2016, the risk and attributable fraction of cardiovascular deaths due to warm and cold temperatures decreased for men and women across all age groups. For all the age groups combined, between 1980–94 and 2002–16, the heat-attributable fraction decreased by –42·06% (95% empirical CI –44·39 to –41·06) for men and –36·64% (–36·70 to –36·04) for women, whereas the cold-attributable fraction was reduced by –30·23% (–30·34 to –30·05) for men and –44·87% (–46·77 to –42·94) for women. Interpretation In Spain, the observed warming of the climate has occurred in parallel with substantial adaptation to both high and low temperatures. The reduction in the RR and the attributable fraction associated with heat would be compatible with an adaptive response specifically addressing the negative consequences of climate change. Nevertheless, the simultaneous reduction in the RR and attributable fraction of cold temperatures also highlights the importance of more general factors such as socioeconomic development, increased life expectancy and quality, and improved health-care services in the country. Funding None. Copyright © 2019 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY-NC-ND 4·0 license. Introduction temperatures and less exposed to moderate and extreme Increasing concentrations of greenhouse gases due to cold temperatures, which has serious implications for human activities are inducing an unequivocal rise in various health outcomes.2,3 global temperatures.1 As a consequence of this anthro­ Several studies addressing the potential future impacts pogenic warming, the world population is progres­ of climate warming on temperature-related mortality sively more exposed to moderate and extreme warm have suggested an increase in heat-related deaths and a www.thelancet.com/planetary-health Vol 3 July 2019 e297 Articles Research in context Evidence before this study magnitude of the increases in minimum mortality temperature We searched PubMed from database inception until and annual mean temperatures support the hypothesis that the Dec 1, 2018, for articles published in English using the search global warming observed in recent decades has been terms “temperature”, ”heat”, “cold”, AND “mortality”, “deaths”, accompanied by substantial adaptation to both warm and cold “cardiovascular mortality”. Studies that assessed temporal temperatures in Spain. variations in the association between heat and mortality Implications of all the available evidence showed a reduction in the relative risk (RR) of mortality due to In Spain, the observed climate warming has occurred in heat across many settings, mostly in developed countries, parallel with substantial adaptation to both warm and cold despite the observed rise in temperatures. By contrast, some temperatures. The reduction in the risk and attributable evidence has indicated a decline in cold-attributable mortality fraction of cardiovascular deaths associated with heat would risk in the past 30 years, but few studies have investigated the be compatible with an adaptive response specifically impact of cold temperatures and available evidence is addressing the negative consequences of climate change. inconclusive. Little is known about the effect of heat and cold Nevertheless, the simultaneous reduction in the risk of exposure on mortality risk and attributable mortality among mortality and impact of cold temperatures also highlights the population subgroups, and particularly the most susceptible importance of more general factors such as socioeconomic populations, by sex, age, and cause of death. development, increased life expectancy and quality, and Added value of this study improved health-care services in the country. These adaptions This is the first study to comprehensively assess the impact of to cold temperatures are not necessarily associated with an the 1°C increase in ambient temperature that has been adaptive response specifically addressing the negative observed in Spain since 1980, and the effect of this increase on consequences of climate change (ie, they would have also the Spanish population by sex, age, and specific cause of death. occurred in a non-warming scenario), and therefore, our The effect of warm and cold temperatures on cardiovascular results show that this adaptive response to climate change disease mortality in Spain, either expressed as RR (ie, level of might be more limited than the contribution of more general vulnerability) or attributable fraction (ie, mortality burden factors. Thus, the substantial adaption observed within the attributable to non-optimum temperatures), differed by sex Spanish population during this period of rapid climate and age group and decreased over the study period for both warming might not be sustained at higher ambient sexes and across all age groups. The concurrent timing and temperatures when warming begins to occur at a faster rate. reduction in cold-related deaths,4,5 resulting in a positive observed rise in temperatures, and the associated or negative long-term net effect in mortality depending increase in the frequency, intensity, and duration of on the location and magnitude of the warming.6 extreme heat events. By contrast, some evidence13,14 However, most of these analyses assumed no changes in suggests that the risk of cold-related mortality has vulnerability (expressed as relative risk [RR]) over declined in recent years, but few studies have assessed time. The eventual incidence of climate warming on the impact of cold temperatures and evidence remains temperature-related mortality will not only depend on inconclusive.15 Moreover, little is known about the effect the change in the levels of exposure to warm and cold of cold and heat exposure on the risk and impact temperatures, but also on any underlying change in the (ie, heat-attributable mortality and cold-attributable vulnerability of the exposed populations.7 mortality) among population subgroups, and parti­cularly A decrease in risk is often considered a sign of the most susceptible populations by sex, age, and cause adaptation, occurring as a result of a physiological of death. From a public health perspective, analysis of acclimatisation response within the population to cause-specific mortality is more important than analysis changes in temperature (intrinsic adaptation), or through of all-cause mortality since total mortality includes many non-climate driven factors contributing to the reduction causes of death that have a weak association with of the risks (extrinsic adaptation), such as socioeconomic temperature, which might hide diverging patterns development or improved health-care services.8 In a between different types of cause-specific mortality, and warming climate, the reduction in heat-related mortality because the mechanisms by which ambient temperatures risks is likely to be the result of a combination of both trigger mortality might vary for different causes of death. adaptation processes, but an eventual decrease in In this study, we assessed sex-specific and age-specific vulnerability to cold temperatures in the context of rising trends in heat-related and cold-related cardiovascular temperatures would only be explained by non-climate disease mortality between 1980 and 2016 in Spain, where driven mechanisms. a rapid increase in annual temperatures (0·33°C per To date, many studies9–12 have reported a reduction in decade) has been observed since 1980. Findings from the RR of mortality due to heat across a number of this study have important implications for health policies settings, mostly in developed countries, despite the designed to adapt to warming temperatures in a country e298 www.thelancet.com/planetary-health Vol 3 July 2019 Articles that has been characterised as a major climatic hotspot to derive estimates of province-specific temperature– within the Mediterranean region.16 mortality associations, reported as RR by sex and age group. The models included a natural cubic B-spline of Methods time with 8 degrees of freedom (df) per year to adjust for Data sources seasonal and long-term trends, and a categorical variable For this country-wide time-series analysis, we extracted to control for the day of the week. We used a distributed daily cardiovascular disease mortality data (as the primary lag non-linear model to model the associations between cause of death) disaggregated by sex, 15-year age groups temperature and mortality. This model is based on the (0–14, 15–29,…, 75–89, ≥90 years), and province for definition of a cross-basis function combining exposure– the study period of Jan 1, 1980, to Dec 31, 2016. Data response and lag-response associations.18 We modelled were extracted from the Spanish National Institute of the exposure–response curve with a quadratic B-spline, Statistics. Causes of death were coded by the Inter- with one internal knot placed at the 75th percentile of the national Classification of Diseases, ninth revision (ICD-9) daily temperature distribution, and the lag-response codes 390–459 for the years 1980–98 and ICD tenth curve was modelled with a natural cubic B-spline with an revision (ICD-10) codes I00–I99 for the years 1999–2016, intercept and three internal knots placed at equally but both classifications contained the same disaggregation spaced values in the log scale. We extended the lag period of causes of death. No obvious differences in cardio­ to 21 days to account for the long-delayed effects of vascular mortality from specific diseases were identified cold temperatures and short-term harvesting (ie, deaths between years 1998 (using ICD-9) and 1999 (using ICD-10; brought forward by only a few days due to temperature).19 appendix p 6). The overall effect of temperature on a specific day on the See Online for appendix We obtained daily mean temperatures from the E-OBS RR of death was defined as the sum of the effects on that For more on the E-OBS gridded gridded dataset (version 16; resolution of 0·25° × 0·25°) day and the 21 subsequent days. We did not use additional dataset see http://ensembles- eu.metoffice.com from the European Climate Assessment and Dataset, knots for the exposure–response function (ie, tenth and For the European Climate and transformed these temperatures into regional 90th percentiles19) because the estimates of RR at extreme Assessment and Dataset see estimates.17 Both mortality and temperature datasets had warm and cold temperatures were not sensitive to the http://www.ecad.eu no missing values. choice of only one internal knot. The Poisson regression model for the whole study period was as follows: Statistical analysis Ambient temperatures are mainly correlated with cardio­ LogE (Y)=intercept + cb + dow + S (time, 8 df × year) vascular and respiratory diseases, but the two disease types have evolved differently during the study period where Y denotes the series of daily mortality counts; (1980–2016) in Spain,12 and therefore, we separated cb the cross-basis matrix produced by distributed lag non- the analyses for a more detailed description of their linear model; dow the day of the week; and S the natural evolution, their relationship with tempera­tures, and the cubic B-spline of time. The modelling choices were tested recent evolution with warming condi­ tions. Statistical in sensitivity analyses by varying the number of knots in analysis was done in two stages. We first applied standard the exposure–response function, the number of lag days, time-series quasi-Poisson regression models allowing for and the number of degrees of freedom used to control for overdispersion for the whole study period (1980–2016) the seasonal and long-term trends (appendix pp 204, 205). and data subsets of 15-year moving periods (1980–94, In the second stage, we did a multivariate random- 1981–95,…, 2002–16) in each of the 48 Spanish provinces effects meta-analysis to estimate the mean RR values A June–September B October–May 12 1980–94 (mean 20·48 °C) 1980–94 (mean 9·61 °C) 2002–16 (mean 21·16 °C) 2002–16 (mean 10·43 °C) 10 8 Frequency (%) 6 4 2 0 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 –11 –8 –5 –2 1 4 7 10 13 16 19 22 25 28 Temperature (°C) Temperature (°C) Figure 1: Distribution of daily mean temperatures in Spain between 1980 and 1994 and 2002 and 2016 Daily mean temperatures between June and September (A) and between October and May (B). www.thelancet.com/planetary-health Vol 3 July 2019 e299 Articles temperature from the country-level and province-specific A Overall (all ages) B Men (all ages) C Women (all ages) 3·0 RR curves. 1980–94 2·8 1991–2005 The mortality burden attributable to non-optimum 2·6 2·4 2002–16 temperatures by period and province, reported as the 2·2 2·0 attributable fraction of deaths, was estimated using the 1·8 methodology of Gasparrini and Leone.21 The overall RR RR 1·6 1·4 corresponding to each day of the series was used to 1·2 1·0 calculate the attributable fraction of deaths on that day 0·8 0 and the next 21 days. We then computed the daily attributable number of deaths by multiplying the daily D Overall (age ≥60 years) E Men (age ≥60 years) F Women (age ≥60 years) attributable fraction by the daily number of deaths. The 3·0 overall number of attributable deaths caused by 2·8 2·6 non-optimum temperatures was given by the sum of 2·4 the contributions from all days of the series with 2·2 2·0 temperatures higher or lower than the value of minimum 1·8 RR 1·6 mortality temperature derived from the best linear 1·4 1·2 unbiased prediction, and its ratio with the total number of 1·0 0·8 deaths provided the temperature-attributable fraction. 0 The components attributable to cold and warm temperatures were in turn computed by separating the G Overall (age 60–74 years) H Men (age 60–74 years) I Women (age 60–74 years) associations corresponding to days with temperatures 3·0 lower or higher than the minimum mortality temperature, 2·8 2·6 respectively. We calculated 95% empirical CIs (eCIs) of 2·4 2·2 attributable risk using Monte Carlo simulations. 2·0 1·8 To ensure that the variation in minimum mortality RR 1·6 1·4 temperature was not an important factor explaining 1·2 the temporal changes in heat-attributable and cold- 1·0 0·8 attributable deaths and the differences between sexes 0 and age groups, we alternatively computed the attri­ butable fractions for reference ranges of temperatures J Overall (age 75–89 years) K Men (age 75–89 years) L Women (age 75–89 years) 3·0 (ie, temperatures warmer than the 85th percentile for the 2·8 definition of heat-related mortality, and colder than the 2·6 2·4 median for cold-related mortality). We also considered 2·2 2·0 the same reference point for computing the heat- 1·8 attributable and cold-attributable fractions for both men RR 1·6 1·4 1·2 and women (ie, the highest minimum mortality 1·0 temperature for men and women in each province by 0·8 age group). 0 We did not assess the associations between temperature M Overall (age ≥90 years) N Men (age ≥90 years) O Women (age ≥90 years) and cardiovascular mortality specifically for individuals 3·0 younger than 60 years because of the small number of 2·8 2·6 daily death counts reported for those age ranges in most 2·4 2·2 provinces, which led the model fitting to fail. Instead, we 2·0 analysed data for individuals aged 60–74, 75–89, and 1·8 RR 1·6 90 years and older because the number of deaths among 1·4 1·2 these age groups was large enough to achieve model 1·0 0·8 convergence, and because they represent the majority of 0 cardiovascular deaths in Spain (93%). –5 0 5 10 15 20 25 30 –5 0 5 10 15 20 25 30 –5 0 5 10 15 20 25 30 All statistical analyses were done with R software Temperature (°C) Temperature (°C) Temperature (°C) (version 3.4.3) using the packages splines (“bs” for Figure 2: RR of death from cardiovascular diseases in Spain for 15-year time periods between 1980 and 2016 quadratic splines and “ns” for natural cubic splines), 95% empirical CIs are shown in the appendix (pp 9, 10). RR=relative risk. dlnm (for the first-stage regression), and mvmeta (for the second-stage meta-analysis). associated with the temperature–mortality curves across provinces,20 and to derive the best linear unbiased Role of the funding source prediction of the temperature–mortality associations in There was no funding source for this study. The each location. We then extracted the minimum mortality corresponding author had full access to all the data and e300 www.thelancet.com/planetary-health Vol 3 July 2019 Articles had final responsibility for the decision to submit for A By sex (all ages) B By sex (age ≥60 years) publication. 21·0 Men 21·0 Women Results 20·5 Overall 20·5 We collected data from 48 provinces in mainland Spain and the Balearic Islands between Jan 1, 1980, and 20·0 20·0 MMT (°C) Dec 31, 2016 (appendix p 4). The dataset included 4 576 600 cardiovascular disease deaths (38·2% of deaths 19·5 19·5 due to natural causes). Cardiovascular mortality rates 19·0 19·0 largely decreased over the study period for both men and women, and for all the age groups, particularly among 18·5 18·5 people aged 90 years or older (appendix p 5). The decline in cardiovascular mortality was due to the reduction in C By sex (age 60–74 years) D By sex (age 75–89 years) deaths from cerebrovascular diseases, atherosclerosis, 21·5 21·5 and acute myocardial infarction observed over the study 21·0 21·0 period (appendix p 6). In parallel, the distribution of hot and cold temperatures has shifted towards higher values, 20·5 20·5 MMT (°C) generally with more moderate and extreme warm days 20·0 and fewer moderate and extreme cold days between 2002 20·0 and 2016 than between 1980 and 1994 (figure 1). 19·5 Temperature-related cardiovascular disease mortality 19·5 19·0 curves by sex and age group for the whole study period had an asymmetric U or V shaped curve (appendix 19·0 18·5 pp 7, 8), indicating a monotonically increasing mortality risk for temperatures higher and lower than the case- E By sex (age ≥90 years) F By age group 21·5 21·5 60–74 years specific minimum mortality temperature. The slope for 75–89 years warm temperatures was in all cases steeper than the 20·5 ≥90 years slope for cold temperatures, and varied by sex and age 20·5 19·5 group. The slopes for women were steeper than those for MMT (°C) men and the slopes for older age groups (75–89 years and 18·5 19·5 ≥90 years) were steeper than those for the younger age 17·5 groups. By contrast, the slopes for cold temperatures 18·5 were slightly steeper for men than for women, with the 16·5 exception of the 60–74 year age group, and showed no 15·5 17·5 clear pattern by age group. 19 –94 19 –96 86 98 88 00 90 02 92 04 94 06 19 –20 8 98 10 20 012 20 –14 6 19 –94 19 –96 86 98 19 –20 0 90 02 9 04 94 06 8 98 10 20 012 20 –14 6 88 0 96 0 –1 96 0 –1 All temperature–mortality risk functions showed a 19 –20 19 –20 19 –20 19 –20 19 –20 19 –20 19 –20 19 –20 19 2–20 19 –20 19 84– 19 4– 00 00 02 02 80 80 –2 –2 82 82 8 19 19 substantial reduction in risk of cardiovascular death at both hot and cold temperatures during the study period Figure 3: Minimum mortality temperature for cardiovascular diseases by sex and age, 1980–2016 (figure 2; appendix pp 9–99). For example, the overall MMT=minimum mortality temperature. RR of death from cardiovascular disease at the first temperature percentile decreased from 1·618 (95% eCI across all age groups (figure 3). The pooled overall 1·558–1·681) during the 1980–94 time period (first minimum mortality temperature increased from 19·5°C percentile 1·20°C) to 1·348 (1·289–1·409) during the in 1980–94 to 20·2°C in 2002–16, which is similar in 2002–16 time period (first percentile 1·65°C), and from magnitude to the observed mean temperature increase of 1·515 (1·416–1·621) during the 1980–94 time period 0·77°C that occurred between these two time periods. to 1·277 (1·226–1·330) during the 2002–16 time period at The minimum mortality temperature decreased with the 99th percentile (99th percentiles 26·27°C [1980–94] and age, and, with the exception of the 60–74 year age group, 26·57°C [2002–16]; figure 2A). The RR corresponding to was higher for men than for women. the first temperature percentile (first percentile 1·41°C) The increase in minimum mortality temperature for the whole study period decreased from 1·702 between 1980 and 2016 strongly correlated with the (1·633–1·775) between 1980 and 1994 to 1·408 (1·337–1·484) increase in mean temperatures observed for the same between 2002 and 2016, and the RR corresponding to the period (subperiod relationship slope range 0·90–1·35; 99th temperature percentile (99th percentile 26·36°C) figure 4). For the whole study period, the association decreased from 1·962 (1·770–2·174) between 1980 and between mean temperatures and minimum mortality 1994 to 1·457 (1·369–1·551) between 2002 and 2016. temperatures for the 48 provinces was also strong and Between 1980 and 2016, the minimum mortality linear (slope range 0·88–1·01, p<0·001; appendix p 100). temperature progressively increased for both sexes, and The magnitude of the association between mean www.thelancet.com/planetary-health Vol 3 July 2019 e301 Articles A Overall (all ages) B Men (all ages) C Women (all ages) 20·5 β=1·08 21·0 β=0·90 20·5 β=1·30 R=0·68 R=0·65 R=0·72 20·5 20·0 20·0 MMT (°C) 20·0 19·5 19·5 19·5 19·0 19·0 19·0 18·5 D Overall (age ≥60 years) E Men (age ≥60 years) F Women (age ≥60 years) 20·5 β=1·03 21·0 β=0·92 20·5 β=1·35 R=0·71 R=0·78 R=0·78 20·0 20·0 20·5 MMT (°C) 19·5 19·5 20·0 19·0 19·0 19·5 18·5 13·00 13·25 13·50 13·75 14·00 14·25 13·00 13·25 13·50 13·75 14·00 14·25 13·00 13·25 13·50 13·75 14·00 14·25 Temperature (°C) Temperature (°C) Temperature (°C) 1980–94 1981–95 1982–96 1983–97 1984–98 1985–99 1986–2000 1987–2001 1988–2002 1989–2003 1990–2004 1991–2005 1992–2006 1993–2007 1994–2008 1995–2009 1996–2010 1997–2011 1998–2012 1999–2013 2000–14 2001–2015 2002–16 Figure 4: Association between the mean temperature and minimum mortality temperature by time period MMT=minimum mortality temperature. β=slope. R=Pearson correlation. temperature and minimum mortality temperature The temporal changes in heat-attributable and cold- remained stable across the 48 provinces between attributable deaths and between-sex and between-age the 23 subperiods of 15 consecutive years (appendix differences were not affected by the variation in the pp 101, 102). minimum mortality temperature (appendix pp 200–03). Heat-attributable and cold-attributable fractions of All sensitivity analyses suggested that the results cardiovascular mortality reduced over the study period reported were not dependent on modelling assumptions. for both sexes and across all age groups (figure 5, figure 6; appendix pp 103–99), which are largely Discussion explained by the large reductions in RR (figure 2). For To the best of our knowledge, this is the first study to example, for all age groups combined, between 1980–94 assess the temporal changes in the effect of ambient and 2002–16, the heat-attributable fraction decreased by temperature on sex-specific and age-specific cardio­ –42·06% (95% eCI –44·39 to –41·06) for men and vascular mortality risk and attributable fraction, and to –36·64% (–36·70 to –36·04) for women, and the cold- determine whether adaptation to heat and cold occurred attributable fraction was reduced by –30·23% (–30·34 to in the context of rapid climate warming. The attributable –30·05) for men and –44·87% (–46·77 to –42·94) for fraction of cardiovascular deaths due to warm temper­ women (figure 5, figure 6). The attributable fraction due atures was higher for women than men in all age groups, to heat was generally larger for women than men in all and the attributable fraction due to cold temperatures the age groups, and the attributable fraction due to cold was higher in men than women, with the exception of temperatures was larger in men than women with the the 60–74 year age group. Moreover, heat-related cardio­ exception of the 60–74 year age group. The heat- vascular disease mortality increased with age for both attributable fraction increased with age for both sexes sexes, and cold-related cardiovascular disease mortality (figure 5G, H), whereas the cold-attributable fraction increased with age for men and decreased for women. increased with age for men (figure 6G) and decreased Our results also showed a progressive increase in the for women (figure 6H). For all age groups combined, minimum mortality temperature for both sexes and during the study period the between-sex differences due across all age groups during the study period, and a to heat decreased (figure 5A), whereas between-sex strong reduction in risk of cardiovascular mortality and differences due to cold temperatures remained stable attributable fraction due to heat and cold temperatures. (figure 6A). These results strongly support the hypothesis that the e302 www.thelancet.com/planetary-health Vol 3 July 2019 Articles climate warming observed in recent decades in Spain A By sex (all ages) B By sex (age ≥60 years) has been accompanied by substantial adaptation to both 0·04 Men hot and cold temperatures. Women In our study, the minimum mortality temperature Overall 0·03 increased simultaneously with the mean increase in Attributable fraction ambient temperature observed during the study period, for both sexes and across all age groups. This change 0·02 in minimum mortality temperature is consistent with earlier studies of non-external causes of mortality 0·01 and all-cause mortality done in France22 and Sweden,23 respectively. In France, the minimum mortality temper­ 0·01 ature increased by 0·4°C per 0·4°C increase in mean winter temperature, and per 0·6°C increase in summer C By sex (age 60–74 years) D By sex (age 75–89 years) temperature between 1982–95 and 1996–2009.22 These 0·04 findings support the hypothesis of human adaptation to increasing temper­atures.22 However, in our study, the 0·03 Attributable fraction concurrent timing of the increase in minimum mortality temperature and annual mean temperatures implies that 0·02 the tem­ poral evolution of heat-attributable and cold- attributable mortality essentially depends on the changes 0·01 in the shape and slopes of the exposure–response curves above and below the minimum mortality temperature, rather than on the warming itself. The increase in 0·01 minimum mortality temperature implies a time-varying definition of warm and cold temperatures, which E By sex (age ≥90 years) F By age group 0·05 parallels the shift in the temperature distribution towards 60–74 years 75–89 years warmer temperatures. We also found a strong positive 0·04 ≥90 years spatial correlation between mean temperature and Attributable fraction minimum mortality temperature, which has been widely 0·03 reported elsewhere,22,24 but additionally, we showed that this association has remained stable with time. 0·02 Physiological and behavioural adaptation of populations to local climate conditions is the main potential 0·01 explanation for the strong association between minimum 0 mortality temperature and annual mean temperatures, both in time and in space. G By age group (men) H By age group (women) The effects of warm and cold temperatures on mortality 0·05 from cardiovascular diseases in Spain, either expressed as RR (ie, level of vulnerability) or attributable fractions, 0·04 Attributable fraction decreased over the study period for both sexes and 0·03 generally for all age groups. Our results for warm temperatures are consistent with those reported in 0·02 previous studies.9,11,12,25 Results for cold temperatures are more difficult to discuss in the context of existing 0·01 literature since less evidence is available13,14 and that which exists remains inconclusive.15 In many studies, the 0 general reduction in risk and attributable mortality 19 –94 19 –96 86 98 88 00 90 02 92 04 94 06 19 –20 8 98 10 20 012 20 –14 6 19 –94 19 –96 86 98 19 –20 0 90 02 9 04 94 06 8 98 10 20 012 20 –14 6 88 0 96 0 –1 96 0 –1 19 –20 19 –20 19 –20 19 –20 19 –20 19 –20 19 –20 19 –20 19 2–20 19 –20 19 84– 19 4– 00 00 02 02 80 80 –2 –2 82 82 due to heat has been associated with socioeconomic 8 19 19 development and structural transformations, such as improvements in housing conditions and health-care Figure 5: Heat-attributable fraction of cardiovascular disease mortality in Spain by age and sex, 1980–2016 Shaded areas represent the 95% empirical CIs. systems (eg, improved treatment of heat-related morbidity),11,14 reduction in risk factors (eg, smoking, healthier diet),9 and planned adaptation policies led by view that the changes in vulnerability to heat among the governments and public health agencies.26,27 From a Spanish population are the result of an acclimatisation theoretical perspective, the increases in annual mean response to the warming. Nevertheless, the reduction in temperature and minimum mortality temperature have risk of mortality due to cold temperatures also highlights been similar in magnitude and thus would favour the the importance of socioeconomic development, improved www.thelancet.com/planetary-health Vol 3 July 2019 e303 Articles capita to €2182), social protection expenditure (from A By sex (all ages) B By sex (age ≥60 years) 0·20 Men €1845 per capita to €5746), and the number of doctors 0·18 Women (from 3930 per 1 million inhabitants to 4760 per 1 million Overall inhabitants) have all increased during the study period.28 0·16 Additionally, the proportion of house­holds with central Attributable fraction 0·14 heating increased from 25·83% in 1991 to 56·86% in 0·12 2011,29 and the proportion of households with air 0·10 conditioning increased from 4·16% in 199129 to 35·5% in 0·08 2008.30 0·06 Since the implementation of the National Plan for 0·04 Preventive Actions against the Health Effects of Excess Temperatures31 by the Spanish Ministry of Health in C By sex (age 60–74 years) D By sex (age 75–89 years) 2004, in response to a heatwave in the summer of 2003, a 0·20 rapid decrease in heat-related mortality was observed, 0·18 specifically between 1989–2003 and 1998–2012. However, 0·16 it is difficult to infer whether this accelerated decline in Attributable fraction 0·14 heat-related mortality is directly associated with the 0·12 introduction of the National Plan31 because this measure is only activated in the case of extreme heat events, 0·10 whereas most of the heat-attributable burden is caused 0·08 by moderate warm temperatures.12 0·06 In this study, the attributable fraction of deaths by sex 0·04 and age group differed for warm and cold temperatures. The attributable fraction due to warm temperatures E By sex (age ≥90 years) F By age group was higher for women than men across all the age 0·28 60–74 years groups, whereas the attributable fraction due to cold 75–89 years 0·24 ≥90 years temperatures was higher among men than women across all ages, with the exception of the 60–74 year age Attributable fraction 0·20 group. Moreover, heat-related mortality increased with 0·16 age for both sexes and the cold-related mortality increased with age for men and decreased for women. The findings 0·12 of previous studies with regard to differences in 0·08 temperature-related mortality between sex and age groups have been contradictory. Some studies have 0·04 shown that men are more susceptible to cold and heat G By age group (men) H By age group (women) than women, whereas other studies have reported the 0·28 opposite.32,33 This spatial heterogeneity might be a result of differences in socioeconomic, cultural, and health- 0·24 related factors. With respect to the effect modification by Attributable fraction 0·20 age, older people (aged ≥60 years) are considered the most at-risk population for heat,11,34,35 whereas differences 0·16 between age groups with regard to cold temperatures are 0·12 found to be less conclusive.32 For example, RRs for cold- related cardiovascular mortality were higher among 0·08 younger age groups (aged <65 years) than older age 0·04 groups (aged ≥65 years) in the USA,34 Japan,36 or South Korea.35 19 –94 19 –96 86 98 88 00 90 02 92 04 94 06 19 –20 8 98 10 20 012 20 –14 6 19 –94 19 –96 86 98 19 –20 0 90 02 9 04 94 06 8 98 10 20 012 20 –14 6 88 0 96 0 96 0 –1 –1 19 –20 19 –20 19 –20 19 –20 19 –20 19 –20 19 –20 19 –20 19 2–20 19 –20 19 84– 19 84– 00 00 02 02 80 80 –2 –2 82 82 The underlying physiological mechanisms by which 19 19 warm and cold temperatures trigger cardiovascular Figure 6: Cold-attributable fraction of cardiovascular disease mortality in Spain by age and sex, 1980–2016 mortality are not well understood, but seem to be largely Shaded areas represent the 95% empirical CIs. mediated by a thermoregulatory pathway. Several studies37 have identified age-related changes in thermo­regulation health care, increased life expectancy and quality, and in response to heat and cold stress. During heat exposure, planned interventions.13,15 In Spain, the gross domestic older individuals (aged ≥60 years) generally respond with product (from €8789 per capita in 1991 to €22 813 in attenuated sweat gland outputs, reduced blood flow to the 2009), the life expectancy at birth (from 77·08 years to skin, smaller increases in cardiac outputs, and less 81·58 years), health-care expenditure (from €605 per redistribution of blood flow from the splanchnic and e304 www.thelancet.com/planetary-health Vol 3 July 2019 Articles renal circulations compared with younger individuals. Blue-Action], 730004 [project PUCS], and 737480 These responses seem to be compatible with the increased [Marie Sklodowska-Curie fellowship ACCLIM]). We acknowledge the E-OBS dataset from the EU-FP6 project ENSEMBLES and the data risk of cardiovascular mortality with age for both men providers from the European Climate Assessment and Dataset project. and women. Further­ more, during exposure to cold References temperatures, older individuals generally respond with a 1 Intergovernmental Panel on Climate Change. Climate Change reduced peripheral vasoconstriction (implies increased 2013. The Physical Science Basis. Working group I contribution to heat loss) and decreased metabolic heat production the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. http://www.climatechange2013.org/images/ compared with younger individuals, although the ability report/WG1AR5_ALL_FINAL.pdf (accessed Jan 31, 2018). to maintain body temperature does not differ with age 2 Watts N, Amann M, Ayeb-Karlsson S, et al. The Lancet Countdown among women.38 These age-specific differences in on health and climate change: from 25 years of inaction to a global transformation for public health. Lancet 2018; 391: 581–630. thermo­ regulatory responses to cold exposure might 3 The Intergrovernmental Panel on Climate Change. Special Report. explain some of the diverging patterns in the effects of Global warming of 1·5°C. https://www.ipcc.ch/sr15/ (accessed cold temperatures observed between men and women in June 11, 2019). our study. 4 Huang C, Barnett AG, Wang X, Vaneckova P, Fitzgerald G, Tong S. Projecting future heat-related mortality under climate change Sex has also been linked to physiological differences scenarios: a systematic review. Environ Health Perspect 2011; in thermoregulation, which might explain some of 119: 1681–90. our results. Women have been reported to have a 5 Vardoulakis S, Dear K, Hajat S, Heaviside C, Eggen B, McMichael AJ. Comparative assessment of the effects of climate higher temperature threshold above which sweating change on heat- and cold-related mortality in the United Kingdom mechanisms are activated, and a lower sweat output than and Australia. Environ Health Perspect 2015; 122: 1285–92. men, which results in less evaporative heat loss and 6 Gasparrini A, Guo Y, Sera F, et al. Projections of temperature-related excess mortality under climate change therefore a larger susceptibility to the effects of heat.39,40 scenarios. Lancet Planet Health 2017; 1: e360–67. Conversely, men have larger decreases in core body 7 Intergovernmental Panel on Climate Change. Climate Change 2014. temperature when exposed to cold,41,42 which might Impacts, adaptation, and vulnerability. Part A: global and sectoral aspects. 2014 https://www.ipcc.ch/site/assets/uploads/2018/02/ explain the higher risk of cardiovascular mortality WGIIAR5-PartA_FINAL.pdf (accessed Jan 31, 2018). observed in response to cold temperatures. 8 McMichael AJ, Woodruff RE, Hales S. Climate change and human Our study had two main limitations. We could not health: present and future risks. Lancet 2006; 367: 859–69. control for ambient air pollution or relative humidity in 9 Bobb JF, Peng RD, Bell ML, Dominici F. Heat-related mortality and adaptation to heat in the United States. Environ Health Perspect 2014; the models due to paucity of data, and therefore, we do 122: 811–16. not know the extent to which values, trends, and 10 Gasparrini A, Guo Y, Hashizume M, et al. Temporal variation in subgroup differences are affected by temporal changes in heat-mortality associations: a multicountry study. Environ Health Perspect 2015; 123: 1200–07. these factors. Although the available literature on the 11 Ng CFS, Boeckmann M, Ueda K, et al. Heat-related mortality: confounding effect of air pollution shows modest34,43 or effect modification and adaptation in Japan from 1972 to 2010. no modifying effect,44 the effects of warm and cold Glob Environ Chang 2016; 39: 234–43. 12 Achebak H, Devolder D, Ballester J. Heat-related mortality trends temperatures on mortality remain unchanged when under recent climate warming in Spain: a 36-year observational relative humidity is accounted for.45 The use of combined study. PLoS Med 2018; 15: 1–17. indices of temperature and humidity, such as apparent 13 Vicedo-Cabrera AM, Sera F, Guo Y, et al. A multi-country analysis on potential adaptive mechanisms to cold and heat in a changing temperature, did not predict mortality more accurately climate. Environ Int 2018; 111: 239–46. than the single measure of temperature,46 and the 14 Chung Y, Ang D, Gasparrini A, et al. Changing susceptibility to assessment of the effect of temperature and humidity non-optimum temperatures in Japan, 1972–2012: the role of climate, separately showed that humidity does not affect demographic, and socioeconomic factors. Environ Health Perspect 2018; 126: 57002–08. mortality.47 In our study we did not describe the drivers of 15 Arbuthnott K, Hajat S, Heaviside C, Vardoulakis S. Changes in the observed reduction in heat-related and cold-related population susceptibility to heat and cold over time: assessing cardiovascular mortality in Spain, since this will be adaptation to climate change. Environ Heal 2016; 15 (suppl 1): 33. 16 Giorgi F. Climate change hot-spots. Geophys Res Lett 2006; 33: 1–4. addressed in a future study. 17 Haylock MR, Hofstra N, Klein Tank AMG, Klok EJ, Jones PD, Contributors New M. A European daily high-resolution gridded data set of HA designed the study, did the statistical analysis, interpreted the surface temperature and precipitation for 1950–2006. results, and wrote the original draft. JB provided the data and J Geophys Res Atmos 2008; 113: 1–12. contributed to the statistical analysis, the interpretation of the results, 18 Gasparrini A, Armstrong B, Kenward MG. Distributed lag and drafting of the manuscript. DD contributed to the interpretation of non-linear models. Stat Med 2010; 29: 2224–34. the results and the drafting of the manuscript. All authors contributed to 19 Gasparrini A, Guo Y, Hashizume M, et al. Mortality risk attributable the submitted version of the manuscript and approved the final version. to high and low ambient temperature: a multicountry observational study. Lancet 2015; 386: 369–75. Declaration of interests 20 Gasparrini A, Armstrong B, Kenward MG. Multivariate We declare no competing interests. meta-analysis for non-linear and other multi-parameter Acknowledgments associations. Stat Med 2012; 31: 3821–39. HA was supported by the Secretariat for Universities and Research of 21 Gasparrini A, Leone M. Attributable risk from distributed lag the Ministry of Business and Knowledge of the Government of Catalonia models. BMC Med Res Methodol 2014; 14: 1–8. (grant numbers B00391 [FI-2018] and 100180 [FI-2019]). JB gratefully 22 Todd N, Valleron A. Space–time covariation of mortality with acknowledges funding from the European Union’s Horizon 2020 temperature: a systematic study of deaths in France, 1968–2009. Environ Health Perspect 2015; 123: 659–64. research and innovation programme (grant numbers 727852 [project www.thelancet.com/planetary-health Vol 3 July 2019 e305 Articles 23 Åström DO, Tornevi A, Ebi KL, Rocklöv J, Forsberg B. Evolution of 35 Son JY, Lee JT, Anderson GB, Bell ML. Vulnerability to minimum mortality temperature in Stockholm, Sweden. temperature-related mortality in Seoul, Korea. Environ Res Lett 2011; Environ Health Perspect 2016; 124: 740–44. 6: 034027. 24 Ballester J, Rodó X, Robine J-M, Herrmann FR. European seasonal 36 Atsumi A, Ueda K, Irie F, et al. Relationship between cold mortality and influenza incidence due to winter temperature temperature and cardiovascular mortality, with assessment of effect variability. Nat Clim Chang 2016; 6: 927–30. modification by individual characteristics. Circ J 2013; 77: 1854–61. 25 Barnett AG. Temperature and cardiovascular deaths in the US 37 Kenney WL, Munce TA. Aging and human temperature regulation. elderly: changes over time. Epidemiology 2007; 18: 369–72. J Appl Physiol 2003; 95: 2598–603. 26 Fouillet A, Rey G, Wagner V, et al. Has the impact of heat waves on 38 Wagner JA, Horvath SM. Influences of age and gender on human mortality changed in France since the European heat wave of thermoregulatory responses to cold exposures. J Appl Physiol 1985; summer 2003? A study of the 2006 heat wave. Int J Epidemiol 2008; 58: 180–86. 37: 309–17. 39 Bittel J, Henane R. Comparison of thermal exchanges in men and 27 Schifano P, Leone M, De Sario M, et al. Changes in the effects of women under neutral and hot conditions. J Physiol 1975; heat on mortality among the elderly from 1998–2010: results from 250: 475–89. a multicenter time series study in Italy. Environ Health 2012; 11: 58. 40 Andérson GS, Ward R, Mekjavić IB. Gender differences in 28 Spanish National Institute of Statistics. Social indicators, 2011 physiological reactions to thermal stress. edition. 2012. http://www.ine.es/dyngs/INEbase/en/operacion.htm Eur J Appl Physiol Occup Physiol 1995; 71: 95–101. ?c=Estadistica_C&cid=1254736176825&menu=resultados&i 41 Wagner JA, Horvath SM. Cardiovascular reactions to cold exposures dp=1254735576508 (accessed Jan 15, 2018). differ with age and gender. J Appl Physiol 1985; 58: 187–92. 29 Spanish National Institute of Statistics. Population and Housing 42 Graham TE. Thermal, metabolic, and cardiovascular changes in Census results. 2012. http://www.ine.es/en/censos2011_datos/ men and women during cold stress. Med Sci Sport Exerc 1988; cen11_datos_resultados_en.htm (accessed Jan 15, 2018). 20: 185–92. 30 Spanish National Institute of Statistics. 43 Brooke AG, Bell ML. Weather-related mortality: how heat, cold, Survey on households and environment 2008. 2010. and heat waves affect mortality in the United States. Epidemiology http://www.ine.es/dyngs/INEbase/en/operacion.htm?c=Estadistica 2009; 20: 205–13. _C&cid=1254736176950&menu=resultados&idp=1254735576508 44 Zanobetti A, Schwartz J. Temperature and mortality in (accessed Jan 15, 2018). nine US cities. Epidemiology 2008; 19: 563–70. 31 Ministerio de Sanidad, Consumo y Bienestar Social. Plan Nacional 45 Hajat S, Kovats RS, Lachowycz K. Heat-related and cold-related de Actuaciones Preventivas por Altas Temperaturas. https://www. deaths in England and Wales: who is at risk? Occup Environ Med mscbs.gob.es/ciudadanos/saludAmbLaboral/planAltasTemp/2018/ 2007; 64: 93–100. Plan_nacional_actuaciones_preventivas.htm (accessed 46 Hajat S, Armstrong B, Baccini M, et al. Impact of high June 13, 2018). temperatures on mortality: is there an added “heat wave” effect? 32 Liu C, Yavar Z, Sun Q. Cardiovascular response to thermoregulatory Epidemiology 2006; 17: 632–38. challenges. Am J Physiol Heart Circ Physiol 2015; 309: H1793–812. 47 Braga A, Zanobetti A, Schwartz J. The effect of weather on 33 Moghadamnia MT, Ardalan A, Mesdaghinia A, Keshtkar A, respiratory and cardiovascular deaths in 12 US cities. Naddafi K, Yekaninejad MS. Ambient temperature and Environ Health Perspect 2002; 110: 859–63. cardiovascular mortality: a systematic review and meta-analysis. PeerJ 2017; 5: e3574. 34 O’Neill MS, Zanobetti A, Schwartz J. Modifiers of the temperature and mortality association in seven US cities. Am J Epidemiol 2003; 157: 1074–82. e306 www.thelancet.com/planetary-health Vol 3 July 2019