The summer of 2022 was the hottest on record in Europe, characterized by intense series of heatwaves, and +1.4 °C warmer than the reference period 1991–20201. With temperatures on the European continent increasing at least twice as fast as the global average, and most recent years breaking one temperature record after another, Europe is already experiencing the adverse impacts of anthropogenic warming, especially an unprecedented increase in the frequency and intensity of heatwaves2. One of the significant impacts of human-induced climate change on human health is heat-related morbidity and mortality3. Up-to-date, heatwaves account for the majority of weather-related fatalities in Europe4. Heightened risks of death from cardiovascular and respiratory diseases related to heat exposure have been identified among people with pre-existing conditions, particularly in vulnerable populations, including the elderly and those in disadvantaged socio-economic conditions3.

Despite the growing awareness of those risks during recent decades, and the implementation of public health strategies to cope with the adverse effects of heat, European societies are still largely vulnerable to high summer temperatures. Ballester et al.5 used epidemiological models to estimate that the record-breaking temperatures observed in Europe during the 2022 summer were associated with 61,672 [95% CI 37,643–86,807] heat-related deaths. Moreover, they found heat-related mortality rates greater than 200 summer deaths per million inhabitants in several Southern European countries, i.e. Italy, Greece, Spain, and Portugal, with decreasing impacts towards higher latitudes. In a recent analysis of the 2023 summer, Gallo et al.6, similarly, found a notably high heat-related mortality burden, i.e. 47,690 deaths [95% CI = 28,853–66,525].

In this study, we aim to quantify the contribution of human-induced climate change to the heat-related mortality burden of the 2022 summer in Europe and analyze it within the context of other less extreme recent years (2015–2021). Towards this aim, we compare observed temperatures (Fig. 1a) and the associated heat-related mortality (Fig. 1b), here referred to as factual temperatures and mortalities, with a counterfactual scenario (Fig. 1c, d). We obtain the counterfactual temperatures by removing the anthropogenic warming signal from the factual temperatures (see the “Methods” section). The difference between the two scenarios, i.e. factual minus counterfactual, provides an estimate of the anthropogenic warming itself (Fig. 1e), and its contribution to the heat-related mortality during the summer of 2022 (Fig. 1f) and preceding years (Table 1).

Fig. 1: Regional temperature and heat-related mortality rate during the summer of 2022.
figure 1

Summer mean temperature (in °C; a, c) and heat-related mortality rate (summer deaths per million; b, d) during the summer of 2022 with observed temperatures (factual temperature and mortality; a, b) and in the absence of anthropogenic warming (counterfactual temperature and mortality; c, d). The resulting differences (factual minus counterfactual) are the anthropogenic warming (°C, e) and the mortality attributed to the anthropogenic warming (summer deaths per million, f).

Full size image
Table 1 Contribution of anthropogenic warming to heat-related mortality during recent summers (2015–2022)
Full size table

The factual temperatures are obtained from the ERA5-Land reanalysis7 for the period from 1950 to 2022, to which we apply well-established techniques within the framework of detection and attribution studies to calculate the counterfactual temperatures8. Specifically, for each region, we linearly regress the annual time-series of regional summer mean temperature onto the annual time–series of global mean surface temperature (GMST) change from HadCRUT59. The slope for each region is then multiplied by the level of global mean surface warming in 2022 relative to the pre-industrial era (i.e. +1.15 °C)9, and the resulting regional summer anthropogenic warming level is then subtracted from the weekly regional factual temperatures to obtain the regional counterfactual temperatures (Fig. 1e).

The regional anthropogenic warming level ranges between 0.63 and 3.90 °C for the summer of 2022 across Europe (Fig. 1e), with lower bounds (5th percentile) between 0.04 and 3.01 °C and upper bounds (95th percentile) between 1.03 and 4.70 °C. The population-weighted European increase in 2022 is at 2.10 °C [95% CI 1.57–2.55 °C]9, twice the global average.

In the next step, we use the same mortality database and epidemiological models as in Ballester et al.5 (see the “Methods” section) to estimate the factual and counterfactual heat-related mortality, and then calculate their difference to isolate the effects of human-induced climate change. Because Ballester et al.10 found that using data at weekly resolution leads to an underestimation of the summer heat-related mortality, which decreases monotonically and nonlinearly with higher temperatures, the underestimation for the counterfactual is likely larger than for the factual. Therefore, we apply the methodology described in Ballester et al.10 to bias–correct the estimates from weekly data models to infer the estimate of summer heat-related mortality that would have been obtained from daily data models. As a result, we estimate that 38,154 [95% CI 21,722–58,571] of the 68,593 [95% CI 42,229–95,891] heat-related deaths in the summer of 2022 would not have occurred without anthropogenic warming, i.e. a percentage of 56% [95% CI 39–77%]. In absolute numbers, most—70% or 26,813—of those deaths occurred in Southern Europe (Table S2). Moreover, the number of heat-related deaths per inhabitant attributed to anthropogenic warming is twice as high in the Southern regions compared to the rest of Europe (Fig. 1f).

We find a percentage of heat-related deaths attributable to anthropogenic warming higher than 50% [30%] at the 0.9 significance level in 16 [30] of the 35 European countries included in this analysis. Hence, anthropogenic warming has considerably exacerbated heat-related mortality across almost the entire continent (Table S2). However, the percentage varies across countries without a significant North–South gradient, e.g., high values can be found in Sweden (77%) and Norway (71%), but a small percentage (26%) is found in neighboring Finland (Table S2). These findings highlight that temperature exposure does not solely determine the health burden imposed by heat. Socio-demographic characteristics and resilience to heat differ between and within countries, and play an important role in the vulnerability to the impacts of human-induced climate change.

In line with former studies5,8,11, we find a higher number of heat-related deaths attributed to climate change among women (22,501 deaths) and people aged 80 years or more (23,881 deaths), as opposed to men (14,026 deaths) and people aged 64 years or less (2702 deaths) (Table 1). Both the absolute number and the percentage of heat-related deaths attributed to climate change are higher in these cases, indicating that the generally higher vulnerability of these groups is exacerbated by anthropogenic warming to a larger extent than in the general population.

Human-induced climate change has not only exacerbated the heat-related mortality in exceptionally hot summers such as in 2022. Among the years of the period 2015–2021, between 44% and 54% of summer heat-related mortality can be attributed to anthropogenic warming (Table 1). In absolute numbers, this corresponds to an annual burden attributed to anthropogenic warming ranging between 19,000 and 28,000 heat-related deaths during 2015–2021. Hence, the heat-related mortality of the record-breaking summer of 2022 attributed to anthropogenic warming (i.e. 38,154 deaths) represents an increase of around two–thirds with respect to the average mortality burden attributed to human-induced climate change (i.e. 23,450 deaths).

Compared with previous studies, we note that the estimated 65% of climate change-attributable heat-related mortality for Switzerland (Table S2) is consistent with the 60% estimate obtained in recent work by Vicedo-Cabrera et al.8. Furthermore, Mitchell et al.12 found a percentage of 20% for London and 70% for Paris for the summer of 2003, which is reasonably close to our estimate for the summer of 2022, i.e. a percentage of 36% for the United Kingdom and 54% for France (Table S2). In a global multi-country, multi-city study, Vicedo-Cabrera et al.13 found a lower percentage (37%) for the 1991–2018 period. Considering the differences in methodology and study area in the previous work, we cannot directly compare the results of both studies. However, our findings, i.e. a higher percentage of heat-related deaths attributable to human-induced climate change during the summer of 2022 than in the previous period, are consistent with the notion of intensifying human-induced climate change and the resulting impacts.

For this study, we use a simple approach to derive counterfactual climates because, in our view and considering recent research on European warming and its drivers, the potential benefits of more complex detrending methods cannot be validated (see the “Methods” section).

Our study is based on the assumption that temperatures shift in response to changes in GMST, as demonstrated by large ensembles of model simulations14,15. It has been shown that GMST also causes changes in temperature variability, although weaker than mean changes (or trends), and the underlying mechanisms remain debated16,17. Using climate model simulations and other more sophisticated methods could potentially increase the accuracy of the results but would introduce new caveats.

Furthermore, we acknowledge using the same epidemiological association for the factual and counterfactual scenarios. Therefore, the contribution of climate change to the heat-related mortality burden must here be interpreted as the epidemiological transformation of two different climate regimes, excluding the potential role of adaptation13. Gallo et al.6 estimated the temperatures observed in 2023 would have caused over 80% more heat-related deaths between 2000 and 2004. Their study shows that adaptation has drastically reduced the risk of heat-related mortality, especially for the elderly population. In another study, Stuart-Smith et al.18 show that 40% of the climate change-attributable deaths between 1969 and 2000 in the canton of Zurich were avoided because of adaptation. Gaining new insights into the effectiveness of adaptation to heat will allow us to better predict the impacts of human-induced climate change.

Summarizing the findings of this study, we attribute—across all sex or age groups—around half of heat-related deaths in Europe during recent summers to human-induced climate change. Our findings highlight that human-induced climate change poses a risk beyond vulnerable populations, extreme temperatures, heatwaves, or Southern regions characterized by high summer temperatures. However, we also find that population groups more susceptible to heat, i.e. women and the elderly, are more adversely affected by anthropogenic warming than the general population.

Fueled by human-induced climate change, increases in temperature and heat-related mortality in Europe have sped up in recent years and will most likely continue do so in the near future unless strong adaptation and mitigation actions are put in place. Compared to 2015–2021, we show that the record-breaking temperatures of the 2022 summer have caused an increase in heat-related mortality by more than 40% and an increase in human-induced climate change-attributable mortality by more than two-thirds.

Hence, our study urgently calls for national governments and agencies in Europe to increase the ambition and effectiveness of heat surveillance and prevention measures, new adaptation strategies, and global mitigation efforts.

Methods

Data sources

We aggregate the hourly gridded 2-m temperature data from the high–resolution ERA5-Land reanalysis7 into weekly regional averages of daily mean 2-m temperature for the period of 1950–2022. We obtain global mean surface temperature (GMST) changes from HadCRUT59. We use the same mortality data as in ref. 5: We obtained weekly counts of all-cause mortality by sex and age groups from Eurostat19, and missing data was complemented by contacting the corresponding national agencies for statistics. The final dataset includes 45,184,044 counts of death (22,000,519 for women and 21,913,050 for men) between January 2015 and November 2022 from 823 contiguous regions inhabited by over 543 million Europeans in 35 countries. All–age data by sex was not available in the United Kingdom, and only at the country level in Germany. Data by sex and age groups were not available in the United Kingdom, Ireland, and Germany.

Counterfactual temperatures

For each region, we linearly regress the annual time series of regional summer mean temperature onto the annual time series of GMST for the 4-year running mean from HadCRUT5.

The slope parameter for each region is then multiplied by the value of GMST in 2022 relative to the pre-industrial era (i.e. +1.15 °C9), and this regional summer warming is then subtracted from the regional factual temperatures to obtain the regional counterfactual temperatures. We also generate counterfactual scenarios for the upper and lower bound of the 95% confidence intervals of the regional anthropogenic warming.

It has recently been shown that in much of Western Europe, circulation changes have considerably exacerbated the largely thermodynamic warming that primarily occurs in response to anthropogenic greenhouse gas emissions20,21,22. As of now, it remains unclear if these circulation changes are entirely unforced, i.e. a manifestation of internal climate variability, or partly externally forced20.

Bearing all this in mind, we deem our simple approach to derive counterfactual climates reasonable because, in our view and considering recent research on European warming and its drivers, the potential benefits of more complex detrending methods cannot be validated.

Epidemiological analysis

We use the same two-step epidemiological model as in ref. 5. In the first stage, we used quasi-Poisson regression models, which allow for overdispersed counts of deaths, to calculate the location-specific temperature–lag–mortality relation in each European region11,23. The models include (i) an intercept, (ii) a natural cubic spline of time with 8 degrees of freedom per year to control for the seasonal and long-term trends, and (iii) a cross-basis function to estimate the exposure–lag–response association between weekly temperatures (temp) and mortality counts (mort):

$$\begin{array}{ll}{log}(E({{{mort}}}))={{{intercept}}}+{ns}({{{time}}},8{{{df}}\; {{per}}\; {{year}}})\\\qquad\qquad\qquad\quad+\,{{{cross-basis}}}({{{temp}}}{{;}}\,0,1,2,3\,{{{weeks}}})\end{array}$$

The lag–response function of the cross-basis is modelled with integer lag values of 0, 1, 2, and 3 weeks, and the exposure–response function with a natural cubic spline with three internal knots at the 10th, 50th, and 90th percentiles of the location-specific weekly temperature distribution.

In the second stage, we use a multivariate multilevel meta-regression analysis24 to pool the location-specific coefficients obtained in the first step. The meta-regression includes (i) country random effects and the location-specific (ii) temperature average, (iii) temperature interquartile range, and (iv) percentage of people aged 80+ years as meta–predictors13. We derived the best linear unbiased predictions of the temperature–mortality relationship in each region from the meta-regression25 to obtain the location-specific minimum mortality temperature and to transform the regional temperature and mortality time-series from January 2015 to November 2022 into the weekly and summer heat-related mortality numbers over the years 2015–202226. Heat-related mortality is calculated for the weeks with average temperatures above the location-specific minimum mortality temperature27. Regional heat-related mortality numbers are aggregated to obtain the national and European burdens28,29. Similarly, we computed 1000 Monte Carlo simulations of the regional heat-related mortality numbers and separately aggregated the numbers in each simulation to calculate the 95% confidence intervals at the national and continental levels11,13. We then follow the methodology proposed by Ballester et al.10 to bias-correct the weekly estimates by using a dataset of daily mortality data spanning over 16 European countries.