According to the European Environment Agency, Europe is the fastest-warming continent in the world, and heat stress is one of the most severe climate risks.¹ In 2023, Europe recorded one of its warmest summers in history. Globally, yearly average temperatures continue to rise as well. Many countries are experiencing an increase in their average temperature and temperature volatility, leading to extreme heat or cold. Thus, people are being exposed to more heatwaves, which have a significant impact on human health, as demonstrated by the high number of excess deaths during the European heatwaves in 2003, 2018 and 2022.^2,3,4

To analyze the mortality impact of heatwaves, one should first define what a heatwave is – but due to regional variations in temperature, there is no standardized definition. Typically, heatwaves are defined by the number of days (duration) with the maximum temperature exceeding a certain threshold. The threshold can be either a fixed value or a percentile of the local temperature distribution.⁵ It is also important to consider the possibility of delayed effects of extreme heat on mortality.⁶ For example, heat stress or dehydration may worsen chronic conditions and thereby lead to higher mortality in the weeks following a heatwave.

In this article, we analyze the impact of heatwaves on mortality in selected European countries from 2000–2019 and in South Korea from 2010–2019. We estimate quasi-Poisson regression models⁷ with actual over expected (A/E) deaths as the target variable and include the number of days in a week with maximum temperatures at or above a fixed threshold to capture the effects of extreme heat duration. We also include one- and two-week lags of extreme heat duration to explore its delayed effects on the mean A/E.

Data Sources

We considered the following groups of countries:

• The DACH countries (Germany, Austria and Switzerland)
• The Benelux countries (Belgium, Netherlands and Luxembourg)
• The Baltic countries (Estonia, Latvia and Lithuania)
• South Korea

Since the land area of Germany is larger than the areas of the other selected countries, the analysis on a national level for Germany would have been less informative and was conducted at the state level instead.

We took the actual weekly deaths from the Short-term Mortality Fluctuations (STMF) data series of the Human Mortality Database (HMD) for all countries except Germany,⁸ for which we used weekly state-level deaths from the German Federal Statistical Office.⁹ Thereby, we obtained more data points for Germany and accounted for state variations in temperature. For the European countries, we included weekly mortality data over the period 2000–2019, preceding the COVID‑19 pandemic. For South Korea, the STMF weekly mortality data was available only for 2010–2019.

For all countries except Germany, we used the HMD population data at the beginning of a year.¹⁰ For Germany, population size at the end of a year for each state was taken from the German Federal Statistical Office,¹¹ and we assumed that population size at the end of a year was equal to population size at the beginning of the next year. Further, we used life tables from the HMD¹² for all countries, including Germany, to compute expected deaths for each country or state in each year with the population size at the beginning of the year and life table of the previous year, which was the newest life table at that time. We divided the expected yearly deaths by 52, removing week 53 where applicable for consistency purposes. Weeks in these data sources are defined according to the ISO 8601 date and time standard.

We used CPC Global Unified Temperature data,¹³ WorldPop population counts¹⁴ and country¹⁵ (or German state¹⁶) polygons to compute the weighted mean of maximum temperature for each country (or German state) for each day over the period 2000–2019.¹⁷ Population counts were used as weights, since high temperatures are expected to have a stronger effect on mortality in densely populated areas, as more people are at risk there.

Modeling Approach

We estimated quasi-Poisson regression models to investigate the impact of extreme heat duration on the mean value of actual deaths while accounting for expected deaths in different countries or groups of countries. To this end, the A/E ratio was taken as the target variable and expected deaths as the weight variable. The number of days with maximum temperatures at or above a certain threshold in a given week, one week prior and two weeks prior were included as categorical predictor variables alongside country (for Europe), year, age group and gender. We computed the 95% confidence intervals for parameter estimates using the delta method.^18,19

Since extreme heat events typically occur during the summer months, we restricted the data we used for modeling to weeks 18–40. Higher levels of the extreme heat duration were combined into one category (3+, 4+ or 5+) due to a low occurrence of those levels, and we chose the level of 0 days as the base level in all models.

Extreme heat in Europe

Figure 1 presents modeling results for the DACH, Benelux and Baltic countries. The maximum temperature threshold used to define extreme heat was set to 30°C, similar to studies analyzing heat-related mortality in Germany and Austria.^20,21 The results in the rest of the article are based on the state-level data for Germany and country-level data for the other countries.

Figure 1 – Effects of Extreme Heat Duration in the DACH, Benelux and Baltic Countries

Source: Author visualization of modeling results based on various data sources²²

Figure 1 shows the estimates of the percentage change in the mean A/E for different levels of extreme heat duration in a given week relative to the base level of 0 days alongside 95% confidence intervals. The effects of extreme heat duration in the three groups of countries show a similar pattern: increasing the duration of extreme heat increases the mean A/E relative to the no‑extreme-heat scenario in a given week, with the largest effect taking place when extreme heat lasts four or five or more days. Specifically, in the DACH and Benelux countries, the mean A/E is estimated to be 17% and 19% higher, respectively, when extreme heat lasts five or more days in a given week relative to the no‑extreme-heat scenario. For the Baltic countries, the estimated effect of extreme heat lasting five or more days in a given week is of similar magnitude but has high uncertainty associated with it, since there was only one week over the period 2000–2019 (week 31 in 2014 in Lithuania) when extreme heat lasted five or more days.

These results are based on an assumption that the effects of enduring extreme heat are the same regardless of which age group we consider. However, studies show that older or very young age, certain chronic conditions, pregnancy, and social deprivation are among the risk factors for heat-related adverse health outcomes.²³

As insurers, we are particularly interested in whether the effects of extreme heat vary by age group. To this end, we combined data for the DACH, Benelux and Baltic countries, as the effects of extreme heat duration appeared to be similar across these countries, and we included the interactions of extreme heat duration and its lags and age group into the model. Age groups were defined as follows:

• 0 to 64
• 65 to 74
• 75 to 84
• 85 and older

Figure 2 shows the estimates of percentage change in the mean A/E for different levels of extreme heat duration in a given week relative to the base level of 0 days for different age groups. For the age groups 0 to 64, 65 to 74, 75 to 84, and 85 and older, the mean A/E is estimated to be 11%, 12%, 18%, and 22% higher, respectively, when extreme heat lasts five or more days relative to the no‑extreme-heat scenario in a given week. Overall, the effects of extreme heat duration appear to be most prominent for the oldest age group and least prominent but still significant for the younger age groups of 0 to 64 and 65 to 74 years. These findings are consistent with other research exploring the association between temperature and mortality.²⁴

Figure 2 – Effects of Extreme Heat Duration in a Given Week in the DACH, Benelux and Baltic Countries (Combined) by Age Group

Source: Author visualization of modeling results based on various data sources²⁵

We explored the delayed effects of extreme heat by adding the interactions of age group and one- and two-week lags of the extreme heat duration predictor. Increases in the mean A/E in a given week were detected only for the extreme heat duration of five or more days one week prior (3% to 9% depending on the age group), whereas small decreases were detected for the extreme heat duration of one to five or more days two weeks prior (see Figure 3). An der Heiden et al. (2020) explained similar findings by the displacement of the time of death by two or three weeks.²⁶

Figure 3 – Effects of Extreme Heat Duration One and Two Weeks Prior in the DACH, Benelux and Baltic Countries (Combined) by Age Group

Source: Author visualization of modeling results based on various data sources²⁷

Extreme Heat in South Korea

Figure 4 presents modeling results for South Korea, covering the period from 2010–2019. We used the maximum temperature threshold of 35°C to define extreme heat in South Korea, since it is consistent with a threshold used by the Korea Meteorological Administration to issue a heatwave warning.²⁸ We could detect weaker or no effects of extreme heat duration when we used lower temperature thresholds.

Figure 4 – Effects of Extreme Heat Duration in South Korea, 2010–2019

Source: Author visualization of modeling results based on various data sources²⁹

Information in Figure 4 shows that the mean A/E is estimated to be 6% higher when extreme heat lasts three or more days relative to the no‑extreme-heat scenario in a given week in South Korea. Over the period 2010–2019, extreme heat duration of three or more days was observed on the country level for only two weeks (weeks 31 and 33 in 2018), leading to higher uncertainty associated with the percentage change estimate.

Needing to use different temperature thresholds to define extreme heat for Europe and South Korea can be explained by the violin-box plots in Figure 5. The plots visualize the daily maximum temperature distribution in weeks 18–40 over the period 2010–2019, showing that most of the maximum temperature values for South Korea are concentrated in a higher temperature range and the median maximum temperature is 7°C to 8°C degrees higher than in the selected European countries.

The estimated effects of extreme heat duration seem to be slightly weaker in South Korea than in Europe when 30°C and 35°C temperature thresholds are used to define extreme heat in Europe and South Korea, respectively. However, more data will be required to reduce the uncertainty associated with these estimates.

Figure 5 – Daily Maximum Temperature Distribution for Selected Countries, Weeks 18–40, 2010–2019

Source: Author visualization of the CPC Global Unified Temperature data³⁰

Summary and Outlook

The analysis of the effects of extreme heat duration in selected European countries and South Korea indicates heat that is extreme relative to a local temperature distribution and that lasts sufficiently long in a given week or one week prior could have adverse effects on mortality across all age groups, with the oldest age groups most affected. Additional deaths caused by extreme heat and heatwaves may result in a potential burden for life insurance companies. However, preventive strategies could mitigate the adverse heat effects.³¹ A natural extension of this analysis could be to also investigate the impact of extreme cold on mortality in the winter months, controlling for the flu effects and accounting for the possibility of longer lags between cold exposure and observed mortality.³²

This article was initially published by the Society of Actuaries in The Actuary Magazine, https://www.theactuarymagazine.org/the-effects-of-heatwaves/