During the summer of 2010, the territories of European Russia and neighbouring countries experienced record-breaking temperature anomalies. The area of the blocking high was established over the vast territory in July, and had dissipated by mid-August. However, the development of anticyclones was uneven, leading to temperature anomalies occurring at different times in various regions. Extreme heat was initially observed at the end of June at a latitude of Moscow, and spread to the Central, Central Black Earth and Volga-Vyatka districts in July. Peak temperatures were recorded in early August, and the heatwave abruptly subsided in the middle of that month. Therefore, to capture the entire progress of the heatwave and its impact on excess mortality, a long time frame of 27–33 weeks in 2010 (05.07–22.08) was chosen for the study.

The study covered the entire territory of European Russia, which spans almost four million square kilometres and consists of 55 regions (with the cities of Moscow and Saint Petersburg considered distinct regions) grouped into eight economic districts (Figure 1). Due to the substantial heterogeneity of climatic conditions in European Russia, several of these territories, such as the Arctic North and the North Caucasus, did not experience extreme heat during the 2010 heatwave.

Populated places in Russia were categorized into rural settlements and urban-type settlements (posyolok gorodskogo tipa) and cities. In the past, the threshold populations for inclusion in the latter categories were 3,000 and 12,000, respectively, and these thresholds were accompanied by a requirement that the settlement have a non-agricultural occupational structure. However, as of 2010, no strict criteria were applied, and the status of the settlement was generally determined by its role in the hierarchy of municipal subdivisions. All the computations were undertaken in an aggregate form for different groups of settlements divided by their size and type. This approach not only allowed us to account for settlement patterns, it also helped us to mitigate the limitations caused by the small population sizes.

The following classification of settlements was utilized: rural settlements; urban settlements (both city and urban-type settlements) with fewer than 100,000 inhabitants; and cities with 100,000–249,999, 250,000–499,999, 500,000–999,999 or 1,000,000+ inhabitants.

Mortality was assessed in both absolute terms (total number of deaths) and relative terms (death rate standardized by the 2013 European population standard) during the 27–33 ISO weeks of 2010. Excess mortality was estimated based on the differences between the observed values and the expected values calculated from the model using the 2005–2009 reference period. This process was performed for each aggregated group of settlements at the regional level.

Interrupted time-series analysis is considered the gold standard in environmental epidemiology, but the applicability of many convenient approaches is limited due to the aggregation of data at the weekly level. For example, the inclusion of the day of the week and other independent variables, as well as the use of distributed lags, can become impractical. Moreover, regression models incorporating only time variables are applicable when weekly data are used.

The baseline mortality level during the heatwave was estimated using a regression model incorporating both long-term changes in mortality and seasonality. Linear and quadratic yearly trends represent long-term changes in mortality (Comas-García and Erdely, 2023), while two pairs of sine and cosine functions with periods of 52 and 26 weeks are employed for seasonal patterns (Nepomuceno et al., 2022). Weekly baseline mortality estimates for each week of 2010 were retrospectively projected using data from the reference period. The resulting formula was stated as follows: log(E(dij))=β0+β1i+β2i2+β3sin(2π52j)+β4cos(2π52j)+β5sin(2π26j)+β6cos(2π26j), where E(dij) is the expected mortality during the jth week of year i. 95% confidence intervals were also calculated for all the estimates using bootstrapping. All the data extraction and processing, calculations and analyses were performed in R 3.6.2.

We employed regression analysis to quantitatively evaluate the impact of settlement type and size on heat-related mortality. Two distinct model types were considered: the first included the category of settlement size as a categorical variable, while the second incorporated interactions between settlement characteristics and thermal stress indicators. The first approach allowed us to investigate the differences in excess mortality between cities of varying sizes and rural areas during heatwaves while controlling for external heatwave effects. The latter approach provided an assessment of the “heat slope”, revealing the varying risk-exposure relationships based on the size and the type of settlement.

The first approach can be expressed as: log(Yijk)=β0+β1k+β2X+β3Z where Yijk represents the relative excess of the observed week-specific age-standardized death rate over the expected values in region i and category of settlements k during the jth week, X is the measurement of thermal stress and Z denotes the vector of air pollution variables. Since k is the factor variable, β1 is a set of coefficients, each representing the change in the age-standardized death rate during the heatwave associated with a specific type of settlement. Thermal stress and air pollution variables from the available parameters described above were analysed, and the most effective models were selected based on information criteria (AIC and BIC) and their performance.

To demonstrate the consistency of the estimates of the coefficients β1 indicating the association between excess mortality and the type of settlement, we also assessed models that included the observed Xj in conjunction with a one-week lag Xj−1. This decision was motivated by the persistent lag effects observed in multiple studies.

The second model can be defined as: log(Yijk)=β0+β1k+β2k*X+β3Z

Here, the settlement category k interacts with the heat stress X, providing the “heat slope” steepness, which denotes the varying associations between heat stress and excess mortality based on the type and the size of the settlement. All the descriptive statistics of the variables used in the regression analysis are presented in Table 1.

During the 27–33 weeks of 2010, the total number of excess deaths, compared to the baseline calculated from the 2005–2009 reference period, was estimated at 56.0 (95% confidence interval (CI): 48.9–62.9) thousand individuals. The distribution of casualties was not uniform, with 10.7 (CI: 10.3–11.0) thousand excess deaths occurring in Moscow. The excess mortality in the Central economic district, excluding Moscow, amounted to 11.3 (CI: 9.8–12.7) thousand cases. The Central Black Earth and Volga districts, which were also among the most severely affected by the heatwave experienced during this period, had 5.7 (CI: 5.1–6.3) and 12.0 (CI: 10.7–13.2) thousand excess deaths, respectively.

The impact of the heatwave was uneven across different types and sizes of settlements (Table 2). Rural areas, which accounted for 26.8% of the total population in European Russia, experienced an estimated 10.0 (CI: 7.8–12.2) thousand excess deaths. Simultaneously, in cities with populations of more than a million (excluding Moscow and Saint Petersburg) the number of excess deaths was 5.5 (CI: 5.1–5.9) thousand, corresponding to 6.4% of the population as of the 2010 census.

Throughout the study period, excess mortality was observed within all weeks, with the impact starting earlier and being relatively modest in Saint Petersburg (Figure 2). In most areas, the peak was observed at 31–32 weeks (02.08–15.08). However, a compensatory decline in excess mortality (also known as a “harvesting effect”) after the end of the heatwave was not identified.

The observed age-standardized death rate during the 2010 heatwave in European Russia was estimated to be 13.9 per 1,000, exceeding the baseline by 3.1 (CI: 2.6–3.5). Interestingly, the initial variations in death rates led to decreased differences in cities of different sizes during the heatwave. The baseline age-standardized death rates ranged between 10.4 and 12.0 per 1,000 in the largest cities (over 1,000,000) and the smallest urban areas (fewer than 100,000), respectively, while the observed death rates were 15.2 and 14.5 per 1,000, respectively.

The highest standardized death rate, 16.2 per 1,000, was observed in the Volga-Vyatka economic district, which was partially attributable to this district having one of the highest baseline levels of mortality (12.0 per 1,000 during the corresponding period). Furthermore, the most substantial increases in age-standardized death rates, exceeding 4.0 per 1,000 people, occurred in the Volga, Volga-Vyatka and Central Black Earth districts (Figure 3).

The relative increase in terms of age-standardized death rates was 28.4% (CI: 23.2%–33.8%) for the entire population of European Russia (Figure 4). The increased risk of death during the heatwave was 16.8% (11.9%–22.0%) in rural areas, while it was estimated at 46.6% (CI: 41.5%–51.8%) in cities with a population exceeding one million (excluding capital cities). In Moscow and Saint Petersburg, age-standardized death rates increased by 67.1% (CI: 63.4%–70.7%) and 22.5% (CI: 20.0%–25.3%), respectively. Overall, cities with more than 500,000 inhabitants experienced significantly greater excess mortality during the heatwave, as these urban areas had the largest increases in death rates compared to the total population (as presented in Table S.1 in the Supplementary material, available online at https://doi.org/10.1553/p-42de-zc5p).

Spatial disparities in heatwave mortality were also observed between the different districts. In the Central, Central Black Earth and Volga economic districts, age-standardized death rates exceeded the baseline by more than 35% (Figure 5). However, the Northeast, North Caucasus and Ural districts did not experience increases of more than 15% (Table S.2, Supplementary material). The geographic variations in excess mortality reflect the patterns of the heatwave, with the most affected regions being located within the area where the extent of the heatwave was the greatest.

The regression analysis revealed substantial variations in the risk of death during the heatwave associated with the type of settlement. According to all the models with various variables included, excess mortality was higher in the largest cities, with a significant increase in risk being detected in cities with more than 500,000 inhabitants. For instance, cities belonging to the largest population groups (500,000–999,999 and 1,000,000+) experienced an increase in the age-standardized death rate that was 19.5% (CI: 13.4%–25.9%) and 21.3% (CI: 13.2%–29.9%) higher, respectively, than that in rural areas. In contrast, in cities and urban settlements with fewer than 100,000 inhabitants, the relative excess over the baseline was much lower, at only 4.0% (CI: 0.4%–7.7%).

The urban-rural dipartites in excess mortality persisted after adjusting for extreme heat, as shown in Table S.3 (Supplementary material). In the analysis, thermal stress, as assessed by any of the metrics, whether with or without one-week lags, was found to be a statistically significant control variable for assessing the differences based on the size of the settlement. While the findings were consistent across all models, the models including the Heat Index demonstrated the best fit and were more appropriate based on information criteria (AIC and BIC). Thus, a one-degree increase in the deviation of mean Heat Index values from 30-year summer averages was associated with a 10.4% (CI: 9.7%–11.2%) increase in the risk of death during the specific week.

The inclusion of variables representing air pollution did not lead to a significant improvement in the model. Their impact was only modest in the presence of thermal stress, and none of the variables provided statistically significant coefficients (as shown in Table S.4, Supplementary material). Models assuming lag effects did not alter the results, with the increases in death rates in cities of different sizes compared to those in rural areas being similar in all cases (as indicated in Table S.5, Supplementary material).

The models that considered the varying associations of extreme heat with excess mortality by the size of the settlement provided additional insights. Comparable temperature extremes led to significantly higher death risks in larger cities. Once more, the Heat Index provided the best fit (as shown in Table S.6, Supplementary material).

A one-degree increase in the weekly Heat Index anomaly was associated with a 17.1% (CI: 14.3%–20.1%) and a 18.4% (CI: 14.3%–22.6%) greater risk of death in the two most populous city categories (500,000–999,999 and 1,000,000+, respectively). In the smallest cities and in rural areas, the “heat slope” was less steep, amounting to 9.3% (CI: 8.0%–10.7%) and 6.5% (5.2%–7.9%), respectively. The impact of air pollution was found to be limited, with minimal changes in the observed associations (Table S.7, Supplementary material).

We examined the relationships between the risk of death during the heatwave and the size and the type of settlement. The results of regression analysis supported the empirical findings, particularly when accounting for environmental factors. We found evidence of the influence of urban heat island effects on mortality rates, with a more significant impact being observed in the largest cities with populations exceeding 500,000. A study conducted in Poland (Graczyk et al., 2022) provided a comparable population threshold for heat-related mortality effects in urban areas.

The impact of urban heat islands tends to be more pronounced in areas characterized by high population density, reduced urban greenery and densely built-up zones (Sera et al., 2019; Choi et al., 2022; Lee et al., 2022). These factors are likely to be important contributors to the observed differences in excess mortality.

In Russia, there is a positive correlation between the proportion of residents living in urban environments and multi-story buildings and the size of the settlement. Consequently, many residents in small towns live in low-rise, “rural” built-up areas (Kolosov and Nefedova, 2014), which experience less pronounced urban heat island effects.

Furthermore, the prevalence of air conditioning is considered one of the most important contributors to the decline in heat-related mortality in many countries (Sera et al., 2020). Since a significant portion of households in Russia were not equipped with air conditioners in 2010, the importance of air conditioning as a mitigating factor during the studied heatwave was diminished (Shaposhnikov et al., 2014; Otrachshenko et al., 2017). However, due to the absence of statistical data at the local level, we were unable to confirm these associations.

The study period was characterized by a significant decline in mortality. Life expectancy began to gradually increase from the mid-2000s onwards, initially in the capital and in the largest cities (Shchur, 2019; Shchur et al., 2021). As a result, a substantial geographic and hierarchical disparity in mortality had emerged by 2010 (Timonin et al., 2020). While mortality during the heatwave, as measured by excess age-standardized death rates, was nearly uniform across various settlement categories, the relative increase in mortality was considerably greater in the largest cities. Due to this disparity, it is reasonable to compare relative figures only when assessing heatwave effects.

The preceding rapid decline in mortality was largely attributed to a decrease in deaths from external causes at working ages, which were closely linked to alcohol consumption (Shkolnikov et al., 2013; Grigoriev et al., 2014; Timonin et al., 2017). Simultaneously, heat-related mortality in Russia, like elsewhere, tends to be mainly concentrated in older age groups (Shaposhnikov et al., 2014; Grigorieva and Revich, 2021). Thus, we cannot assume that recent mortality trends had a direct influence on the results, although they might have altered the calculation of the baseline, potentially impacting the outcomes. Therefore, the regression coefficients for the dummy variables representing city size in the interaction models may partially reflect the imprecise estimation of the baseline. However, none of these coefficients were found to be statistically significant.

Excess mortality during heatwaves is generally attributed to extreme heat stress, which affects individuals in various ways, including by increasing the occurrence of heat-related illnesses and by worsening pre-existing health conditions. While air temperature is often regarded as the primary determinant of thermal stress, a number of studies have employed alternative parameters, such as biometeorological indices, to represent the perceived heat, which may be influenced by multiple additional factors.

In our analysis, we incorporated multiple biometeorological indices in addition to two-meter air temperatures to more accurately represent the impact of thermal stress on heat-related mortality. Although all of these parameters significantly explained spatial differences in excess mortality during the heatwave, formal statistical tests revealed that the Heat Index was the most comprehensive predictor in our study. The selection of the exact indicator of thermal stress is not governed by a universal rule, but rather depends on data accessibility, convenience and model fit (Barnett et al., 2010; Vaneckova et al., 2011; Lo et al., 2023).

The region with the highest observed risk of death coincided with the area that was most affected by the heatwave, and that experienced the highest level of heat stress. In other parts of European Russia, including in the North and Northeast, Ural and North Caucasus regions, there were slight increases in death rates. Thus, in these areas there were only minor urban-rural differences in excess mortality.

An investigation of the temperature-mortality relationship using interaction models revealed a steeper increase in the risk of death in the largest cities. However, due to the relatively coarse spatial resolution of the gridded datasets from which temperatures and air pollution were extracted, the measurements used in the regression analysis might only represent the “background” conditions, while disregarding the influence of the local microclimate. Therefore, it is likely that the observed relationship reflects the variations in urban heat island effects in cities of different sizes rather than variations in heat-related mortality risks for other reasons. Thus, it may be assumed that the urban heat island effects were much more pronounced in cities with more than 500,000 inhabitants, though they also occurred in smaller cities.

As demonstrated by Spangler et al. (2019), gridded datasets based on extrapolations provide more accurate measurements of temperature-mortality relationships in remote and rural areas, while underestimating them in large metropolitan areas. This discrepancy could be attributable to the failure to account for local conditions, which are also prevalent when using data from in situ observations, as most ground stations are situated at airports outside the urban environment.

The 2010 summer heatwave in European Russia was accompanied by widespread forest fires, which generated smog and the associated health problems (Shaposhnikov et al., 2014). The burning of forests leads to elevated concentrations of various pollutants and suspended microparticles in the air. In addition to forest fires, air pollution during the heatwave was attributed to air stagnation and photochemical transformations. However, distinguishing the specific influences of these factors can be complex.

Although CAMS considered both anthropogenic emissions and pollution from fires based on data retrieved from satellite observations, the association between air pollutant concentrations and the risk of death during the heatwave was found to be weak when accounting for thermal stress. None of the tested air pollutants could be identified as the primary modifier of heat-related mortality, as their inclusion did not substantially improve the model. The inclusion of formaldehyde, which is typically not considered an outdoor air pollutant, produced comparable outcomes. Previous studies have associated outdoor formaldehyde concentrations with nonaccidental mortality (Ban et al., 2022), and there is evidence that these concentrations increased during the 2010 summer heatwave (Sitnov, 2011).

The strong correlation between air pollutant concentrations complicates the separation of their effects and the identification of the most significant of these concentrations. While this challenge could be addressed through the use of reanalysis data, which provide approximate estimates based on modelling, this approach cannot precisely represent local conditions (Cobelo et al., 2023). This highlights the importance of considering local-level microclimate characteristics when assessing heat-related death risks and estimating the effects of urban heat islands (Analitis et al., 2012).

Several additional limitations of the study should be acknowledged. The aggregation of all the data on a weekly basis may have provided a temporal scale that was too coarse for heat-related mortality analysis. Consequently, peak and extreme observations were not captured due to averaging by weeks. Therefore, despite the prolonged duration of the 2010 heatwave, many short-term effects were not identified in our study. The influence of lagging in assessments of the effects of temperature and air pollution on excess mortality has been well documented and discussed in previous research (Anderson and Bell, 2009). However, since strong lag effects are observed within several days, lagging did not substantially alter our week-specific estimates.

Due to the aggregation of our data at the regional level, numerous local patterns could be overlooked. The coarse spatial resolution of gridded datasets limited our ability to identify local conditions, which was particularly critical in urban areas due to the improper consideration of urban heat islands. Since the regions of European Russia often cover significant areas, the weather and climate within them might vary. To mitigate this problem, we reweighted the observations by the population of municipalities; however, it is important to acknowledge that some inconsistencies may still persist. Furthermore, numerous studies have found evidence of a link between heat-related mortality and abrupt changes in temperature (Cao et al., 2009; Lin et al., 2013; Vicedo-Cabrera et al., 2016), which could not be identified within the scope of our study.

Despite the relatively high level of aggregation in both spatial and temporal scopes, the total number of deaths in specific regions was insufficient to provide reliable estimates. Shaposhnikov and Revich (2016) argued that in territories with fewer than 200,000 inhabitants, the confidence intervals for mortality are excessively wide. Of the 184 groups of settlements by region that we examined in our study, 27 had a population size under than this threshold. The problem of small sample sizes was partially mitigated by grouping by weeks. However, in certain regions, particularly in rural areas, the estimates of excess mortality were not statistically significant.

The uneven distribution of population within regions may have resulted in specific settlement patterns being overlooked when reweighting gridded data for the total population. This issue was especially significant for the largest urban agglomerations, which also included the surrounding areas of large cities affected by urban heat islands. This may have distorted the results, especially in the Moscow region, where a substantial share of the population resides within the area of continuous urban development around Moscow.