Climate Change, Summer Temperature, and Heat-Related Mortality in Finland: Multicohort Study with Projections for a Sustainable vs. Fossil-Fueled Future to 2050

Abstract

Background: Climate change scenarios illustrate various pathways in terms of global warming ranging from "sustainable development" (Shared Socioeconomic Pathway SSP1-1.9), the best-case scenario, to 'fossil-fueled development' (SSP5-8.5), the worst-case scenario.

Objectives: We examined the extent to which increase in daily average urban summer temperature is associated with future cause-specific mortality and projected heat-related mortality burden for the current warming trend and these two scenarios.

Methods: We did an observational cohort study of 363,754 participants living in six cities in Finland. Using residential addresses, participants were linked to daily temperature records and electronic death records from national registries during summers (1 May to 30 September) 2000 to 2018. For each day of observation, heat index (average daily air temperature weighted by humidity) for the preceding 7 d was calculated for participants' residential area using a geographic grid at a spatial resolution of $1\mathrm{km}\times 1\mathrm{km}$. We examined associations of the summer heat index with risk of death by cause for all participants adjusting for a wide range of individual-level covariates and in subsidiary analyses using case-crossover design, computed the related period population attributable fraction (PAF), and projected change in PAF from summers 2000-2018 compared with those in 2030-2050.

Results: During a cohort total exposure period of 582,111,979 summer days (3,880,746 person-summers), we recorded 4,094 deaths, including 949 from cardiovascular disease. The multivariable-adjusted rate ratio (RR) for high ($\ge 21°C$) vs. reference ($14-15°C$) heat index was 1.70 (95% CI: 1.28, 2.27) for cardiovascular mortality, but it did not reach statistical significance for noncardiovascular deaths, $\text{RR}=1.14$ (95% CI: 0.96, 1.36), a finding replicated in case-crossover analysis. According to projections for 2030-2050, PAF of summertime cardiovascular mortality attributable to high heat will be 4.4% (1.8%-7.3%) under the sustainable development scenario, but 7.6% (3.2%-12.3%) under the fossil-fueled development scenario. In the six cities, the estimated annual number of summertime heat-related cardiovascular deaths under the two scenarios will be 174 and 298 for a total population of 1,759,468 people.

Discussion: The increase in average urban summer temperature will raise heat-related cardiovascular mortality burden. The estimated magnitude of this burden is $>1.5$ times greater if future climate change is driven by fossil fuels rather than sustainable development. https://doi.org/10.1289/EHP12080.

Figure 1.

Selection of participants living in six Finnish cities from two cohort studies, 2000–2018. Summertime is between and 1 May and 30 September, a total of 150 d. Note: HeSSup, Health and Social Support Study.

Figure 2.

Spatial distribution of monthly heat index ($1\mathrm{km}\times 1\mathrm{km}$ resolution) in Finland, averaged over 2000–2019. The number to the left of each heat map is the countrywide monthly average heat index (°C). The corresponding monthly average temperature is slightly higher than the monthly average heat index: 8.2°C (May), 12.9°C (June), 16.4°C (July), 14.1°C (August), and 9.2°C (September). Note: AUG, August; JUL, July; JUN, June; SEP, September.

Figure 3.

Validation of the density-based definition of heat island using additional high-resolution temperature and NDVI measurements. (A) Data from a $12\mathrm{km}\times 12\mathrm{km}$ area of the city Turku, June 2018. Urban heat islands, defined by a population density of $\ge 500$ people per $250\mathrm{m}\times 250\mathrm{m}$ grid, are shown by black circles. Colors ranging from blue to red indicate temperatures in these grids. (B) The distribution of NDVI by urban heat island status in the same area and at the same time period. The $\text{mean}\pm \text{SD}$ of the NDVI was lower in urban heat islands ($0.40\pm 0.14$) than elsewhere ($0.65\pm 0.12$, $t$-test, $p<0.0001$). Note: AUG, August; JUL, July; JUN, June; NDVI, Normalized Difference Vegetation Index; SD, standard deviation; SEP, September.

Figure 4.

Spatial distribution of decadal May–September heat index change in Finland between 1980–1999 and 2000–2019. The number to the left of each heat map is the countrywide average decadal change in heat index between 1980–1999 and 2000–2019. The corresponding average decadal change in monthly temperature is: 0.57°C (May), $-0.02°\mathrm{C}$ (June), 0.58°C (July), 0.63°C (August), and 0.66°C (September).

Figure 5.

Case-crossover analysis of exposure to high heat index and risk of summertime all-cause and cause-specific mortality. Pooled individual-level data from two cohort studies in six Finnish cities, 2000–2018, were used. The number of participants in each analysis is the same as the number of deaths (range: 694–4,094). Conditional logistic regression with bidirectional control sampling was used for analysis. The analysis compares the odds of being exposed to high heat ($\ge 21\mathrm{°}\mathrm{C}$) in case time compared with control times. In the design “1 year before and after death,” control dates are 1 y before and after the date of death. In the design “Same day of the weeks at the month,” control dates are the same day of the week during the case month as the death day. All time-invariant covariates are controlled by the study design.

Figure 6.

Heat exposure and risk of summertime cardiovascular and noncardiovascular death by heat indicator in population model and case-crossover analyses. Pooled individual-level data from two cohort studies in six Finnish cities, 2000–2018, were used. Population models are based on Poisson regression analysis. Mortality rate and age, sex, and calendar year-adjusted rate ratio for participants exposed to high heat index ($\ge 21°\mathrm{C}$) compared with those unexposed (heat index 14–20°C) are shown. Case-crossover models are based on conditional logistic regression with bidirectional control sampling. The analysis compares the odds of being exposed to high heat ($\ge 21\mathrm{°}\mathrm{C}$) in case time (the date of death) compared with control times (1 y before and after the date of death). All time-invariant covariates are controlled by the study design. The number of participants in the case-crossover analysis is the same as the number of deaths. Note: Tmax, mean of daily maximum temperatures.

Figure 7.

Heat exposure and risk of summertime cardiovascular death in population subgroups from a stratified population model and case-crossover analyses. Pooled individual-level data from two cohort studies in six Finnish cities, 2000–2018, were used. Population models and related tests for interaction are based on Poisson regression analysis. Mortality rate and age, sex, and calendar year-adjusted rate ratio for participants exposed to high heat index ($>21°\mathrm{C}$) compared with those unexposed (heat index 14–20°C) are shown. Case-crossover models and related tests for interaction are based on conditional logistic regression with bidirectional control sampling. The analysis compares the odds of being exposed to high heat ($\ge 21\mathrm{°}\mathrm{C}$) in case time (the date of death) compared with control times (1 y before and after the date of death). All time-invariant covariates are controlled by study design. The number of participants in case-crossover analysis is the same as the number of deaths. Note: NDVI, Normalized Difference Vegetation Index.

Figure 8.

Observed and predicted burden of summertime heat-related cardiovascular death burden by climate change scenarios. (A) Observed decadal change in summertime heat index between 2000 and 2019 and projected decadal change in summertime heat index between 2030 and 2050 in Finland. (B) Summertime heat-related cardiovascular death burden as indicated by PAFs in participants living in six Finnish cities for 2000–2018 and 2030–2050 by climate change scenario. The whiskers in the bars represent 95% confidence intervals. Estimations in (B) are based on pooled individual-level data from two cohort studies in six Finnish cities, 2000–2018 ($N=\mathrm{363,754}$). Note: PAF, population attributable fraction; SEP, September; SSP, Shared Socioeconomic Pathway.