For all periods, mortality attributable to cold was higher than that of heat. Mortality on modestly cold days accounted for the majority of mortality attributed to non-optimal temperature, which is in line with other studies (Gasparrini et al. 2015; Vicedo-Cabrera et al. 2018; Lee et al. 2018).
The estimates of the total AF for the last two periods are higher than those reported previously for Stockholm: Gasparrini et al. (2015) quantified a total AF of 3.9%, with cold accounting for 3.7% and heat for 0.2% in the Greater Stockholm area for the period 1990–2002 (our estimates for the same time period were total AF, 4.73%, with 4.51% due to cold and 0.22% due to heat). The present study used data from the county of Stockholm, which includes less urbanized settings and more disperse socioeconomic status on both an individual and neighborhood level.
We find modest evidence suggesting heat-related AF to be declining over time and strong evidence of a declining trend over time related to extreme heat. From the 1960s and onwards, the AF on modestly hot days was basically zero, which may partially be explained by increasing MMT and changes in the temperature-mortality relationships over time. Our findings of declining heat AF over time are in line with a recent study of 305 locations in ten countries, where the authors reported declining heat-related AFs over time in seven countries (Vicedo-Cabrera et al. 2018). Furthermore, Lee et al. (2018) reported decreasing heat-related AF over time in Taiwan and Korea, whereas it increased in Japan.
We find no strong evidence of a declining trend over time in cold-related AF. Cold-related AF peaked at 7.9% during the 1920s and 1930s, with more recent estimates of around 5.5% during the two latest periods of investigation. The magnitudes of the cold-related AFs, as well as the stable trend over time, show some similarities to those reported by Vicedo-Cabrera et al. (2018), implying that trends in cold-related impacts remain unclear.
It may not be appropriate to assume that higher temperatures would increase the number of heat-related deaths and decrease the number of cold-related deaths, nor may it be appropriate to not account for adaptation, which is likely to offset some of future increases in temperature-related mortality, as evidence suggests MMT not to be a stationary measure over time (Åström et al. 2016; Todd and Valleron 2015).
For Stockholm, Sweden, absolute and relative MMT increased over the course of the twentieth century (Åström et al. 2016), while the effect of extreme temperatures on mortality decreased (Åström et al. 2013), suggesting partial adaptation to increasing temperatures. Decreasing effects of temperature on mortality over long-time periods have been reported elsewhere (Davis et al. 2003a, b, Sheridan et al. 2009, Carson et al. 2006; Ekamper et al. 2009). However, it remains inevitably difficult to separate adaptation and acclimatization to increasing temperatures from general societal improvements associated with development, such as standard of living and the quality and accessibility of health care facilities (Sellers and Ebi 2017).
Using historic data, we found that the AF on cold days remained stable over time, which suggests that mortality during the colder months may not simply decline as temperatures increase in the future. How winter mortality will continue to evolve will remain a difficult component in the projections of future temperature-related mortality (Ebi and Mills 2013), as evidence suggests winter mortality may modify heat-related mortality the following summer (Rocklöv et al. 2009; Stafoggia et al. 2009).
With ongoing climate warming and associated adaptive processes, the temperature-mortality relationships on both sides of MMT may change, with the magnitude and direction of change uncertain. More research is needed to enhance estimates of the burden related to cold and heat in the future.