Global, regional, and national burden of mortality associated with short-term temperature variability from 2000–19: a three-stage modelling study

Summary Background Increased mortality risk is associated with short-term temperature variability. However, to our knowledge, there has been no comprehensive assessment of the temperature variability-related mortality burden worldwide. In this study, using data from the MCC Collaborative Research Network, we first explored the association between temperature variability and mortality across 43 countries or regions. Then, to provide a more comprehensive picture of the global burden of mortality associated with temperature variability, global gridded temperature data with a resolution of 0·5° × 0·5° were used to assess the temperature variability-related mortality burden at the global, regional, and national levels. Furthermore, temporal trends in temperature variability-related mortality burden were also explored from 2000–19. Methods In this modelling study, we applied a three-stage meta-analytical approach to assess the global temperature variability-related mortality burden at a spatial resolution of 0·5° × 0·5° from 2000–19. Temperature variability was calculated as the SD of the average of the same and previous days’ minimum and maximum temperatures. We first obtained location-specific temperature variability related-mortality associations based on a daily time series of 750 locations from the Multi-country Multi-city Collaborative Research Network. We subsequently constructed a multivariable meta-regression model with five predictors to estimate grid-specific temperature variability related-mortality associations across the globe. Finally, percentage excess in mortality and excess mortality rate were calculated to quantify the temperature variability-related mortality burden and to further explore its temporal trend over two decades. Findings An increasing trend in temperature variability was identified at the global level from 2000 to 2019. Globally, 1 753 392 deaths (95% CI 1 159 901–2 357 718) were associated with temperature variability per year, accounting for 3·4% (2·2–4·6) of all deaths. Most of Asia, Australia, and New Zealand were observed to have a higher percentage excess in mortality than the global mean. Globally, the percentage excess in mortality increased by about 4·6% (3·7–5·3) per decade. The largest increase occurred in Australia and New Zealand (7·3%, 95% CI 4·3–10·4), followed by Europe (4·4%, 2·2–5·6) and Africa (3·3, 1·9–4·6). Interpretation Globally, a substantial mortality burden was associated with temperature variability, showing geographical heterogeneity and a slightly increasing temporal trend. Our findings could assist in raising public awareness and improving the understanding of the health impacts of temperature variability. Funding Australian Research Council, Australian National Health & Medical Research Council.

Global, regional, and national burden of mortality associated with short-term temperature variability from 2000 to 2019: a three-stage modelling study Supplementary Appendix Table of contents Text S1. eMethods. Table S1. Descriptive statistics by country/region included in MCC study. Table S2. Missing rates by country/region included in MCC study. Table S3. Meta-regression models for explaining variation in overall TV effects: Cochran Q test for heterogeneity, I 2 statistics for residual heterogeneity. Table S4. The average annual daily temperature variability in 2000 and 2019 by continent and region. Figure S1. The long-term trend of temperature variability after seasonal-trend decomposition by region from 2000 to 2019. Table S5. Percentage change in mortality associated with an interquartile increase in temperature variability by country. Table S6. Average annual percentage excess in mortality and excess deaths per 100,000 residents due to temperature variability in 2000-19 by continent and region. Table S7. Average annual percentage excess in mortality and excess deaths per 100,000 residents due to temperature variability between 2000-19 by the indicators of Köppen-Geiger climate classification. Figure S2. Scatter plots of percentage excess in mortality associated with temperature variability from 2000 to 2019. Figure S3. Scatter plots of excess death rate associated with temperature variability from 2000 to 2019. Figure S4. Leading 20 countries of excess deaths in 2000 and 2019. Table S8. Average annual global percentage excess in mortality and global excess deaths per 100,000 residents due to temperature variability in 2000-19 on different exposure days. Table S9. Results of sensitivity analyses on global percentage excess in mortality and global excess deaths per 100,000 residents. Table S10. Results of sensitivity analyses on overall TV-mortality association based on 500 locations with relative humidity data. Table S11. Average annual percentage excess in mortality under the counterfactual scenario of grid-specific mean temperature variability by continent and region.

Sensitivity analyses
The range of each parameter was decided based on prior knowledge and research.
(1) A maximum exposure period of seven days for short-term TV exposure is commonly used in prior research 1-5 . However, the use of a single length of exposure is insufficient to provide evidence on the short-term impact of TV, thus we use alternative lag from 1 to 6 days and from 8 to 10 days to check if the association still exists.
(2) A lag period of 21 days for mean temperature was commonly used to include the long delay of the effects of cold temperatures 6 . Here, we used longer lag days (24 or 28 lag days) in the sensitivity analyses in case a lag of 21 days was not enough to capture the temperature effects on mortality.
(3) We chose the most commonly used value of the degree of freedom in the main analysis and applied neighboring values on each side to make sure that our results can exist independently of any particular. Small degrees of freedom will fail to capture the main long-term patterns closely, whereas too many will result in overfitting--that a very 'wobbly' function which may compete with the variable of interest to explain the shortterm variation of interest, widening confidence intervals of relative risk estimates 7 . (4) We adjusted relative humidity in the first stage using data from 500 locations with relative humidity data. As 500 locations failed to cover Africa and areas with polar and alpine climates, we were unable to estimate the global mortality burden associated with TV using data from these locations. Thus, we compared the TV-mortality associations with and without adjustment of relative humidity in the sensitivity analyses.   Figure S1. The long-term trend of temperature variability after seasonal-trend decomposition by region from 2000 to 2019.      Table S11. Average annual percentage excess in mortality under the counterfactual scenario of grid-specific minimum temperature variability by continent and region.

Percentage excess in mortality (%)
Counterfactual scenario of grid-specific minimum TV Counterfactual scenario of no TV variation