Fig. 3: The trends in heat-related excess mortality by period, cause, and individual characteristics in China assuming no adaptation or population changes.

The excess mortality (red bar) was represented as the fraction of deaths (%) attributable to high temperature, with 95% empirical confidence intervals (the vertical black line). A distributed lag nonlinear model was used to estimate the district/county-specific temperature-mortality association with 14 days of lag adjusted for time trends and day of the week, which were pooled in a multivariate meta-analysis. Then, estimates of the attributable fraction of deaths due to high temperature, defined as temperatures above the optimum temperature, were calculated at the national level. Monte Carlo simulations generating 1000 samples were computed to produce empirical confidence intervals. The analyses were separately repeated by cause of death and individual characteristics. Source data are provided as Supplementary Data 3.