Avoidable heat-related mortality in China during the 21st century

Rising temperatures driven by increased carbon emissions are a major climate risk to public health. The last decade saw peak global greenhouse gas emissions, with frequent extreme heat events worldwide. Heat exposure elevates core temperature and cardiac stress, leading to heat stroke, respiratory and cardiovascular diseases, and death; notable examples include the 1995 Chicago heatwave, the 2003 European heat event, the 2010 Russian heatwave, and the 2021 North American heatwave. As climate change advances, heat-related risks will intensify, especially in densely populated metropolitan areas, underscoring the imperative to reduce emissions to protect health. The IPCC’s latest assessment integrates climate and socioeconomic development via Shared Socioeconomic Pathways (SSPs) linked to Representative Concentration Pathways, with CMIP6 providing updated climate projections. Prior studies have projected heat risks in China by combining demographic projections and mortality records. However, there has been limited quantification of how much heat-related risk can be avoided by maximizing emission reductions, a gap that hinders effective infrastructure planning. China is already warmer than the global average and has experienced frequent extreme heat since 2000. Continued population growth together with rising temperatures suggests the need for expanded health infrastructure and targeted prevention. This study aims to quantify future avoidable heat-related mortality associated with climate change after accounting for population changes in China and to estimate the independent contributions of future temperature and population changes to heat-related deaths. The results inform strategic planning for health infrastructure to adapt to warming and demographic shifts, and the approach can be applied in other regions.

Study design and data: The study assessed heat-related mortality associated with anthropogenic climate change using observed non-accidental mortality and model-simulated temperatures. Daily non-accidental mortality (ICD-10 A00–R99) was obtained for 195 surveillance sites across 17 provincial-level areas in China (mostly 2010–2016; Guangzhou and Xining 2012–2016). Population projections were taken from gridded SSP datasets (SSP1, SSP2, SSP3, SSP5) at 0.5° resolution for 2010–2100, incorporating fertility, mortality, migration, education, and China’s two-child policy. Climate simulations: Eleven CMIP6 models (first realizations) provided historical (1900–2014) and future (2021–2100) daily mean surface air temperatures. Two historical forcings were used: ALL (anthropogenic + natural) and NAT (natural-only, counterfactual without human influence). Future ALL simulations were run under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. The NAT series was used as a counterfactual projection without anthropogenic emissions. Model outputs were bilinearly interpolated to 0.5° × 0.5° to match population grids. Anthropogenic contributions were computed as ALL minus NAT; proportions were (ALL−NAT)/ALL. Regions and periods: Seven subregions followed standard Chinese geographic divisions: Northeast (NE), North (NC), Northwest (NW), East (EC), Central (CC), Southwest (SW), and South (SC). Baseline period was 1995–2014; futures were 2021–2100, analyzed in four 20-year windows: 2021–2040, 2041–2060, 2061–2080, 2081–2100. Temperature–mortality relationship: A two-stage approach estimated exposure–response functions. Stage 1: At each site, quasi-Poisson regression with a distributed lag non-linear model (DLNM) modeled non-linear and lagged temperature effects on mortality, adjusting for time (natural cubic spline, 7 df/year), relative humidity (natural cubic spline, 3 df), and day of week. A 21-day lag captured delayed effects. Stage 2: Multivariate meta-analysis (restricted maximum likelihood) pooled site-specific associations within regions, and best linear unbiased predictions (BLUP) provided cumulative temperature–mortality curves for each region. Heterogeneity was assessed with Cochran’s Q and I2 statistics. Computation of heat-related mortality and deaths: For each grid cell and day, relative risks (RR) from regional exposure–response curves yielded attributable fractions (AF = (RR−1)/RR). Daily attributable deaths were AD_{x,d} = Mort_{x,y} × Pop_{x,y} × AF_{x,d}. Annual heat-related deaths (HD_{x,y}) were the sum over days with temperature above the regional minimum mortality temperature (MMT); annual heat-related mortality (HM_{x,y}) = HD_{x,y} / Pop_{x,y}. Regional totals were sums over grid cells; regional mortality was regional deaths divided by regional population. Regional MMTs were approximately: NE 22 C, NC 25 C, NW 24 C, EC 25 C, CC 26 C, SW 26 C, SC 25 C. Independent contributions of temperature and population: To disentangle drivers, Random Forest models were trained per subregion and period using gridded inputs of population, annual mean heat-related daily temperature (mean of days above MMT), and annual heat-related deaths. Gini importance provided independent contributions (percentage) of temperature and population changes to heat-related deaths. Uncertainty: Uncertainties arose from exposure–response estimation and climate model spread. Monte Carlo simulation generated 1000 samples of reduced BLUP coefficients assuming multivariate normality, applied across all CMIP6 models. Multi-model ensemble means were reported with 95% confidence intervals as the 2.5th and 97.5th percentiles. Additional analyses: Anthropogenic warming was quantified from ALL–NAT differences historically and into the future. An exploratory analysis linked historical mortality declines to economic and medical development (GDP per capita and hospital beds per 1000) and assessed a counterfactual where such development did not occur (using NAT baseline temperatures and 1949–1977 mortality levels, excluding 1958–1962 disaster years).

- Present-day attribution: Anthropogenic climate change currently accounts for about 17% (95% CI: 13–21%) of heat-related mortality in China. The anthropogenic share is higher at high latitudes and altitudes; NW shows 43% (95% CI: 37–49%), while EC is lowest at 16% (95% CI: 12–19%). About 30% of China’s land already experiences anthropogenic warming >1.0 C. - Future heat-related mortality under ALL (with emissions): National heat-related mortality is projected to reach ~1.5× current levels in 2021–2040 (0.027–0.032%, 95% CI: 0.021–0.039%) and may quadruple by 2081–2100 under SSP5-8.5 (0.105%, 95% CI: 0.085–0.125%). EC, SC, and SW will experience the highest heat-related mortality and absolute deaths; NE and NC the lowest. - NAT counterfactual (no anthropogenic emissions): Heat-related mortality would remain low and similar to current levels (0.015–0.017%, 95% CI: 0.008–0.025%). However, NW and SC would still experience 110–140% of current heat-related deaths due to natural climate variability and population changes. - Avoidable burden via mitigation: Averaged over 2021–2100, human-induced warming would account for 48–72% (95% CI: 40–76%) of total heat-related mortality, about 2.5–4 times the present-day share. This implies 40–70% of future heat-related deaths could be avoided by reducing anthropogenic emissions. Nationally, anthropogenic warming would cause about 15,576–21,532 (95% CI: 6605–31,183) heat-related deaths annually in 2021–2040, rising to 23,139–87,612 (95% CI: 15,470–106,736) by 2081–2100, with EC contributing ~40% of the national total. - Emission-scenario divergence and timing: Scenarios are similar in 2021–2040, with divergence growing thereafter. Under SSP5-8.5, anthropogenic heat-related mortality reaches 0.090% (95% CI: 0.071–0.110%) in 2081–2100, roughly triple other scenarios. Mitigation benefits on mortality emerge gradually over decades rather than immediately. - Drivers of change: Nationally, temperature increases and population changes contribute ~78% and ~22%, respectively, to future heat-related deaths. Temperature dominates (75–95%) in NW, NE, NC, SW; population contributes more (about 55%) in CC and SC; in EC, contributions are more balanced (temperature 50–75%, population 25–50%). - Economic/medical development counterfactual: Without historical anthropogenic emissions (and thus without associated economic/medical advances), current heat-related mortality would be 57% higher than observed, and up to 96% higher in SC, despite reduced warming. - Spatial warming patterns: Anthropogenic warming is and will remain stronger at high latitudes and altitudes (NE, NW, NC, SW), reaching 7.4–8.3 C (95% CI: 5.6–9.8 C) in 2100 under SSP5-8.5, versus 5.4–6.4 C (95% CI: 4.3–7.6 C) in other regions.

The study demonstrates that while natural climate variability currently dominates heat-related mortality in China, anthropogenic warming is already responsible for a significant and growing share, particularly in high-latitude and high-altitude regions. Over the 21st century, without mitigation, heat-related mortality and deaths will increase substantially, with densely populated EC and SC bearing the largest absolute burdens. Mitigation can avoid a large fraction of future heat-related deaths (about 40–70% on average for 2021–2100), and the avoided fraction increases over time as anthropogenic influence grows. However, due to the inertia of the climate system, near-term benefits (next two decades) are limited; thus, adaptation and health system preparedness are essential in the short term. The analysis of independent contributions shows temperature change is the principal driver in sparsely populated, rapidly warming regions (NW, NE, NC, SW), while population dynamics dominate in CC and SC. This suggests region-specific strategies: emission mitigation is crucial everywhere but especially impactful in high-latitude/altitude regions; in CC and SC, managing demographic changes and spatial population distribution, alongside strengthening health infrastructure, may yield substantial benefits. An important nuance is the dual role of anthropogenic emissions: they drive warming but have historically coincided with economic and medical advancements that reduce mortality. A counterfactual lacking such development indicates current heat-related mortality could have been markedly higher. Future socioeconomic development could partially offset warming impacts through improved healthcare and adaptation, implying the avoidable-deaths estimates may be conservative without explicitly modeling adaptation gains. Overall, sustained emission reductions combined with targeted adaptation and health system investments will be necessary to manage heat risks effectively.

Heat-related mortality in China is presently driven mainly by natural climate variability, but the anthropogenic contribution is increasing and will likely dominate within the next 20–40 years. Cutting human-induced emissions could prevent approximately 48–72% (95% CI: 40–76%) of future heat-related mortality in China over 2021–2100, with especially large benefits in high-latitude and high-altitude regions (NW, NE, NC, SW). Temperature increases will primarily drive future heat deaths in most regions, whereas population changes will dominate in CC and SC. Given limited near-term mitigation effects due to climate inertia, immediate investment in emergency response and healthcare infrastructure is essential, alongside region-specific adaptation strategies and consideration of population distribution. Future research should integrate climate, demographic, and socioeconomic dynamics, including adaptation and changing baseline mortality, to refine projections and inform planning.

- Humidity was controlled in the DLNM during historical model fitting, but future projections used temperature alone as the predictor; relative humidity changes under future scenarios were not explicitly projected in exposure–response, potentially biasing estimates. - Other meteorological and environmental factors (e.g., air pollutants, wind speed, solar radiation) that may affect the temperature–mortality relationship were not incorporated in future projections. - Population projections may diverge from actual future demographics; datasets lack age structure, and changing age distributions could alter vulnerability and baseline mortality. - Adaptation and socioeconomic development were not explicitly modeled in future exposure–response or baseline mortality, despite evidence that adaptation shifts minimum mortality temperatures and reduces heat sensitivity. - Uncertainties stem from exposure–response estimation and CMIP6 model spread; multi-model ensemble means with 95% CIs were used, but additional uncertainties (e.g., technological change) remain. - The counterfactual analysis of economic/medical development is simplified and may not capture complex interactions between emissions, development, and health outcomes.