Mortality attributable to ambient PM2.5 exposure across regions in China from 2005 to 2020

China has implemented rigorous clean air policies since 2013 to address PM 2.5 pollution particularly affecting economically developed regions. The Fusion relative risk model was recently developed to better capture the association between long-term PM 2.5 exposure and mortality risk, but it has not been used in China. We estimated cause-specific PM 2.5 attributable deaths among adults (≥ 25 years) across China over 2005–2020 at a 0.1° × 0.1° spatial resolution using the Fusion model, and also adopted the Global Exposure Mortality Model (GEMM) for comparison. We investigated 31 provinces and targeted five mega city clusters: Beijing–Tianjin–Hebei cluster (BTH), Yangtze River Delta cluster (YRD), Pearl River Delta cluster (PRD), Cheng–Yu cluster (CY) and Middle reaches of the Yangtze River cluster (MYR). The Fusion model estimated PM 2.5 attributable deaths increased from 2.38 million (95 % UI: 2.06-2.64) in 2005 to 2.68 million (95 % UI: 2.33-2.95) in 2013 (12.6 %), then declined by 11.6 % to 2.37 million (95 % UI: 2.04-2.65) in 2020. Ischemic Heart Disease (IHD) and Stroke contributed most to the total attributable deaths (33.4 % and 35.0 % in 2020). Compared with the GEMM, the Fusion model generated higher attributable mortality estimates for IHD, Stroke and Chronic Obstructive Pulmonary Disease, but lower estimates for Lung Cancer and Lower Respiratory Infections. PM 2.5 attributable deaths were clustered in densely populated and highly polluted regions, with Henan and Shandong bearing the highest mortality burden among the 31 provinces. PM 2.5 attributable deaths in BTH, YRD, PRD, CY and MYR declined by 5.3 %, 11.5 %, 18.0 %, 25.4 % and 18.6 % respectively over 2013–2020, with greater declines in attributable mortality rates (18.9 %, 23.8 %, 33.8 %, 27.3 % and 24.1 %). Future clean air policies in China should consider regional disparities and continue prioritizing highly polluted and densely populated urban areas, including the five mega city clusters.