A new perspective of simultaneous exposure of risk factors to explore the role of ambient temperature associated with mortality

Extensive epidemiological evidence has evaluated the effect of a single climate-induced environmental stressor on cause-specific mortality. However, this exposure pattern often fails to reflect the real-world context of simultaneous exposure to multiple stressors. The conventional additive models face several methodological limitations-high susceptibility to multicollinearity, low predictive accuracy, and a lack of straightforward interpretability of risk estimation-which challenge comprehensive health risk assessment. We establish a Sequence-to-Sequence architecture combined with local feature attribution method to investigate the effect of ambient temperature on the association between coexisting meteorological factors and air pollutants and mortality for Jiangsu Province, China from 2014 through 2019. The model usually generated satisfactory performance in capturing the temporal trajectory of deaths of non-accidental causes (NAC), cardio-cerebrovascular diseases (CD), and respiratory diseases (RD) with extreme outcomes. Quantitative insights in temperature-related excess deaths at given temperature benefited from optimal local explanations. E.g., annually, an average of 20 300, 12 661, 2 511 excess deaths of NAC, CD, and RD, respectively, were associated with non-optimal temperatures over the entire study period; corresponding statistics of 1 192, 806, and 98 were attributable to extreme low temperatures, and those of 3 604, 2 238 and 480 were attributable to temperature drops. The magnitude and pattern of interaction effects even from the same pairwise features might differ by cause-specific mortality in each city. Our findings highlight the significance of simultaneous exposure in tailored public health policies. The methodology would have good performance with data availability in risk assessments at national and global scale.