Application and Comparison of Three Time-Series Causal Discovery Methods in Environmental Epidemiology

Objective To systematically elaborate the theoretical foundations and technical paradigms of causal analysis methods for time-series data and to explore their application efficacy in environmental epidemiology through empirical studies.Methods A multi-method causal discovery framework was adopted to perform an empirical analysis of the time-series data of environmental factors and influenza incidence rates across 31 provinces in China. The methodological system included: the PCMCI+ algorithm based on conditional independence tests, the VAR-LiNGAM method based on structural equation models, and the CCM model based on nonlinear state-space reconstruction.Results All three models confirmed that mean temperature exerted a negative causal effect on influenza incidence rates with a one-week lag (p < 0.05), with relative causal strengths of -0.019, -0.047, and -0.078, respectively. However, neither the PCMCI+ method nor the VAR-LiNGAM method identified any reverse causal associations from influenza incidence rates to environmental factors (p > 0.05).Conclusion The three methods demonstrate complementary advantages in the context of environmental epidemiology. By constructing a multi-method cross-validation framework, the inherent limitations of a single method can be effectively overcome, facilitating more robust causal analysis to be carried out in environmental health risk assessment.