Background: While many studies have explored the correlation between environmental factors and influenza, research on their potential causal associations remains limited. Further, the impact of temperature changes between neighboring days (TCN) has not been thoroughly investigated.
Methods: Influenza-like illness (ILI) data, meteorological indicators, and air pollutant levels were collected in Jinan, China (2015-2019). Gradient boosting decision trees (GBDT) were used to identify key environmental variables. The distributed lag nonlinear models (DLNM) and empirical dynamic modeling (EDM) framework were then applied to explore their nonlinear associations with and potential causal effects on influenza infection. Subgroup analysis was also performed by different age groups.
Results: GBDT identified absolute humidity (AH), atmospheric pressure (AP), sulfur dioxide (SO2), and ozone (O3) as key factors, along with TCN as a key variable of interest. DLNM results revealed J-shaped and bimodal exposure-response relationships for TCN and AH, respectively, with increased relative risks (RRs) under low AH. SO2 was positively associated with influenza risk. The highest RRs were 1.190 (95% confidence interval (CI): 1.070-1.322) observed at 1012 hPa for AP and 3.373 (95% CI: 2.650-4.294) at 125 [Formula: see text]g/m3 for O3. EDM results indicated that long-lag AP and SO2 may have a potential driving effect on influenza incidence. Positive effects on influenza were observed when TCN > -5 °C, AP > 1100 hPa, and across the full SO2 range. Children aged 0-4 were more sensitive to AP and SO2, while the aged 5-59 were more affected by TCN, AH, and O3.
Conclusion: This study demonstrated that both the DLNM and EDM methods consistently revealed the complex and nonlinear effects of specific environmental factors on influenza infection. Specifically, TCN > 5 ℃, AP > 1100 hPa, and SO2 > 60 [Formula: see text]g/m3 were associated with increased risk of influenza infection. Children aged 0-4 years and individuals aged 5-59 years exhibit different susceptibility patterns to environmental factors. These findings can inform public health strategies for influenza prevention, particularly in the context of increasing air pollution and climate variability.
Keywords: Air Pollution; Convergent Cross Mapping; Distributed Lag Nonlinear Models; Influenza; Meteorological Factors; Temperature Variability.
© 2025. The Author(s).