Ozone as an environmental driver of influenza

Influenza poses a significant global health and economic burden, causing hundreds of thousands of deaths annually. Understanding the environmental drivers of influenza dynamics is crucial for developing effective mitigation strategies. While the roles of temperature and humidity are well-studied, the potential influence of ambient ozone (O3) remains relatively unexplored. Previous research in Hong Kong hinted at a negative association between O3 and influenza transmissibility, aligning with laboratory findings suggesting O3's virucidal properties and potential to boost host immunity. This study expands on this limited research by investigating this relationship using comprehensive data from the United States, a geographically diverse region with varying climates compared to the subtropical Hong Kong. The aim is to rigorously determine whether this O3-influenza relationship holds true in temperate climates as well, using an integrated methodological framework to strengthen causal inference.

Existing literature extensively examines the relationship between influenza activity and absolute humidity (AH) and temperature (T). Studies have reported varying associations, highlighting the complex interplay of factors influencing influenza dynamics. A limited number of population-level studies explored the O3-influenza link, with mixed findings. Some studies in Hong Kong and Brisbane reported positive associations, but these lacked robust controls for confounding factors or focused on aggregated respiratory illnesses rather than influenza specifically. However, a more recent time-series analysis in Hong Kong showed a negative association between ambient O3 and influenza transmissibility. In vitro and in vivo studies provided further support, indicating that O3 can inactivate influenza viruses and may modulate host immunity, potentially reducing infection severity.

This study employed a rigorous methodological framework integrating three distinct data analysis approaches to enhance causal inference: 1. **Convergent Cross Mapping (CCM):** This causality test, rooted in state-space reconstruction theory, assesses the ability of a hypothesized driver variable (O3, AH, or T) to predict the response variable (influenza activity). The method checks for convergence in predictive skill as data increase, thus distinguishing true causality from spurious correlation. Seasonal surrogates were used to rule out spurious relationships from shared seasonality. Multivariate S-map analysis quantified the effect size. 2. **Peter-Clark-momentary-conditional-independence plus (PCMCI+):** This graphical modeling approach constructs a directed acyclic graph (DAG) to depict causal relationships between variables. The method uses conditional independence tests to identify and orient edges, accounting for autocorrelation and common drivers. Analysis was conducted at both state and nationwide levels. 3. **Generalized Linear Model (GLM):** This regression-based approach assessed the statistical association between environmental variables and influenza activity, controlling for secular trends, seasonality, and autocorrelation. State-level regression coefficients were then pooled using meta-analysis. Weekly data on influenza, ILI, O3, AH, and T were obtained from publicly available sources for the period of October 3, 2010 to September 27, 2015. A composite influenza activity measure combined laboratory-confirmed cases and ILI reports to improve accuracy. Analyses focused on the influenza season (October to May), excluding states with incomplete data.

All three analytical methods consistently revealed a negative impact of ambient O3 on influenza activity at a 1-week lag. Specifically: * **CCM:** Nationwide meta-analysis showed that O3 was a significant driver of influenza activity (Pmeta < 1.0 × 10⁻³), with a stronger effect than AH and T. State-specific analysis revealed mostly negative effects, indicating a consistent inhibitory influence of O3 across different regions. * **PCMCI+:** Both state-level and nationwide causal graphs consistently indicated a direct negative effect of O3 on influenza activity at lag 0 and lag 1. AH showed mixed effects, while T's effect was largely indirect, mediated through O3. * **GLM:** Meta-analysis of state-level GLM results confirmed a statistically significant negative association between O3 and influenza activity at a 1-week lag (p < 5.9 × 10⁻¹⁰). AH and T were also negatively associated, but at different lags. Table 1 in the original paper presents a detailed summary of the findings across all three methods and lags. Figure 2 demonstrated the CCM causality tests for each environmental factor and Figure 3 is a map of the United States showing state-specific effect strength estimates for ozone (O3) affecting influenza activity at a 1-week lag. Figure 4 shows the results for PCMCI+ and Figure 5 for GLM.

This study provides robust evidence supporting a negative association between ambient ozone and influenza activity. The consistent findings across three independent analytical methods with differing theoretical underpinnings strengthen the causal interpretation. While previous studies yielded mixed results, the current study, utilizing a more comprehensive methodology and addressing potential confounding factors, offers a more conclusive picture. The negative effect of O3 is likely not due to its direct virucidal effects at typical ambient concentrations, but rather through its immune-modulatory properties. O3 exposure triggers various immune responses, including the release of IL-33, a cytokine that plays a role in both antiviral defense and tissue repair. This suggests a mechanism whereby O3 primes the immune system, enhancing its ability to combat influenza infection.

This study demonstrates a significant negative association between ambient ozone and community-level influenza activity using a robust, multi-method approach. The observed effect is likely mediated by ozone's impact on the immune system rather than direct viral inactivation. These findings highlight the importance of considering air quality in public health interventions and underscore the need for further research exploring the intricate immunological mechanisms involved. Future research should investigate this relationship at finer spatial and temporal scales, incorporating subtype-specific data and accounting for potentially confounding factors like social dynamics and public health interventions. Further molecular studies to confirm the proposed mechanisms are also warranted.

Several limitations should be considered when interpreting the results. The reliance on passive surveillance data, while comprehensive, may be subject to reporting biases. Variations in testing capacity and healthcare-seeking behaviors across states and years could affect the accuracy of the influenza activity estimates. The aggregation of influenza cases across subtypes might mask lineage-specific responses to O3 and other environmental factors. Finally, the analysis was conducted using state-level weekly data over five years; therefore, the findings may not capture the full complexity of influenza dynamics that could vary on finer spatial and temporal scales.