Ozone as an environmental driver of influenza

Fang Guo, Pei Zhang, Vivian Do, Jakob Runge, Kun Zhang, Zheshen Han, Shenxi Deng, Hongli Lin, Sheikh Taslim Ali, Ruchong Chen, Yuming Guo, Linwei TianNature Communications 2024 · DOI

TL;DR

While temperature and humidity are the "usual suspects" in seasonal influenza modeling, this study identifies ambient ozone ($O_3$) as a significant, previously overlooked environmental driver. Using a robust "triangulation" of three distinct causal inference methods, researchers found that higher ozone levels consistently inhibit influenza activity in the U.S. at a one-week lag.

What Problem Does This Paper Address?

Influenza imposes a massive global health burden, yet our understanding of its environmental drivers remains incomplete. We know that cold, dry air (low temperature and low absolute humidity) helps the virus spread, but these factors don't fully explain the complex dynamics of flu outbreaks.

The role of ambient ozone—a common air pollutant and powerful oxidant—has remained largely in the shadows. While some studies in subtropical regions like Hong Kong suggested a link, it wasn't clear if this relationship held true in temperate climates or if the association was truly causal. The challenge lies in the "messiness" of environmental data: seasonality, non-linear interactions, and autocorrelation make it difficult to prove that ozone causes a drop in flu cases rather than just being a bystander.

Methods

To move beyond simple correlation, the authors employed a "totality of evidence" approach, applying three mathematically diverse methods to weekly state-level data from 46 U.S. states (2010–2015):

  1. Convergent Cross Mapping (CCM): A technique from state-space reconstruction theory designed to detect causality in non-linear dynamical systems. It looks for the "signature" of one variable within the time series of another.
  2. PCMCI+: A sophisticated graphical modeling algorithm that identifies causal networks (Directed Acyclic Graphs) while accounting for high-dimensional dependencies and time lags.
  3. Generalized Linear Model (GLM): A standard statistical regression approach used here to quantify the effect size and provide a familiar benchmark for comparison.

By "triangulating" these methods—each with different underlying assumptions and biases—the researchers sought a result that was robust to the choice of mathematical framework.

Datasets

The study utilized several publicly available longitudinal datasets:

  • Influenza Activity: Weekly laboratory-confirmed data from the CDC, specifically a proxy variable ("Flu") calculated as the product of influenza-like illness (ILI) consultations and the proportion of positive specimens.
  • Ozone ($O_3$): Daily maximum 8-hour average concentrations from the Tropospheric Ozone Assessment Report (TOAR).
  • Weather: Temperature and absolute humidity data from the NOAA National Center for Environmental Information (NCEI).
  • Scope: The analysis focused on the "flu season" (October through May) across 46 states, excluding four due to data gaps.

Key Insights and Ideas

  • Consistent Inhibition: All three methods converged on a single truth: ambient ozone has a negative impact on influenza activity. When ozone goes up, flu activity tends to go down about a week later.
  • The "Aha" Mechanism: Why would a pollutant help fight the flu? The authors suggest a two-pronged biological mechanism. First, ozone is virucidal; its oxidizing power can directly inactivate the virus. Second, and perhaps more interestingly, low-level ozone exposure may prime host immunity. Specifically, it triggers the release of Interleukin-33 (IL-33), an "alarmin" that boosts antiviral T-cell responses and helps repair lung tissue.
  • Ozone vs. The Field: In the CCM analysis, the causal signal for ozone was often stronger and more consistent across different states than the signals for temperature or humidity, suggesting it is a primary, not secondary, driver.

Key Results

  • Lagged Effect: The inhibitory effect was most consistently demonstrated at a 1-week lag.
  • Quantitative Impact: According to the GLM results, a one standard deviation increase in ozone concentration (approx. 8.3 ppb) was associated with a significant reduction in logit-transformed influenza activity ($p < 5.9 \times 10^{-5}$).
  • Causal Confirmation: The CCM meta-significance for the nation reached $p = 1.9 \times 10^{-7}$, providing powerful evidence that the relationship is not merely a seasonal coincidence.

Limitations and Future Work

The authors are transparent about several caveats:

  • Surveillance Bias: The study relies on passive surveillance data, which is subject to variations in how people seek healthcare and how states report data.
  • Subtype Blindness: The analysis aggregated all influenza types (A and B). Future research could investigate if ozone affects specific strains (like H1N1 vs. H3N2) differently.
  • Scale: State-level and weekly aggregation might smooth over finer-grained local or daily interactions.
  • The "Pollutant Paradox": While ozone appears to inhibit the flu, it is still a harmful respiratory irritant. The authors emphasize that these findings should inform environmental management and public health protection, not suggest that ozone pollution is "good."

Further molecular and clinical studies are needed to confirm the hypothesized IL-33 immune-priming pathway in human populations.

Summary updated

Abstract

Under long-standing threat of seasonal influenza outbreaks, it remains imperative to understand the drivers of influenza dynamics which can guide mitigation measures. While the role of absolute humidity and temperature is extensively studied, the possibility of ambient ozone (O<sub>3</sub>) as an environmental driver of influenza has received scant attention. Here, using state-level data in the USA during 2010-2015, we examined such research hypothesis. For rigorous causal inference by evidence triangulation, we applied 3 distinct methods for data analysis: Convergent Cross Mapping from state-space reconstruction theory, Peter-Clark-momentary-conditional-independence plus as graphical modeling algorithms, and regression-based Generalised Linear Model. The negative impact of ambient O<sub>3</sub> on influenza activity at 1-week lag is consistently demonstrated by those 3 methods. With O<sub>3</sub> commonly known as air pollutant, the novel findings here on the inhibition effect of O<sub>3</sub> on influenza activity warrant further investigations to inform environmental management and public health protection.
BibTeX
@inproceedings{guo2024ozone,
  author = {Fang Guo and Pei Zhang and Vivian Do and Jakob Runge and Kun Zhang and Zheshen Han and Shenxi Deng and Hongli Lin and Sheikh Taslim Ali and Ruchong Chen and Yuming Guo and Linwei Tian},
  title = {{Ozone as an environmental driver of influenza}},
  booktitle = {Nature Communications},
  year = {2024},
  doi = {10.1038/s41467-024-48199-z},
}
Figure