Interpretable AI-driven causal inference to uncover the time-varying effects of PM2.5 and public health interventions on COVID-19 infection rates
Abstract Although COVID-19 appears to be better controlled since its initial outbreak in 2020, it continues to threaten citizens in different communities due to the unpredictability of new strains. The global viral pandemic has resulted in over 700 million infections and 7 million deaths worldwide,...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer Nature
2024-12-01
|
| Series: | Humanities & Social Sciences Communications |
| Online Access: | https://doi.org/10.1057/s41599-024-04202-y |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Although COVID-19 appears to be better controlled since its initial outbreak in 2020, it continues to threaten citizens in different communities due to the unpredictability of new strains. The global viral pandemic has resulted in over 700 million infections and 7 million deaths worldwide, with 22 million cases occurring in the United Kingdom (UK). Emerging evidence has suggested that outdoor PM2.5 pollutants can significantly contribute to COVID-19 infection. However, the time-varying effects of outdoor PM2.5 pollutants on COVID-19 infection rates, particularly in the context of public health interventions, remain poorly understood. This study addresses this knowledge gap by developing a novel AI-driven Bayesian causal deep learning framework to investigate the time-varying causal impacts of PM2.5 concentrations and public health interventions on COVID-19 infection rates in the UK. The proposed framework is designed to identify the time-varying causal relationships between outdoor PM2.5 pollution, key public health intervention measures, and infection rates, while addressing confounding biases and non-linearity in observational temporal-spatial data. It capitalizes on an encoder-decoder architecture for causal inference, where the encoder captures the time-varying causal relationships using a graph neural network, and the decoder provides time-series prediction based on the identified causal structures using a recurrent neural network. Evaluation results demonstrate that the proposed framework outperforms all statistical and deep learning baselines in predicting infection rates. The key findings based on causal effect estimations suggest that short-term outdoor PM2.5 pollution significantly contributed to infection rates, particularly during the early phase. School closure was most effective in early waves, while public transport closure became critical in later stages. These findings offer new insights for public health policymaking. Early-stage interventions to reduce air pollution and enhance indoor ventilation can be crucial for effective pandemic preparedness. Moreover, adaptive public health policies that evolve based on the pandemic phases, such as transitioning from school closures to transport restrictions, can optimize infection control efforts. Beyond COVID-19, the proposed data-driven causal inference and interpretability techniques can be applied to other infectious disease outbreaks and environmental health challenges, providing an interpretable and transferable framework to facilitate evidence-based policymaking. |
|---|---|
| ISSN: | 2662-9992 |