MOAI: A methodology for evaluating the impact of indoor airflow in the transmission of COVID-19

Axel Oehmichen, Florian Guitton, Cedric Wahl, Bertrand Foing, Damian Tziamtzis, Yike GuoarXiv 2021 · OpenReview

Abstract

Epidemiology models play a key role in understanding and responding to the COVID-19 pandemic. In order to build those models, scientists need to understand contributing factors and their relative importance. A large strand of literature has identified the importance of airflow to mitigate droplets and far-field aerosol transmission risks. However, the specific factors contributing to higher or lower contamination in various settings have not been clearly defined and quantified. As part of the MOAI project (https://moaiapp.com), we are developing a privacy-preserving test and trace app to enable infection cluster investigators to get in touch with patients without having to know their identity. This approach allows involving users in the fight against the pandemic by contributing additional information in the form of anonymous research questionnaires. We first describe how the questionnaire was designed, and the synthetic data was generated based on a review we carried out on the latest available literature. We then present a model to evaluate the risk exposition of a user for a given setting. We finally propose a temporal addition to the model to evaluate the risk exposure over time for a given user.

Introduction

Quantifying COVID-19 transmission risk in indoor environments remains a significant challenge. While enclosed spaces are known hotspots, the specific impact of ventilation and airflow is often overlooked or oversimplified. Current epidemiological models typically fall into two extremes: Well-Mixed Room (WMR) models, which are computationally efficient but physically unrealistic, and Computational Fluid Dynamics (CFD), which are highly accurate but prohibitively expensive for large-scale use.

The MOAI framework addresses this "missing middle" by combining crowdsourced environmental data with machine learning to estimate infection risk. Crucially, it does so through a decentralized, privacy-first architecture that utilizes Trusted Execution Environments (TEEs) to protect user data.

The MOAI Architecture

The core of MOAI is the Secretarium platform, which ensures data privacy by design. By utilizing TEEs (such as Intel SGX), the system guarantees that data remains encrypted not just at rest and in transit, but also during computation. Even the developers cannot access raw user data.

High-level architecture of the MOAI framework

As shown in Figure 1, the system facilitates a secure loop where the general public provides environmental data via check-ins, which the MOAI smart contract processes to provide insights to contact tracers without revealing individual identities.

Secure Communication and Identity

To maintain anonymity while ensuring data integrity, MOAI employs a sophisticated handshake protocol between the mobile application and the backend enclaves. This process involves the generation of unique ECDSA identity keys and ephemeral ECDH keys for each connection, ensuring that even the service providers cannot link specific data points to individual users.

Secure handshake protocol between app and TEE

Methodology

Crowdsourced Environmental Proxies

Rather than relying on expensive sensors, MOAI uses user-reported "proxies" for complex physics. Through simple questionnaires, users provide data on ventilation (e.g., open windows), mask usage, and occupancy. These qualitative observations serve as effective stand-ins for fluid dynamics parameters in the risk model.

Machine Learning and Temporal Risk

The researchers evaluated multiple classifiers, including Logistic Regression and Light Gradient Boosting Machine (LightGBM), to predict exposure risk based on 22 categorical fields. To account for the cumulative nature of viral exposure, a temporal risk model was introduced to calculate an individual's risk over time, applying a weight decay factor ($e^{-\lambda\delta}$) to reflect the diminishing relevance of older exposures relative to the virus's incubation period.

Results and Discussion

The LightGBM model emerged as a top performer, achieving an Accuracy of 0.697 and an AUC of 0.755 on synthetic datasets of up to 1 million records. While disease transmission is inherently stochastic, these results indicate a strong predictive signal. Furthermore, the system demonstrated high scalability, with training times for large datasets remaining under five minutes, making it viable for national-scale deployment.

Limitations

The current model relies on synthetic data generated via Hidden Markov Models (HMM) informed by UK NHS statistics. Future work will focus on validating these models with real-world, anonymized datasets and addressing potential vulnerabilities such as QR code fraud, which remains an open challenge for venue-based tracing.

Conclusion

MOAI demonstrates that it is possible to build a scalable, privacy-preserving system for indoor risk assessment. By leveraging TEEs and environmental proxies, the framework provides a viable path forward for data-driven pandemic response that respects individual privacy.

Summary updated

Abstract

Epidemiology models play a key role in understanding and responding to the COVID-19 pandemic. In order to build those models, scientists need to understand contributing factors and their relative importance. A large strand of literature has identified the importance of airflow to mitigate droplets and far-field aerosol transmission risks. However, the specific factors contributing to higher or lower contamination in various settings have not been clearly defined and quantified. As part of the MOAI project (https://moaiapp.com), we are developing a privacy-preserving test and trace app to enable infection cluster investigators to get in touch with patients without having to know their identity. This approach allows involving users in the fight against the pandemic by contributing additional information in the form of anonymous research questionnaires. We first describe how the questionnaire was designed, and the synthetic data was generated based on a review we carried out on the latest available literature. We then present a model to evaluate the risk exposition of a user for a given setting. We finally propose a temporal addition to the model to evaluate the risk exposure over time for a given user.
BibTeX
@misc{oehmichen2021moai,
  author = {Axel Oehmichen and Florian Guitton and Cedric Wahl and Bertrand Foing and Damian Tziamtzis and Yike Guo},
  title = {{MOAI: A methodology for evaluating the impact of indoor airflow in the transmission of COVID-19}},
  year = {2021},
  eprint = {2103.17096},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2103.17096},
}
Figure