TL;DR
This paper tackles the challenge of estimating the international importation risk of the SARS-CoV-2 Omicron BQ.1 variant, particularly from Western Africa, using incomplete mobility data. They developed a deep neural network model that leverages available mobility and flight data to predict passenger flow and, consequently, the risk of virus introduction to other countries, highlighting France and Spain as high-risk destinations.
What Problem Does This Paper Address?
The emergence of new SARS-CoV-2 variants, like Omicron BQ.1, poses a significant public health threat, especially when they exhibit immune escape properties and can cause reinfections even in highly vaccinated populations. The rapid spread of such variants is heavily influenced by international travel. Estimating the risk of a variant being imported into a country is crucial for informing public health policies and implementing timely control measures. However, obtaining comprehensive and accurate global mobility data is a major challenge. This paper specifically addresses the problem of estimating the importation risk of Omicron BQ.1 from Western Africa to other non-African countries, where mobility data is often incomplete, particularly due to the limitations of platforms like Facebook's mobility reports which only capture flows above a certain user threshold.
Methods
The core of this paper's methodology lies in its novel approach to estimating missing mobility data and then using that to calculate importation risk.
- Deep Neural Network for Mobility Estimation: The authors developed a deep neural network (DNN) model to predict the proportion of occupied seats on flights between countries. This DNN takes as input time, and country-specific information such as latitude, longitude, GDP per capita, and government policy indexes (stringency, containment, health, and vaccination policies). This approach aims to overcome the limitations of incomplete mobility data (e.g., from Facebook) by learning complex relationships between these factors and actual travel patterns.
- Estimating Traveler Numbers: Once the proportion of occupied seats is predicted, this is combined with flight data (number of available seats) to estimate the number of travelers between countries.
- Importation Risk Calculation: The estimated traveler numbers are then used to calculate the importation risk. This involves estimating the number of asymptomatic and presymptomatic cases in Western Africa (based on reported cases and variant characteristics) and then multiplying this by the estimated traveler flow to a destination country. A Poisson process model is used to estimate the probability of at least one infected case being imported.
Datasets
The study utilized the following datasets:
- Facebook Mobility Data: Provided by Data For Good at Meta, this dataset offers insights into user movement between countries but has a threshold for reporting (only flows with at least 1,000 daily users are captured).
- OpenSky Flight Data: This dataset provided information on available flight seats between countries.
- GISAID: Used for Omicron genome sequence data, specifically for identifying the emergence of BQ.1 in Nigeria.
- World Population Prospects: Used for population estimates.
- Oxford COVID-19 Government Response Tracker: Provided indexes on government policies.
- Other publicly available data: Including GDP per capita, latitude/longitude, and Facebook user data by country.
The paper does not introduce a new benchmark dataset but rather focuses on a novel method to process and augment existing, albeit incomplete, datasets.
Key Insights and Ideas
- DNNs as a powerful tool for inferring mobility: The study demonstrates that deep learning can effectively infer missing mobility data by learning complex relationships between various socio-economic and policy factors and actual travel patterns, significantly improving upon simpler assumptions.
- Western Africa as a potential export hub for BQ.1: The research highlights the significant risk of Omicron BQ.1 being imported from Western Africa to other continents due to substantial people flow.
- Geographic disparities in risk: The model identifies specific countries, particularly in Western Europe (France, Spain) and North America (Canada, US), as having a higher importation risk, likely due to direct flight connections.
- Early arrival correlation with high import probability: The analysis suggests a logarithmic relationship between import probability and the arrival time of infected cases, implying that higher risk countries might see earlier introductions.
Key Results
- The developed DNN model achieved an R² score of 0.67 when predicting the proportion of occupied aircraft seats, a significant improvement over a previous study's fixed-ratio assumption (R² of 0.11).
- France and Spain were identified as having the highest importation risk for Omicron BQ.1 from Western Africa during the study period.
- Other countries like Canada, the United States, Italy, the United Kingdom, Germany, Portugal, Belgium, Turkey, and the United Arab Emirates also showed notable importation risks.
- The study provides a visualization of importation risk across the globe, indicating that Western Europe and Northern America faced higher risks compared to other regions.
Limitations and Future Work
- Assumptions on travel behavior: The study acknowledges that assumptions about symptomatic infected individuals not traveling and the prevalence of BQ.1 in Western Africa mirroring Nigeria's might influence the accuracy of the risk estimation.
- Incomplete data: Despite the DNN's ability to handle incomplete data, the inherent limitations of the input datasets (e.g., Facebook's reporting threshold) can still impact the precision of the estimations.
- Indirect flights: The paper notes that ignoring indirect flights could lead to an underestimation of export risk.
The authors suggest extending their approach to other types of viruses and developing a platform for policymakers to monitor importation risks. They also plan to further refine their model to provide more granular insights.
Summary updated