Predicting COVID-19 positivity and hospitalization with multi-scale graph neural networks

Area of Science:

• Epidemiology
• Data Science
• Public Health

Background:

• The COVID-19 pandemic presented significant challenges to healthcare systems globally.
• Accurate prediction of disease spread, including positivity and hospitalization rates, is vital for effective healthcare planning and resource allocation.

Purpose of the Study:

• To develop and validate a novel multi-scale graph neural network for predicting COVID-19 positivity and hospitalization.
• To leverage high-resolution spatio-temporal data and integrate population mobility and vaccination rates for improved prediction accuracy.

Main Methods:

• A multi-scale graph neural network model was developed, integrating fine-scale geographical zone data.
• The model employed message passing to capture interactions between areas, utilizing population mobility and other relevant features.
• Performance was evaluated against baseline and existing deep learning models.

Main Results:

• The proposed model demonstrated superior performance compared to baseline and deep learning models in predicting both COVID-19 positivity and hospitalization.
• Low prediction errors were achieved for both tasks, highlighting the model's accuracy and reliability.
• The model's effectiveness in predicting hospitalizations is particularly significant due to the critical role of hospitals during the pandemic.

Conclusions:

• This work presents a novel approach to spatio-temporal prediction of COVID-19 cases and hospitalizations using high-resolution data.
• The integration of multi-scale data, mobility, and vaccination rates offers a significant advancement in disease forecasting.
• The developed method has the potential to enhance future healthcare planning and resource management during pandemics.