COVID-19 forecasting using new viral variants and vaccination effectiveness models.

Area of Science:

• Epidemiology
• Infectious Disease Modeling
• Machine Learning Applications

Background:

• Rising COVID-19 cases in highly vaccinated regions necessitate booster strategies.
• Forecasting COVID-19 incidence is challenging due to variant evolution and co-factors.
• Existing models struggle with the variability of infections and vaccination impacts.

Purpose of the Study:

• To develop a machine learning model for forecasting COVID-19 daily positive cases (DPC).
• To assess the impact of vaccination effectiveness, waning immunity, and viral variant infectivity on transmission.
• To explore cross-country data transferability for vaccination effect projection.

Main Methods:

• Long short-term memory (LSTM) network for time-series forecasting of DPC, serious cases, and deaths.
• Integration of vaccination data (including waning effects) from Israel and Japan.
• Inclusion of public mobility and social media interaction data to model behavioral changes.
• Analysis of viral variant infectivity and population effectiveness of vaccination.

Main Results:

• LSTM model accurately estimated vaccination effectiveness and waning protection parameters.
• Vaccination effectiveness against the Delta variant was estimated at 0.24 (2nd dose, 5 months) and 0.95 (3rd dose, 2 weeks).
• Population-level vaccine effectiveness, considering waning immunity, is critical for COVID-19 prevention, with estimated thresholds in Tel Aviv (0.3) and Japan (0.4).

Conclusions:

• Vaccination effectiveness and viral variant infectivity are key predictors for future COVID-19 case forecasting.
• Population-level vaccine effectiveness, accounting for waning protection, is more influential than vaccination rates alone.
• The study demonstrates a viable method for projecting vaccination effects using international data.