A Bayesian predictive analytics model for improving long range epidemic forecasting during an infection wave.

Area of Science:

• Epidemiology
• Computational modeling
• Public health

Background:

• Policymakers faced uncertainty in planning Non-Pharmaceutical Interventions (NPIs) during the early COVID-19 pandemic.
• Existing epidemiological models were found to be too reliant on manual inputs and slow to adapt to data changes.

Purpose of the Study:

• To develop and evaluate novel algorithmic models for more accurate and proactive forecasting of epidemic trends.
• To improve local policymaking by providing reliable predictions amidst uncertainty.

Main Methods:

• Developed four new models incorporating daily reported deaths and infections.
• Explicitly addressed missing data, including under-reported cases.
• Modeled delays in test reporting and simulated weekly death forecasts.
• Utilized a lighter model variant for faster forecasting post-initialization.

Main Results:

• Proposed models demonstrated improved forecasting, particularly for long-range predictions and post-peak scenarios.
• Models showed greater proactivity in identifying trend changes and adapting to post-peak data.
• Accounting for under-reported cases significantly enhanced model stability.
• Modeling retroactive data additions had minimal impact on performance.

Conclusions:

• The developed models offer a more robust and adaptive approach to epidemic forecasting compared to traditional methods.
• These enhanced predictive tools can better support public health decision-making during epidemics.
• The models' ability to handle data uncertainties and adapt to trends is crucial for effective NPI planning.