Forecasting the COVID-19 Pandemic: Lessons learned and future directions

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Accurate forecasting of Coronavirus Disease 2019 (COVID-19) infection and mortality rates is vital for public health policy and resource allocation.
• A comprehensive understanding of the relative performance of various COVID-19 modeling methodologies is lacking.

Approach:

• Introduced the non-mechanistic MIT-LCP forecasting model.
• Assessed and compared MIT-LCP against seven other mechanistic and non-mechanistic models using data from November 2020 to April 2021.
• Evaluated model performance within the United States' Centers for Disease Control and Prevention (CDC) Forecast Hub.

Key Points:

• Mechanistic models perform better for long-term forecasts when recent data is less informative.
• Non-mechanistic models are more effective for short-term forecasts using recent, representative data.
• Both modeling approaches have distinct strengths depending on the forecasting horizon and data availability.

Conclusions:

• Understanding the optimal use cases for different forecasting models enhances pandemic preparedness and management.
• The study provides insights into selecting appropriate models for various forecasting scenarios.
• This research contributes to more reliable pandemic response strategies through informed model selection.