Evaluating short-term forecasting of COVID-19 cases among different epidemiological models under a Bayesian framework

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Accurate forecasting of COVID-19 cases is critical for global public health decision-making.
• Short-term daily and weekly projections are essential for informing governments and health sectors.
• Predictive modeling plays a vital role in generating these crucial public health insights.

Purpose of the Study:

• To evaluate and compare the short-term forecasting performance of different predictive models for COVID-19 cases.
• To calibrate stochastic growth models and the susceptible-infectious-removed (SIR) model within a unified Bayesian framework.
• To assess the accuracy and interpretability of these epidemiological models against a benchmark.

Main Methods:

• Utilized a Bayesian framework to calibrate stochastic epidemiological models, including variants of growth models and the SIR model.
• Implemented rolling-origin cross-validation for robust comparison of short-term forecasting capabilities.
• Analyzed data from 20 countries with the highest confirmed COVID-19 cases as of August 22, 2020.

Main Results:

• No single model demonstrated superior performance across all evaluated regions.
• Stochastic epidemiological models consistently outperformed the autoregressive moving average (ARMA) model.
• The evaluated epidemiological models showed improved accuracy and interpretability compared to the ARMA model.

Conclusions:

• Stochastic epidemiological models provide valuable tools for short-term COVID-19 case forecasting.
• The choice of model for COVID-19 prediction should consider regional variations and specific public health needs.
• While no universal best model exists, epidemiological approaches offer advantages over traditional time-series methods for this application.