Short-Range Forecasting of COVID-19 During Early Onset at County, Health District, and State Geographic Levels Using...

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• COVID-19 forecasting relies on historical data, with effectiveness varying by geographic scope and public behavior.
• Numerous platforms provide COVID-19 forecasts, but evidence is needed for short-term method selection at various administrative levels.

Purpose of the Study:

• To evaluate the performance of seven forecasting methods for cumulative COVID-19 case counts.
• To identify forecasting model assumptions that minimize error in short-term COVID-19 case growth predictions.
• To inform public policy and public understanding of health statistics.

Main Methods:

• Compared seven forecasting methods: Naïve, Holt-Winters (HW), growth rate, moving average (MA), autoregressive (AR), ARMA, and ARIMA.
• Utilized historical county-level COVID-19 case data from Virginia (March 7 - April 22, 2020).
• Evaluated 1-, 3-, and 7-day ahead forecasts using Median Absolute Error (MdAE) and Median Absolute Percentage Error (MdAPE).

Main Results:

• The MA method with a 3-day look-back achieved the lowest MdAE, significantly outperforming most alternatives.
• Forecasting methods assuming stationary means of prior case counts were more effective than those assuming non-stationary means.
• Significant differences in MdAPE were observed across different geographic levels.

Conclusions:

• The MA method is effective for short-range (1-7 day) COVID-19 cumulative case count forecasting at county, health district, and state levels.
• Exponential growth models were less suitable for early pandemic stages with public awareness.
• Geographic resolution is a critical factor in selecting appropriate COVID-19 forecasting methods.