Predictive performance of international COVID-19 mortality forecasting models.

Area of Science:

• Epidemiology
• Public Health
• Data Science

Background:

• Accurate forecasting of COVID-19 mortality is crucial for effective pandemic response.
• Decision-makers require reliable information on the predictive performance of available models.

Purpose of the Study:

• To screen and evaluate the predictive performance of global COVID-19 mortality forecasting models.
• To assess model accuracy based on extrapolation time, geographic region, and estimation month.
• To evaluate the prediction accuracy of the timing of peak daily mortality.

Main Methods:

• Screened 386 public COVID-19 forecasting models, selecting 7 global, public, date-versioned models.
• Examined predictive performance for mortality using median absolute percent error (MAPE).
• Assessed prediction accuracy for the timing of peak daily mortality.

Main Results:

• Models released in October demonstrated a global median absolute percent error (MAPE) of 7-13% at six weeks.
• Median absolute error for peak timing prediction increased from 8 days (1 week) to 29 days (8 weeks).
• Performance was consistent for both first and subsequent mortality peaks.

Conclusions:

• Global COVID-19 mortality models exhibit robust predictive performance, even with complex influencing factors.
• Forecasting peak mortality timing becomes less accurate with longer extrapolation periods.
• A public framework and codebase are available for ongoing model comparison and performance evaluation.