A modular approach to forecasting COVID-19 hospital bed occupancy

Area of Science:

• Epidemiology
• Public Health Surveillance
• Health Informatics

Background:

• Hospital bed occupancy monitoring was crucial for Australia's COVID-19 surveillance.
• Short-term forecasts of bed occupancy were generated from 2021 to 2023 to inform public health policy.

Purpose of the Study:

• To develop and evaluate a model for forecasting COVID-19 patient ward and intensive care unit (ICU) bed occupancy.
• To assess the performance of 21-day occupancy forecasts across Australian states and territories.

Main Methods:

• A stochastic model simulating patient progression through the hospital system was employed.
• The model was fitted to reported occupancy data using approximate Bayesian inference.
• Forecasts utilized independently generated case incidence data as input, decoupling infection and occupancy modeling.

Main Results:

• Forecasts exhibited downward bias before epidemic peaks and upward bias after peaks.
• Forecast accuracy was highest in more populous states and territories.
• Performance evaluation covered March to September 2022 across all Australian jurisdictions.

Conclusions:

• Weekly COVID-19 hospital burden forecasts informed Australia's national public health response.
• A modular modeling approach allowed independent development and leveraged ensemble case forecasts.
• The forecasting system supported evidence-based decision-making during the pandemic.