Enhancing the prediction of hospitalization from a COVID-19 agent-based model: A Bayesian method for model parameter...

Area of Science:

• Epidemiology
• Computational Modeling
• Public Health

Background:

• Agent-based models (ABMs) are crucial for estimating hospital bed demand during pandemics.
• Existing COVID-19 ABMs often lack key hospitalization predictors like comorbidities and testing status.
• This omission limits model interpretability for stakeholders.

Purpose of the Study:

• To introduce a novel application of Bayes' theorem for enhanced COVID-19 ABM parameterization.
• To incorporate comorbidities and testing status into hospitalization probability estimates.
• To improve the accuracy and transparency of ABM predictions for decision-making.

Main Methods:

• Applied Bayes' theorem using aggregated hospital data.
• Updated input parameters for a North Carolina COVID-19 ABM.
• Incorporated discrete factors: comorbidities and COVID-19 testing status.

Main Results:

• Demonstrated sensitivity of ABM predictions to input data variations.
• Highlighted enhanced interpretability and accuracy of the proposed method.
• Validated the approach using real-world data from North Carolina.

Conclusions:

• A straightforward Bayesian approach can significantly improve COVID-19 ABMs.
• Incorporating comorbidities and testing status enhances model transparency and utility.
• This method supports better decision-making, even with limited data during pandemics.