Generating simple classification rules to predict local surges in COVID-19 hospitalizations

Area of Science:

• Epidemiology
• Public Health
• Computational Biology

Background:

• COVID-19 surges risk overwhelming US hospital capacity due to low vaccination rates, new variants, and relaxed mitigation.
• Current predictive models focus nationally, leaving local policymakers without timely tools for surge warnings.
• Local hospitalization trajectories vary based on demographics, vaccination, and behavior.

Purpose of the Study:

• Develop a framework for simple classification rules to predict local COVID-19 hospitalization surges.
• Provide early warnings to local decision-makers about potential hospital capacity exceedance within 4-8 weeks.
• Create a tool usable without complex numerical computations.

Main Methods:

• Utilized a simulation model of SARS-CoV-2 transmission and COVID-19 hospitalizations.
• Trained classification decision trees using real-time hospital occupancy, COVID-19 admissions, and genomic surveillance data.
• Developed rules robust to data changes and future uncertainties.

Main Results:

• Classification rules demonstrated accuracy, sensitivity, and specificity ≥80% in predicting local surges.
• Performance was validated across numerous simulated scenarios capturing future COVID-19 uncertainties.
• Rules are simple, visual, and easy for local decision-makers to implement.

Conclusions:

• The developed framework provides effective, user-friendly classification rules for predicting local COVID-19 hospitalization surges.
• These rules can aid local policymakers in proactive resource management and mitigation strategy implementation.
• The approach addresses the critical need for localized, real-time public health surveillance tools.