Extrapolating Sentinel Surveillance Information to Estimate National COVID Hospital Admission Rates: A Bayesian Modeling Approach

Abstract

The COVID-19-Associated Hospitalization Surveillance Network (COVID-NET) was established in March 2020 to monitor trends in hospitalizations associated with SARS-CoV-2 infection. COVID-NET is a geographically diverse population-based surveillance system for laboratory-confirmed COVID-19-associated hospitalizations with a combined catchment area covering approximately 10% of the US population. Data collected in COVID-NET includes monthly counts of hospitalizations for persons with confirmed SARS-CoV-2 infection who reside within the defined catchment area. A Bayesian modeling approach is proposed to estimate US national COVID-associated hospital admission rates based on information reported in the COVID-NET system. A key component of the approach is the ability to estimate uncertainty resulting from extrapolation of hospitalization rates observed within COVID-NET to the US population. In addition, the proposed model enables estimation of other contributors to uncertainty including temporal dependence among reported COVID-NET admission counts, the impact of unmeasured site-specific factors, and the frequency and accuracy of testing for SARS-CoV-2 infection. Based on the proposed model, an estimated 6.3 million (95% uncertainty interval (UI) 5.4–7.3 million) COVID-19-associated hospital admissions occurred in the United States from September 2020 through December 2023. Between April 2020 and December 2023, model-based monthly admission rate estimates ranged from a minimum of 1 per 10,000 population (95% UI 0.7–1.2) in June of 2023 to a highest monthly level of 16 per 10,000 (95% UI 13–19) in January 2022.