Obtaining prevalence estimates of COVID-19: A model to inform decision-making

Objectives The primary aim was to evaluate whether randomly sampling and testing a set number of individuals for active or past COVID-19 while adjusting for misclassification error captures a simulated prevalence. The secondary aim was to quantify the impact of misclassification error bias on publicly reported case data in Maryland. Methods Using a stratified random sampling approach, 50,000 individuals were selected from a simulated Maryland population to estimate the prevalence of active and past COVID-19. Data from the 2014-2018 and 2018 American Community Surveys were used. The simulated prevalence was 0.5% and 1.0% for active and past COVID-19, respectively. Bayesian models, informed by published validity estimates, were used to account for misclassification error when estimating the prevalence of active and past COVID-19. Results Failure to account for misclassification error overestimated the simulated prevalence for active and past COVID-19. Adjustment for misclassification error decreased the point estimate for active and past COVID-19 prevalence by 55% and 29%, respectively. Adjustment for sampling method and misclassification error only captured the simulated past COVID-19 prevalence. The simulated active COVID-19 prevalence was only captured when set to 0.7% and above. Adjustment for misclassification error for publicly reported Maryland data increased the estimated average daily cases by 8%. Conclusions Random sampling and testing of COVID-19 is needed but must be accompanied by adjustment for misclassification error to avoid over- or underestimating the prevalence. This approach bolsters disease control efforts. Implementing random testing for active COVID-19 may be best in a smaller geographic area with highly prevalent cases.