A Bayesian Approach to Improving Spatial Estimates After Accounting for Misclassification Bias in Surveillance Data for COVID-19 in Philadelphia, PA

Surveillance data obtained by public health agencies for COVID-19 are likely inaccurate. Using a Bayesian approach, we adjusted for misclassification to improve spatial estimation of COVID-19 in Philadelphia, PA at the ZIP code level.  We modeled true prevalence as a function of area deprivation index and spatial random effects at the ZIP code level. There were 111,497 documented tests as of June 10, 2020 in Philadelphia, among whom 23,941 (21%) were classified by the laboratory as positive for SARS-CoV-2, translating to an observed prevalence of 1.5%. After accounting for bias in the surveillance data, the posterior citywide true prevalence was 2.8% (95% credible interval: 2.7%, 3.0%). Overall the posterior surveillance sensitivity and specificity were 58.9% (95% credible interval: 41.5%, 75.9%) and 99.8% (95% credible interval: 99.7%, 100%), respectively. Underreporting tends to understate discrepancies in burden for more affected areas, potentially leading to bias in setting priorities.