Speaker
Description
Asymptomatic SARS-CoV-2 infections have contributed substantially to disease transmission during the COVID-19 pandemic. Estimating their frequency, denoted $\psi$, is important for understanding SARS-CoV-2 epidemiology and optimizing. Current estimates of $\psi$ are inaccurate. Two-part serosurveys are used to estimate $\psi$. They employ assays to detect infection and then a questionnaire to ascertain symptoms. Of the individuals who test positive, the fraction declaring symptoms is taken as an estimate of $\psi$. Infection tests, however, are not perfect. Based on their sensitivity and specificity, they yield false positives and negatives. Further, the symptoms elicited by COVID-19 are not unique. They are also elicited by a host of other conditions, including influenza and common cold. Thus, serosurveys may misclassify individuals experiencing symptoms from other conditions as symptomatic for COVID-19, underestimating $\psi$. No formalisms have been proposed to correct for these confounding effects, despite the hundreds of serosurveys reported during the pandemic. We developed such a formalism. We obtained an analytical expression that employs data from serosurveys and yields an accurate estimate of $\psi$. We verified the formalism against synthetic datasets. Applying it to 50 COVID-19 serosurveys, we found that $\psi$ was substantially larger (median ~60%) than previously reported (~40%). Further, our formalism showed that $\psi$ became more consistent across surveys within nations and better associated with age. Our findings help better understand COVID-19 epidemiology and inform efforts to unravel origins of asymptomatic infections.
