Estimating SARS-CoV-2 seroprevalence and epidemiological parameters with uncertainty from serological surveys

Daniel B Larremore; Bailey K Fosdick; Kate M Bubar; Sam Zhang; Stephen M Kissler; C. Jessica E. Metcalf; Caroline Buckee; Yonatan Grad

is the (which was not peer-reviewed) The copyright holder for this preprint Figure S3 : Uncertainty in serological data produces uncertainty in estimates of epidemic peak height and timing, even when the test has perfect sensitivity and specificity. Serological test outcomes for (A) n = 100 tests and (B) n = 1000 tests produce are shown as bar graphs for four tests with sensitivity and specificity values as indicated. Serological test samples were not generated stochastically but instead according to expectation to highlight how sensitivity and specificity affect inference. Posterior seroprevalence estimates for (C) n = 100 and (D) n = 1000 scenarios reveal that Bayesian estimate place posteriors over the correct values (15%) but with uncertainty that depends on n (compare C to D) and on test characteristics (compare peak heights of yellow and purple to blue and orange). Samples from the seroprevalence posterior produce a distribution of epidemic curves for scenarios of 25% and 50% social distancing (see Methods), leading to uncertainty in (E) height of epidemic peak and (F) timing of epidemic peak. Uncertainty is mitigated but not eliminated in the n = 1000 scenario, just as uncertainty is mitigated but not eliminated using a perfect serological test. Boxplots reflect 100 samples from SEIR dynamimcs; whiskers span 1.5×IQR, boxes span central quartile, lines indicate medians, and outliers not shown. See Methods for SEIR simulation details and parameters.