We have all become familiar with the importance of monitoring the number of COVID-19 cases for setting policies and managing limited health care resources. But how do inaccuracies in tests for the virus impact the curves that describe the epidemic?
Paul teamed up with Drexel University epidemiologists Igor Burstyn and Neal Goldstein to study this question, using publicly available COVID-10 testing data from Alberta and Philadelphia. They develop a Bayesian method, placing priors on disease prevalence and the sensitivity and specificity of the tests. With their approach and with the available data, they glean information on disease prevalence and test sensitivity, and re-calculate epidemic curves along with uncertainty measures of these curves. Their analysis indicates that the data are consistent with the hypothesis that the number of truly infected is being under-estimated. The under-diagnosis is more pronounced when there are both more positive cases and the prevalence of positive tests is higher, i.e., in Philadelphia related to Alberta.
The research appears in this recently posted preprint on medRxiv entitled "Towards reduction in bias in epidemic curves due to outcome misclassification through Bayesian analysis of time-series of laboratory test results: Case study of COVID-19 in Alberta, Canada and Philadelphia, USA” by Igor Burstyn, Neal D. Goldstein and Paul Gustafson.