Title: Estimation of log-odds ratio from group testing data using a Bayesian approach
Authors: Viktor Skorniakov - Vilnius University (Lithuania) [presenting]
Ugne Cizikoviene - Vilnius University (Lithuania)
Remigijus Leipus - Vilnius University (Lithuania)
Abstract: Group testing (GT) is an important testing strategy with many applications spanning broad spectra of areas with a clinical setting (CS) among all the rest. In the case of CS, GT reduces to the testing of pooled specimens obtained by pooling untested individual specimen. A typical example occurs when, instead of testing $k$ individual blood samples, one pools them, performs a single test, and finds out whether an infection is present in the pool. In case of a rare disease, the pool often tests negatively resulting in significant test cost savings. In case pool tests positively, retesting takes place if one wishes to identify all infected. This is called an identification task. However, if one wishes only to estimate disease prevalence (DP), subsequent retesting is not needed and leads to an estimation task (ET). There is a substantial body of literature indicating that GT is effective for ET when DP is low. We investigate the effectiveness of Bayesian GT based ET by conducting a pilot simulation study devoted to the case of two populations when the estimation of the odds ratio is a goal and test is imperfect.