|
Abstract:
|
Group testing is widely used as a means to reduce costs in high volume disease screening applications. The initial stage of the testing involves forming pools by mixing a fixed number of individual specimens, such as urine, blood, swab, etc. If disease traits are rare, group testing is more efficient but can increase the number of false negative test results, a testing error commonly referred to as ``dilution effects.'' The deleterious effects of dilution on statistical inferences include underestimation of the individual-level disease probabilities, misleading covariate effects, etc. In this article, we propose a new method to account for dilution effects in group testing regression models. We express the pool-level sensitivity in terms of a (parametric or nonparametric) submodel that can be easily constructed using cumulative distribution functions. Unlike existing works, our approach does not require any information about the underlying biomarker distributions. We develop a general Bayesian framework to model data from most group testing protocols. We show using simulation and real data applications that our models provide reliable inferences in the presence of dilution effects.
|
