Diagnostic research in the presence of an incomplete or imperfect reference standard

Abstract

Diagnostic accuracy research is a vital step in the evaluation of new diagnostic technologies. It is the ability of a test to correctly discriminate between patients that have and do not have the target disease. In studies of diagnostic accuracy, results of the tests under study are compared with results of a reference standard applied to the same patients. In this research framework, the reference standard is the best available method to verify the presence or absence of the target disease, and thus provides the final classification of patients into target disease present or absent. This process is known as disease verification. Ideally, the reference standard provides error-free disease classification. In some situations, it is not possible to verify the disease outcome with the preferred reference standard in all patients or any patient at all. Failure to apply the reference standard may result in various types of disease verification problems. Biased and exaggerated estimates of accuracy of a test can lead to inefficiencies in testing in clinical practice, unnecessary costs, and could trigger physicians to making incorrect treatment decisions. This thesis examines the problem of verification bias in studies of diagnostic accuracy. In particular, we aimed to investigate the available methods to alleviate the various problems of verification bias and, more importantly, to improve the methodology and analysis of primary diagnostic accuracy studies and diagnostic meta-analyses in the presence of various forms of verification bias.