Abstract:
|
Current advances in technology provide less invasive or less expensive diagnostic tools for identifying disease status. When a diagnostic test is evaluated against an invasive or expensive gold standard test, one often finds that not all patients undergo the gold standard test. The naive sensitivity and specificity estimates computed by using only the patients with verified disease are often biased. This bias is called a verification bias. Many authors have examined the consequences of verification bias and proposed bias correction methods. Some of these methods are based on assumption of independence of disease and test result conditional on the verification status. When this assumption is not valid, possible non-ignorable missing data mechanism needs to be considered. However, this often involves uncheckable assumptions and requires sensitivity analysis. In this talk, we present a global sensitivity analysis for assessing the performance of a diagnostic test. Bounds and region of all possible pairs of sensitivity, and specificity estimates are given in a "bow tie" plot. The methodology is illustrated with clinical examples.
|