|
Abstract:
|
Currently, the FDA uses an estimation approach called the patient infected status algorithm (PISA) to evaluate nucleic acid amplification tests for detecting Chlamydia trachomatis (CT) and Neisseria gonorrhea. In recent years, researchers have demonstrated that this overly simplistic approach for estimating test sensitivity and specificity parameters of a diagnostic test is biased, arbitrary and is not necessarily better than using a single imperfect test. In this work, I will calculate the PISA-based estimates of two non-nucleic acid amplification tests (non-NAATs) from a study conducted in the north-western part of the U.S. and compared these estimates with the package insert PISA-based estimates of two NAATs for detecting CT. These PISA-based results demonstrate that culture has the "highest" sensitivity and the GenProbe test has the "highest" specificity, indicating the currently FDA recommended CT evaluation algorithm, PISA, can result in biased and highly unusual inference. Using simulations, I will further demonstrate that even a coin test in the presence of conditional dependence can have a PISA-based sensitivity and specificity estimates which are seriously biased.
|
