Abstract:
|
The improvement on estimation precision of repeated measurements is well understood in many statistical applications, but not all. Diagnostic tests, which are often evaluated for their accuracy both in detecting a condition when one is present and in not detecting one when there is none. Improving accuracy in one detection condition often comes at the cost of the other. The connection between sensitivity and specificity of a diagnostic test makes the effect of repeated testing, using either the same test or different tests, less than clear.
In this study, we evaluate the value in repeating diagnostic testing to improve accuracy. Specific focus has been placed on impact of the decision rules used in determining the final diagnosis, e.g. whether a positive is returned if any tests are positive or only when all tests are positive, or anything in between. Diagnostic accuracy is evaluated by weighted loss functions for false negatives (FN) and false positives (FP). Loss ratios of FN/FP = K and 1/K, where K > 1, are examined. Optimal decision rules-those which result in the smallest average losses with respect to the two classes of loss ratios-are sought using simulated scenarios.
|