Abstract:
|
We examine whether the contribution of a test to the accuracy of a sequence of tests depends on the test order or test characteristics. Assuming a fixed sequence structure (Believe the Positive or Believe the Negative) we measure the contribution of each test in a particular ordering to the probabilities of true and false positive for the entire sequence of tests. We weight this ordering not only by misclassification costs and prevalence, but also by testing cost. The testing cost of a sequence is related to the number of tests required to diagnose a subject. A particular test and therefore a particular sequence may be evaluated by examining the trade-off between testing cost and contribution to accuracy of the sequence. These evaluations are made at the optimal threshold settings for each sequence. At times, the threshold values for a particular test are so extreme that few, if any, subjects are classified by that particular test. For such scenarios, we remove the test and consider that particular sequence invalid. Simulation results demonstrate our findings and suggest the factors that affect how tests should be ordered in a sequence so as to maximize diagnostic accuracy.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.