Soft computing considerations on diagnostic accuracy of multiple tests
S. Shibata · 2002
In order to find the optimum combination of multiple tests, this paper considers how diagnostic accuracy changes with the number of tests and with five kinds of combination methods: (A) sequential tests, (B) sequential tests after changing the order in (A), (C) simultaneous tests using the believe-the-negative rule, (D) simultaneous tests using the believe-the-positive rule, and (E) simultaneous tests using the believe-all-positives-and-all-negatives rule. The diagnostic accuracy is defined by sensitivity (/spl alpha/) and specificity (/spl beta/). An overall 2/spl times/2 table for multiple tests was inducted from the 2/spl times/2 tables for each test. The overall diagnostic accuracy was calculated from the overall 2/spl times/2 table and was expressed by a generalized formula. As it was proved that the combination methods A, B and C are "equivalent", the five methods could be summarized into three methods: (1) for A, B and C, /spl beta/ increased but /spl alpha/ decreased with an increasing number of subjects n; (2) for D, /spl alpha/ increased but /spl beta/ decreased with n; (3) for E, /spl alpha/ and /spl beta/ increased with n, although a problem exists concerning how to judge the remaining subjects who are between positive-in-all-tests and negative-in-all-tests. Soft computing considerations ware adopted for this problem.