Some Tricks and Tools for Intelligent Use of Multiple Tests

Abstract

The problem of devising optimal diagnostic strategies is fascinating. Ultimately, the conflicting interests of various parties lend it the flavor of a political game. Meanwhile, medical facts and probability calculus can be exploited in an exact manner. To avoid combinatorial explosion, however, clever heuristic, non-exact algorithms like those that go into chess—playing programs must sometimes be devised. It is sobering to note, first, that in many test problems of high combinatorial complexity the empirical data describing the joint distribution of test outcomes are often too incomplete to warrant sophisticated mathematics. Second, the mathematically optimal diagnostic strategy is invariably surrounded by many close contenders, of which even a simple program that leaves many stones unturned is likely to find one. Third, optimality criteria, such as utility maximization, are themselves based on soft data.