Medical diagnostic decision rules based on mutually dependent diagnostic factors.
There are numerous situations in which clinicians are not free to choose informative and independent factors (tests) in the solution of their diagnostic or prognostic problems. It is vital to use a combination of several mutually dependent factors in such cases. The paper develops a Bayesian approach to the creation of diagnostic and prognostic rules based on mutually dependent factors. Following the work of Philosophov (Philosophov, Medical Decision Making, 15, 264-275, 1995), we distinguish between true factor values and factor estimates-the result of factor measurement by means of an imperfect medical instrument (technology). A constructive solution of the problem is achieved by introducing the concept of clinical variants of a disease--a set of possible forms of disease, each of which is described by a specific combination of prognostic and other relevant factors. The appropriate use of mutually dependent factors can yield gains in decision efficiency, even in the simplest case of two binary factors. The gain achieved can increase with the number of factors used and variety of factor values. An example based on real clinical material illustrates the approach and the gains it can achieve.