Computer-based Diagnostic Support Systems

To the Editor:—I read with great interest the study by Elstein et al. on the use of computer based diagnostic support systems.¹ I felt, however, that one of the restrictions they placed on data provided as input to computer programs may have unrealistically limited program effectiveness. So-called definitive tests were withheld from the programs. I believe their definition of definitive is too broad. Consider their example of a high white blood cell level in a patient with appendicitis. While I think giving the computer program the results of a biopsy or a nuclear medicine scan would be too much, it is highly unrealistic to expect a computer program to be effective without basic laboratory tests that would be ordered as a matter of course for a patient with this specific presentation.

As the authors state, the real use of computer-based diagnostic programs is to bring to the practitioner's attention diagnoses not otherwise thought of. An example from my practice might serve to illustrate the problem I have identified. An otherwise healthy young woman I recently saw complained of fatigue. I was surprised to find she had an “unexplainably” low serum potassium level. I initially did not need the help of a diagnotic program to figure out which basic laboratory tests to order, such as serum pH and urine electrolytes. The program did become useful, however, in suggesting diagnoses that didn't occur to me after I had the initial laboratory results. Both diagnostic programs I tried (QMR and Iliad) suggested what ultimately was the correct diagnosis, renal tubular acidosis. Had I initially withheld the “definitive” test(s), such as serum pH, the program would have given me a useless, long, and unsorted list of conditions. For computer-diagnostic programs to be useful to practicing physicians, studies must use these programs in a more realistic manner.