Biomedical Informatics 4th Edition by Edward Shortliffe, James Cimino

Exercise 5

In one evaluation study, the decision-support system ONCOCIN providedadvice concerning cancer therapythat was approved by experts in only79 % of cases (Hickam etal. 1985) .In another study, the HyperCriticCDS system for the management ofhypertension offered the same com-ments that were generated by a panelof experts in only 45 % of cases (Vander Lei, et al. 1991) . Such system per-formance is fairly typical for computerprograms that suggest patient therapy.Do you believe that this performance isadequate for a computational tool thatis designed to help physicians to makedecisions regarding patient care? Whatproblems might CDS systems encoun-ter as their developers attempt to makethe systems more comprehensive in theadvice that they offer? Why might it bemore difficult for computer systems tooffer acceptable recommendations forpatient therapy than seems to be thecase for diagnosis? What safeguards,if any, would you suggest to ensure theproper use of such systems? Would yoube willing to visit a particular hysicianif you knew in advance that she madedecisions regarding treatment that wereapproved by expert colleagues lessthan 80 % of the time? If you wouldnot, what level of performance wouldyou consider adequate? Justify youranswers.