A computer-based model for providing empiric antibiotic therapy in a hospital.

A new model for computer-based medical decision making, called QID for Quick Infectious Disease, is described. It is based on the principles of decision analysis, using probabilities and utilities. In this study, QID was applied to the problem of recognition and empiric treatment of hospital-acquired infections. The model makes use of three knowledge bases: (1) the knowledge needed for generating a differential diagnosis from disease manifestations and risk factors; (2) estimates of expected morbidity and mortality for each disease, both with and without optimal treatment; and (3) tables of sensitivities of each organism-specific disease to each antibiotic, as well as cost and risk associated with use of each antibiotic. Using information about each patient, QID determines the expected utility of each antibiotic in terms of “Good Days of Life Saved†(GDS) and suggests the best antibiotic(s). A 4-year effort was required to build the QID software and knowledge bases. The knowledge required for QID to diagnose and suggest treatment for 85 different pathogens in 11 sites of infection, was provided by infectious disease experts in knowledge engineering sessions. In an experiment to measure the decision-support utility of QID, 40 university-based physicians were presented with abstracts of four nosocomial cases and asked to prescribe antibiotics before, and then after, the aid of QID. Physicians' antibiotic prescriptions significantly (p < .001) improved, after seeing the suggestions made by QID, as scored by two infectious disease specialists. In 53% of the experimental cases there was improvement, as judged by at least one of the two specialists. The average improvement was 7.8% before the judges knew the isolated pathogen, and 9% after. For the most difficult case (meningitis), the improvement in scores averaged 20%. In the cases in which the physicians decided to change antibiotics, GDS was the principal influencing factor. Case selection proved to be important in determining the outcome of the evaluation of QID. The average improvement in physicians' choices of antibiotic after using QID was inversely proportional to the prevalence of the disease as well as the mean scores before the physicians were exposed to QID's recommendations.