MS

thesisHeart failure is a common disorder, which is easy to diagnose, but remains potentially lethal. Effective therapy exists but is underused. Recognizing this problem, the Agency of Health Care Policy and Research published guidelines for heart failure management in June 1994. Practice guidelines have been proposed as a way to reduce inappropriate variation in patient care, and thus as a way to improve the quality of care. Computerization with real time feedback is seen as one of the ways to implement guidelines more effectively. This thesis discusses the development and evaluation of a prototype system called HeartMan, a set of computerized guidelines for heat failure management. Two evaluation using historical patient data were performed. The first experiment was designed to evaluate the accuracy and completeness of the knowledge. Data on 10 heart failure patients (50 visits) were abstracted and the messages generated by HeartMan for the last 5 patients were formally reviewed by a heart failure expert. Ninety percent of the 177 messages were considered appropriate. The rules that produced neutral or inappropriate messages were revised. New rules were created after assessing the completeness of the knowledge base. The second experiment was designed to evaluate the accuracy of the revised knowledge base and to assess the potential impact of HeartMan on patient care. Ten new cases (50 visits) from Cardiology and Internal Medicine clinics were studied. Of the 350 messages generated, over 90% were considered appropriate, 82% could have averted a severe quality problem, and 37% were also discrepant with the treating physician's actions. The reason of discrepancy between Heartman and the treating physician included error of omission and differences in management styles. This latter group of messages offered the greatest opportunity for improvement in clinical practice by using the system. A small number (2%) of inappropriate messages could have produced a severe quality problem and were also discrepant with the physician. These experiments support the conclusions that a rule-based knowledge base can adequately model complex domains, and computerized guidelines for heart failure management might have a potential positive impact on the quality of patient care