Flexible reasoning about patient management using multiple models

摘要

This paper reports on the Heart Failure Program, which uses multiple models and multiple reasoning operators to provide patient management information for physicians. The program uses a causal probabilistic knowledge base of pathophysiology for reasoning diagnostically, a quantitative physiologic model for reasoning about the effects of interventions, and a case base for an alternate form of diagnostic reasoning. Using these knowledge bases are reasoning operators to turn patient data into evidence for causal reasoning, using the evidence to assert specific physiologic states, generating a diagnosis or differential from the causal knowledge base or from the case base, using the current diagnostic state to determine what further information would be useful, using the diagnostic state to suggest therapies, predicting the possible effects of therapies, and using the diagnostic hypotheses or the effect predictions to generate graphical explanations for the user. By combining the models and reasoning methods, each potential use of the program benefits because each operator provides conclusions that simplify the task of other operators. The result is a program with uses in many phases of the patient management process.