Evaluating Expert Critiques

A number of workers in Artificial Intelligence have argued that, for some medical domains, critiquing the users decisions is an appropriate approach. Others have stressed the importance of integrating decision-support systems with existing information systems. Little research, however, has been directed to the issues of what consists a relevant critique and whether such a critique could be generated using data obtained from automated medical records. We therefore performed a study in which a general practitioner (GP) was asked to provide us with the computer-based medical records of five patients with hypertension. A printout of these medical records was submitted to an internist who had a recognized interest and experience in the treatment of hypertension. The internist was asked to comment on the treatment of the hypertension as documented in the medical records. Subsequently the comments of the internist were submitted to a panel of three GPs; these GPs were asked to judge the relevance of the comments. Finally the comments of the internist were shared with the GP who had treated the patient.