[Utilization of expert systems in psychiatry].

Are expert systems liable to be used as consultants in psychiatry? Most expert systems deal with an over-restricted part of psychiatry and cannot be a real help in everyday care. Moreover, most of them are not actually validated (the comparison between the system's and the expert's conclusions in a few cases is not enough). Another problem is that they reflect the uncertainties of nosographic problems. Validation of such systems needs the careful checking of the logical structure of the underlying nosography, the fitness of the structure's knowledge base and the fitness of the inference engine. Moreover, the naïve use of the system by untrained clinicians is the best means of validation since it provides real life proof of the ability of expert systems to make diagnoses in unselected cases where the need for a common diagnostic reference is clear (for example, epidemiologic, psychopharmacological ornosographic research). Some of the best known expert systems in the field of psychiatry are reviewed and another expert system, Adinfer, is presented. Developed since 1982, Adinfer is a forward-tracking level O system (in its simplified version for micro-computers). The knowledge base is a translation of the DSM-III-R into production rules. The program has been included in several software packages and used in many clinical studies, both among psychiatrists and physicians. The program has been validated with 1,141 unselected cases, and with 47 physicians: an 83% agreement rate was found between the system's and the physician's diagnoses, taking into account that the clinicians were asked to give their conclusions according to their usual nosography.(ABSTRACT TRUNCATED AT 250 WORDS)