Using OMIM (On-line Mendelian Inheritance in Man) as an expert system in medical genetics

Abstract

Expert systems have been used in Medicine for many years, but they are usually highly sophisticated and not well integrated into day-to-day practice. On the other hand, bibliographic databases such as Medline and others are easily accessible and are widely used. We report here the use of OMIM (On-line Mendelian Inheritance in Man), one of these bibliographic databases, as an expert system in Medical Genetics. The description of 93 syndromes was used as search-key and the diagnoses proposed by OMIM were analyzed to determine whether the correct diagnosis was among them. The proposed diagnoses were automatically ranked by OMIM from the most probable (weight = 100) to the least probable (weight = 1). OMIM suggested a total of 1538 +/- 692.2 diagnoses per search. In order to deal with a reasonable number of proposed diagnoses, we only considered the diagnoses with a weight of 50 or more. With this limit, OMIM proposed a mean of 37.0 +/- 24.6 diagnoses per case. The overall accuracy was 76%. A correct answer with a perfect weight of 100 was proposed in 29% of the case. The diagnostic accuracy of OMIM increased linearly when weights lower than 100 were considered. When the rank alone was analyzed, the accuracy of OMIM increased very rapidly from position 1 to 5 with a subsequent almost linear increase. If one only considered the first five proposed diagnoses, the accuracy of OMIM was just above 50%. This study shows that bibliographic databases are not only restricted to the provision of references but could also be used as expert systems and are therefore of great value to medical geneticists.