Pasche Emilie, Teodoro Douglas, Gobeill Julien, Ruch Patrick, Lovis Christian
Medical Informatics Service, University Hospitals of Geneva and University of Geneva, 1211 Geneva 14, Switzerland.
Stud Health Technol Inform. 2009;150:569-73.
We aim at proposing a rule generation approach to automatically acquire structured rules that can be used in decision support systems for drug prescription. We apply a question-answering engine to answer specific information requests. The rule generation is seen as an equation problem, where the factors are known items of the rule (e.g., an infectious disease, caused by a given bacteria) and solutions are answered by the engine (e.g., some antibiotics). A top precision of 0.64 is reported, which means, for about two third of the knowledge rules of the benchmark, one of the recommended antibiotic was automatically acquired by the rule generation method. These results suggest that a significant fraction of the medical knowledge can be obtained by such an automatic text mining approach.
