AIM: The aim of this study is to evaluate Bayesian Networks (BN) as Clinical Decision Support Systems (CDSS) in medical fields, in general, and in Emergency, specifically. Moreover, we intend to study its usefulness in comparison with the traditional systems applied to diagnosis, prognosis and therapeutics.
METHODS: For this state of the art, we selected articles 234 articles using three online databases, with proper and specific queries – ISI Web of Knowledge, Scopus and Medline. We have two or three reviewers (to solve divergences when necessary) that were responsible for the selection of the articles according to the following inclusion criteria: the included articles must be applied to diagnosis, prognosis or therapeutic related to Bayes’ theorem; papers must provide details so that the study can be analyzed; they must be written in English. The articles would be excluded if it was a meta-analysis or a review or if it wasn’t applied to humans. The same article could be classified in more than one medical field. After selection, articles were classified according to the medical field and healthcare domain they referred to.
RESULTS: Ignoring 32 wrongly included articles, we found 128 articles regarding diagnosis (58%), 41 mentioning prognostic (18%) and 54 about therapeutic (24%). Although we found articles related to 33 different medical areas and the majority of the papers were set in the area of Oncology (24) we followed our aim presented before. We also found 9 articles related to diagnosis in Medical Emergency. We have excluded one of them because the full article was unavailable, leaving us with 8 articles. While analyzing the selected 8 articles, we were able to identify the most relevant information and to put it into a proper database. Then we proceed the study of the selected variables. We concluded that 4 articles were nonrandomized trials without control group and with one unique exception, they all considered BN to be a successful CDSS. Although two of the articles did not mention BN to be a successful CDSS, the accuracy of this method ranged from medium to as high as 99%, as in the other papers.
Key-words: Bayesian Network, CDSS, Diagnosis, Emergency