Rosy Tsopra1, Alain Venot1, Catherine Duclos1. 1. INSERM, U1142, LIMICS, F-75006, Paris; Université Paris 13, Sorbonne Paris Cité, F-93000, Bobigny; Sorbonne Universités, Univ Paris 06, F-75006, Paris, France.
Abstract
UNLABELLED: Clinical practice guidelines (CPGs) are documents giving recommendations based on expert reasoning, weighing up the pros and cons of treatments on the basis of the available evidence. We propose a new approach to the construction of clinical decision support systems (CDSS), making use of the evidence-based medical reasoning used by experts in CPGs. In this study, we determined whether this approach could retrieve the recommendations for antibiotic prescription for empirical treatment in primary care. METHODS: We manually extracted, from CPGs, all the properties of antibiotics underlying recommendations for their prescription or non-prescription. We then used these properties to establish an algorithm in the form of a sequence of conditions, leading to a list of recommended antibiotics. The optimal sequence was determined by studying, for each sequence, the degree of similarity between the list of antibiotics recommended in CPGs and the list obtained with the algorithm. RESULTS: 12 antibiotic properties were used in the form of conditions in an algorithm. For 95% of clinical situations, 10 sequences retrieved the recommended treatment. DISCUSSION: This algorithm could be used in a CDSS for antibiotic treatment and would be useful for experts drawing up CPGs.
UNLABELLED: Clinical practice guidelines (CPGs) are documents giving recommendations based on expert reasoning, weighing up the pros and cons of treatments on the basis of the available evidence. We propose a new approach to the construction of clinical decision support systems (CDSS), making use of the evidence-based medical reasoning used by experts in CPGs. In this study, we determined whether this approach could retrieve the recommendations for antibiotic prescription for empirical treatment in primary care. METHODS: We manually extracted, from CPGs, all the properties of antibiotics underlying recommendations for their prescription or non-prescription. We then used these properties to establish an algorithm in the form of a sequence of conditions, leading to a list of recommended antibiotics. The optimal sequence was determined by studying, for each sequence, the degree of similarity between the list of antibiotics recommended in CPGs and the list obtained with the algorithm. RESULTS: 12 antibiotic properties were used in the form of conditions in an algorithm. For 95% of clinical situations, 10 sequences retrieved the recommended treatment. DISCUSSION: This algorithm could be used in a CDSS for antibiotic treatment and would be useful for experts drawing up CPGs.