Affiliations: Institute for High Performance Computing and Networking (ICAR), National Research Council of Italy (CNR), Naples, Italy
Note: [] Corresponding author. Massimo Esposito and Domenico Maisto, Institute for High Performance Computing and Networking (ICAR), National Research Council of Italy (CNR), Via P. Castellino 111 Naples, Italy. E-mails: [email protected] (Massimo Esposito), [email protected] (Domenico Maisto).
Note: [] Corresponding author. Massimo Esposito and Domenico Maisto, Institute for High Performance Computing and Networking (ICAR), National Research Council of Italy (CNR), Via P. Castellino 111 Naples, Italy. E-mails: [email protected] (Massimo Esposito), [email protected] (Domenico Maisto).
Abstract: Clinical practice guidelines are expected to promote more consistent, effective, and efficient medical practices and improve health outcomes, especially if provided in the form of clinical decision support. However, most clinical guidelines, especially when expressed in the form of condition-action recommendations, embody different kinds of structural errors that compromise their practical value. With this respect, this paper presents a novel method for verifying the reliability of condition-action clinical recommendations encoded in the form of fuzzy rules, with the final aim of determining inconsistency, redundancy and incompleteness anomalies in a very simple and understandable fashion. The method is based on general definitions of inconsistency, redundancy and incompleteness for fuzzy clinical rules in terms of similarity between antecedents and consequents, bringing them near the imprecise character of fuzzy decision support systems. A key issue relies on the formalization of fuzzy degrees for these anomalies that can be simply interpreted by the final users as measurements suggesting the modifications to be performed to the clinical rules in order to eliminate or mitigate the existing undesired effects. The method has been profitably assessed on two sample sets of clinical rules: the first one identified from the relevant clinical literature and the second one extracted automatically by machine learning techniques from a widely known clinical database. The achieved results prove simplicity and usability of our method in detecting structural anomalies and in adjusting a rule base by exploiting information carried out during the verification phase.
Keywords: Structural verification, fuzzy decision support systems, clinical guidelines
