A fuzzy framework for encoding uncertainty in clinical decision-making

Abstract

In recent decades, technological advances coupled with research efforts have made possible to develop very complex Decision Support Systems (DSSs) able to exhibit highly sophisticated reasoning capabilities in order to improve clinical decision-making, and, thus, promote more efficient care practices. One of the most significant factors influencing, and in particular limiting, the adoption of clinical DSSs is represented by the modality of representation and computerization of clinical guidelines in form of patient-specific recommendations. Until now, many knowledge representation formalisms have been developed, mainly focused on time-oriented guidelines. However, they can generate an unrealistic over-simplification of reality, since they are not able to completely handle uncertainty and imprecision typically affecting clinical guidelines. In this respect, this paper proposes a novel fuzzy framework expressly thought for building guideline-based DSSs, by efficiently modelling and handling the peculiarities of clinical knowledge affected by uncertainty and imprecision and encoded in the form of guidelines. This framework has been devised with the aim of: (i) offering a set of patterns for easily inserting and editing clinical recommendations belonging to a guideline as a group of one or more fuzzy rules expressing positive evidence and one fuzzy ELSE rule including negative evidence; (ii) defining a set of Fuzzy Guideline Systems (FGSs), one for each guideline encoded, characterized by ad-hoc configurations for the mathematical operators necessary to evaluate rules and generate the outcome expected; (iii) implementing a multi-level inference scheme able to treat different FGSs as a whole and efficiently enable their interconnection, i.e. the chaining among the groups of fuzzy rules belonging to each FGS; (iv) exposing a set of graphical facilities for guiding the definition of fuzzy rules to be embedded into a clinical DSS and enabling their automatic encoding and execution by using an XML-based machine executable language. A usability evaluation has been performed, showing a good satisfaction of medical users with respect to the framework implemented, and, thus, proving both its feasibility and usefulness.