Article Text
Abstract
Background The Guideline Elements Model (GEM) has been widely used to translate natural language clinical practice guidelines (CPGs) into clinical decision support (CDS) using a highly replicable, guideline-centric approach. A CPG recommendation-to-CDS translation process, which uses GEM-processed content to support an oncology rapid learning system (RLS) prototype, is examined here.
Objectives To develop rules for a breast cancer-specific CDS prototype using GEM-processed guideline content.
Methods We created five breast cancer patient scenarios with expert input from oncologists based on nine published CPGs. Using the Yale Center for Medical Informatics-developed GEM Cutter III editor, we parsed the narrative CPG recommendations into an XML-based, machine-readable format. GEM-processed content was then encoded into a Drools business rule management system to develop an integrated platform prototype for rules, workflow, and event processing. We used meta-tags to create value sets for key components of each recommendation by selecting terms from UMLS vocabularies, including SNOMED CT and LOINC.
Results Forty-five recommendations spanning nine CPGs were processed and converted into Drools rules. We identified 138 decision variables and 91 actions within the selected recommendations. From these, we encoded 148 concepts associated with value set meta-tags and 238 decision rules.
Discussion The level of difficulty required to encode the recommendations was directly related to the specificity, complexity, and decidability of each recommendation; there was significant variability among the recommendations.
Implications for Guideline Developers/Users CPG developers may need new processes in order to optimise recommendations for incorporation into CDS systems.