Development and implementation of a guideline-based decision support system for etiological diagnosis of Sustained Monomorphic Ventricular Tachycardia (SMVT)

Abstract Background Guideline (GL)-based decision support systems (DSSs) play a significant role in supporting physicians in their decision-making processes by providing structured clinical guidance. While these systems have shown significant success in enhancing practitioner performance and patient outcomes in general practice, their application in the context of ventricular arrhythmias remains limited. Moreover, the clinical etiological diagnosis of ventricular arrhythmias presents various challenges for healthcare professionals, such as adherence to updated diagnostic criteria and understanding of complex arrhythmia mechanisms. Purpose In this study, we aim to design, implement, and evaluate a guideline-based decision support systems for etiological diagnosis of sustained monomorphic ventricular tachycardia (SMVT), which could potentially address critical challenges associated with the disease diagnosis and management. Methods To comprehensively capture the semantics embedded in clinical guidelines, we first developed a dedicated knowledge graph tailored to the clinical decision-making. This involved extracting entities and relationships pertinent to key decision procedures from ESC guidelines for ventricular arrhythmias, clinical textbooks or literatures. Subsequently, we translated instances within the knowledge graph into archetypes within openEHR, amplifying the interoperability of clinical information. To enhance the interpretability of the DSS, we established executable diagnostic and treatment processes to offer visual representation of clinical steps in diagnosis and treatment. Ultimately, we constructed a decision support engine to implement the GL-DSS, and its effectiveness was evaluated with anonymous real-world data. Results The proposed GL-DSS was found to be quite feasible and could provide etiological diagnosis for SMVT. The knowledge base for the constructed GL-DDS included more than 2000 entities and relations, which played a vital role in representing disease pathogenesis. The proposed system can map the clinical entities to a set of openEHR archetypes, thus mitigate the interoperability issues of the hospital information systems and CDSS data integration. Also, this module improved system usability and ensured that guidelines could be easily translated and updated into decision support tools. Real-word medical records from 20 patient were collected for model evaluation, the results showed that the GL-DSS covered most of the entities and relations between diseases and diagnostic evidence recorded in clinical texts and achieved a diagnosis accuracy of 80%. Conclusions The results demonstrated that our proposed GL-DSS could completely and accurately capture the diagnostic information under the concrete EHR data. Thus, the implement decision support system could have the potential to improve clinical decision-making, patient care, and outcomes in this specialized medical domain.