Computerized CHA2DS2Vasc classification in remote atrial fibrillation alerts: an ontology-based approach
Abstract
Introduction: Atrial fibrillation (AF) notifications in remote monitoring represent a high medical burden in terms of alert management. In the AKENATON project (Automated Knowledge Extraction from medical records iN Association with a Telecardiolgy Observation Network), we developed a method based on the integration of clinical information with data transmitted by implantable cardiac devices for clinical decision support purposes.
Methods: Patient (pt) data were extracted from available textual or structured documents by an information extraction tool, and integrated with data extracted from device memories into a specific data model. An ontology (knowledge model) offering reasoning capabilities based on this data model was used to classify AF alerts. The thrombo-embolic risk (Low (L)/Medium (M)/High (H)/Critical (C)) based on CHA2DS2VASc score, medication and AF episode duration (Figure 1) was estimated by the system for 60 pts and compared to manual review by a domain expert.
Results: At the level of the CHA2DS2VASc calculation (8 criteria per pt), 446 out of the 480 criteria to be estimated by the system were correct, resulting in 58 (97\%) pts with correct CHA2DS2VASc score classification (0/1/2+). Medication was adequately evaluated in 57 (95\%) pts. The final alert classification (11 L, 13 M, 31 H and 5 C) was correct for all pts minus one case of risk over-estimation (C instead of H).
Conclusion: This work proves the ability of the system to better classify alerts and improve alert response based on clinical data. Such method could be of great use for remote monitoring and extended to other use cases and domains.