CI-452770-1 USE OF ARTIFICIAL INTELLIGENCE ALGORITHMS TO REDUCE TRANSMISSIONS IN IMPLANTABLE LOOP RECORDERS

Abstract

Implantable loop recorders (ILRs) are frequently used for arrhythmia diagnosis and management. Remote monitoring (RM) facilitates earlier response to alerts. However, this has resulted in significant increase in clinic workloads due to growing alert burden. A large component of this excess workload has related to inaccuracies related to artefact or over-sensing. The use of artificial intelligence (AI) may help to reduce and streamline workflow. To assess the impact of the AccuRhythmTM in a large independent remote monitoring dataset to determine the impact on the number of transmissions and the how the implementation of AI has improved the burden of downloads from Medtronic LINQ II. The AccuRhythmTM AI algorithm applies deep learning algorithms to the LINQ II data addressing false alerts. For this study, the PaceMateTM RM system was utilized. This system is a partially automated, vendor-neutral RM software system. The system presents RM data acquired from all vendors in a standardized format and includes all RM alerts transmitted by CIED patients undergoing RM via PaceMate LiveTM. All patients, who transmitted at least one episode during the one-month period before (January 2022) and after (April 2022) implementation of the AccuRhythmTM AI algorithm were evaluated. We included 661 patients with a mean age of 64.9±15.4 years who transmitted in both January and April. There was a statistically significant reduction in alerts transmitted per patient from January (ranging from 0 to 63 alerts per patient; mean 1.4, median 0, IQR=1) compared to after activation of the AI algorithm in April (range 0 to 36 alerts per patient; mean 0.98, median 0, IRQ=1; p<0.001). (Figure) This represented a 20.4% reduction in alerts per patient with the AI algorithm. In addition to the reduction in alerts transmitted, there was also a significant reduction in the number of episodes per patient attached to each transmission from January (range 0 to 211 episodes per patient; mean 4.5, median 0, IQR=3) compared to April (range 0 to 110 episodes per patient; mean 3.9, median 0, IQR=1; p<0.001), (Figure) representing a 39.7% reduction in episodes per patient with the AI algorithm. The implantation of an AI algorithm for LINQ II devices, significantly reduced alert burden by 20% and episode burden by 40% in a large, unselected group of patients. The implementation of AI algorithms have an essential role in facilitating the reduction of clinic workloads due to RM.