890 Artificial Intelligence Empowers Remote Monitoring of Implantable Loop Recorders

Abstract

Background The implantable loop recorder (ILR) is a subcutaneous device used to diagnose heart rhythm disorders, allowing the detection of any arrhythmic episodes which are related to the onset of patient-reported symptoms. Although the importance of implantable cardiac monitors has been widely recognized, it is also true that due to their subcutaneous nature, they are inclined to detect false-positive episodes which generate several warnings requiring additional checks from the health care personnel. For this reason implanted cardiac devices monitoring has been recently implemented by the use of artificial intelligence (AI), which applies deep learning algorithms for speeding up the telemetry follow-up checks without affecting the accuracy. Considering the increased need of remote monitoring during pandemics, such achievement should reduce the overwork superimposed on healthcare providers, but to date it is not known to what extent this technology confers a quantitative gain. Our main purpose is to demonstrate how AI can enhance remote monitoring of patients with implantable cardiac monitors, reducing false positives and improving the accuracy of information received by cardiologist. Methods we included twenty-six patients who undergoing ILR implantation between May 2019 and April 2022. Mean age was 58 ± 28 years and there was a higher prevalence of male subjects (n = 20; 76,9%). Fifteen patients received ILR implemented by the use of AI, eleven patients received ILR with standard remote monitoring system. We did follow up from April 2022 to September 2022 with dedicated remote monitoring platforms. The main events detected by ILR were atrial fibrillation (AF), pauses > 3 sec, supraventricular and ventricular arrhythmias, patient's symptom. Results among AI-ILR a total of 25 events were recorded and 92% (n= 23) of these were correctly identified, meanwhile 4% (n= 1) event was considered as false positive. AI implementation showed a sensitivity of 93% (95% CI 0,7018 to 0,9966) and specificity of 90% (95% CI 0,5958 to 0,9949) Among the ILR not supported by AI a total of 26 events were registered, in 76,9% (n=20) of these there was a true correspondence between the event reported and the real event, while in 19,1% (n=5) there wasn't correspondence with a sensitivity of 91% (95% CI 0,6461 to 0,9957) and specificity of 64% (95% CI 0,3876 to 0,8366). Based on these data, AI improves the accuracy of diagnosis, reducing the incidence of false positives with an important saving time for health care providers. Conclusions with our analysis we support the hypothesis that AI algorithms can augment diagnostic capabilities and provides an important contribute to clinical decision-making. The implementation of ILR monitoring with AI is still at the beginning, but continuously further investigations will determine the real added value of these algorithm. The wider adoption of AI technology in cardiovascular medicine seems eventually inevitable in the digital world of 21st century healthcare.