Performance of a cloud-based AI solution for reducing false positive alerts in implantable cardiac monitors detected episodes
N Varma, M Wadhwa, A Lazarus, G Faedda, P Mabo, E CrespinAbstract
Introduction
Implantable Cardiac Monitors (ICM) provide long-term arrhythmia diagnostics but generate high False Positive (FP) event rates, despite incorporating manufacturer proprietary Artificial Intelligence (AI) algorithms.
Purpose
A new cloud-based AI solution was developed to mitigate FP burden while maintaining patient safety. Clinical performance was tested across a set of compatible ICMs, stratified by the presence (AI models) or absence (Non-AI models) of a manufacturer-proprietary AI pre-filtering algorithm.
Methods
A retrospective clinical evaluation was conducted on 2,659 ICM detected episodes (excluding patient-activated episodes) from 1,710 patients (age 67.5 ± 15.0 y, male 27%, 62.2% from US) implanted with ICMs from Medtronic (43.6%), Biotronik (22.4%), Abbott (18.0%), and Boston Scientific (15.9%), followed at 11 US and 19 EU centers via an universal monitoring platform. To ensure a diverse dataset, the selection method included at least one episode per device for each ICM-detected event type (bradycardia, pause, atrial and ventricular arrhythmias). Coprimary endpoints were defined for safety (sensitivity) and performance (specificity). PPV was a secondary endpoint. 95% confidence intervals (CI) were computed with a bootstrap estimation. All episodes were reviewed by an independent adjudication committee.
Results
Among 2,659 episodes, 2,073 (78.0%) were considered interpretable (excluding "Uninterpretable" and "Uncertain SVT") by the Adjudication Committee. The AI algorithm demonstrated a high sensitivity of 98.3% in AI models and 94.3% in Non-AI models. A disparity in specificity of 61.6% for AI-models vs 75.6% for Non-AI models was observed, balanced by a consistent Positive Predictive Value (PPV) across both groups of 73.9% and 74.4%, respectively.
Conclusion
This new cloud-based AI solution demonstrates the capacity to significantly improve clinical efficiency through reducing false positive burden while maintaining a high safety profile.