56-77: Adapting the threshold for atrial fibrillation detection in insertable cardiac monitors based on evidence of irregular sinus rhythm

Abstract

Background: Frequent premature atrial contractions and sick sinus syndrome are primary causes of inappropriate atrial fibrillation (AF) detection in insertable cardiac monitors (ICM). An algorithm was developed to reduce inappropriate AF detection based on adapting the threshold for detection in the presence of irregular RR intervals and p-wave evidence. Methods: The AF detection algorithm in Reveal LINQ ICM looks for evidence of AF based on differences in the pattern of RR intervals over a 2-minute period. The p-wave evidence based algorithm (P-SENSE) reduces evidence for AF detection if p-waves are detected. The adaptive P-SENSE enhancement uses the presence of p-wave evidence during periods of RR irregularity as evidence of the presence of sick sinus or ectopy to adaptively increase the threshold for AF detection. The algorithm was developed using Holter data from the XPECT study which collected two leads of surface ECG and continuously uplinked ICM ECG over a 46 hour period. ICM detections were compared with Holter annotations to compute episode and duration detection performance. Results: Valid Holter recordings were analyzed from the first 56 patients in the XPECT study with a total follow-up duration of 2168 hours (39 hours per patient). True AF was observed in 16 patients, yielding 89 true AF episodes ≥2 minutes and 201 hours of AF. In the nominal (and aggressive) mode of operation, the algorithm correctly identified 97.9% (97.8%) of total AF duration and 99.5% (99.6%) of total sinus or non-AF rhythm duration. The algorithm detected 89% (89%) of all AF episodes ≥2 minutes, and 60% (74%) of detected episodes had AF in the nominal mode of operation. The adaptive P-SENSE algorithm in nominal (or aggressive) mode was able to reduce false detects by 76% (87%) compared to an algorithm without P-SENSE and 56% (63%) compared to the nominal (aggressive) P-SENSE without any loss in true episode detection (Figure).