Improved AF rhythm discrimination using QRS morphology

Background Implantable cardiac monitors (ICMs) are valuable tools for long-term ECG monitoring, especially for AF. However, AF detection algorithms based solely on R-R interval variability are prone to a high number of false positives due to ectopic beats and noise. Objective The purpose of this analysis is to report on the simulated performance of an AF algorithm when using QRS morphology information. Methods Twelve-lead ECG data from 487 patients from 6 different clinical studies (collected in both in-clinic and ambulatory settings), including 35 AF patients, were used to evaluate the AF algorithm. The ECG vector V2–V3 was used to approximate the implanted ICM electrode configuration. The algorithm made a classification of AF or Non-AF for every 2-min window. Morphology Evaluation: The V2–V3 signal morphology of each detected beat in the 2-min window was compared against a QRS template built during sinus rhythm. A match score was assigned to each beat based on the similarity to the template. The percentage of mismatched beats per 2-min window (for example, due to noise or ectopic beats) was used to determine the degree of R-R interval variability not associated with AF. The user-programmed levels for ectopy rejection and AF sensitivity will determine the thresholds used to reject potential AF rhythms. Results A total of 44,716 2-min windows were evaluated. Morphology information was available in 228 windows of true AF along with 51 windows of false positives for AF. Morphology assessment was not conducted for 4 patients in persistent AF throughout the data recording (i.e. no sinus rhythm). Using the morphology evaluation, the results ranged from 41.2% reduction in false positives (21/51 windows) with no impact on sensitivity, to 56.9% (29/51 windows) reduction in false positives with a cost of missing 5 true AF windows (97.8% sensitivity) depending on how the ectopy rejection level is tuned. Conclusions Assessment of EGM morphology in an ambulatory AF detection algorithm can meaningfully reduce false positives with minimal impact to sensitivity.