Novel PVC detection algorithm in an implantable cardiac monitor for longitudinal risk monitoring

Abstract

Abstract Background Guideline therapy is recommended in patients with frequent premature ventricular contractions (PVCs). Longitudinal monitoring of the PVC burden can effectively guide treatment options, stratify patient risk, and manage HF progression. Implantable cardiac monitors (ICMs) may provide a suitable long-term diagnostic tool to monitor PVC burden and guide clinical therapy. Objective Quantify the performance of a novel QRS morphology and timing-based PVC detection algorithm as implemented for an ICM. Methods The algorithm utilizes self-adapting timing-based ectopy detection and dynamic measures of QRS morphology to identify unifocal and multifocal PVCs. The training dataset consisted of 200 sECG episodes recorded in implanted BIOMONITOR (BM) devices, collected from 48 patients. The algorithm was then evaluated on 227 sixty-second duration sECG episodes from 79 BM patients in the CERTITUDE registry. Each beat was adjudicated as normal, PVC, or PAC by at least two clinical experts. In a total of 22,433 beats, 756 true PVCs were assessed (3.4% PVC prevalence). Overall, patient averaged, and generalized estimating equation (GEE) estimates for sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated, as well as patient-specific PVC burden (PVCs/Total Beats). Results Patient-averaged PVC detection sensitivity was 73.8%, with a PPV of 89.2%, specificity of 99.6%, and NPV of 99.1%. Using a GEE model to adjust for multiple episodes per patient, the algorithm achieved detection sensitivity of 66.6±9.8% (95% CI) and PPV of 86.6±6.5%, with specificity of 99.5±0.4% and NPV of 99.0±0.5%. The main reasons for missed PVC detections were morphological similarities to normal beats, insufficient prematurity, or lack of a compensatory pause. Additionally, the algorithm faithfully represented the PVC burden, with a median difference of -0.24% between the PVC burden calculated by the detection algorithm and the adjudicated burden. Algorithm burden was highly correlated with the true PVC burden for each patient, with a Pearson correlation coefficient of 0.80 (p < 0.0001). Conclusions The PVC detection algorithm was highly specific, achieving a specificity of 99.6% and a sensitivity of 73.8%, and accurately representing the overall PVC burden for each patient. This PVC detection algorithm would provide a valuable clinical diagnostic marker to monitor PVC burden trends to guide therapy as well as assess the longitudinal risk of HF. Measurement of PVC burden combined with the 5.5-year longevity of the BIOMONITOR ICM could provide an innovative approach to optimize clinical management of patients.Example PVCs detected by the algorithm