Accuracy of implantable loop recorders: multicenter, multidevice comparison

Background: Implantable loop recorders (ILRs) frequently generate false-positive alert transmissions, increasing clinical burden. Objectives: This study sought to evaluate and compare arrhythmia detection accuracy among currently used ILRs: Medtronic (MDT) LINQ II with AccuRhythm AI, Boston Scientific (BSX) LUX-Dx with a Dual-Stage Algorithm, Biotronik (BIO) Biomonitor III with RhythmCheck, and Abbott (ABT) Confirm Rx (SharpSense). Methods: A multicenter, multivendor database of 437,351 electrocardiogram-verified episodes from 6,756 patients (January 2022 to March 2023, across 25 centers) was used. A random sample of 1,140 patients (∼283 per device) was selected based on a prospectively determined power analysis. All alerts were independently adjudicated by trained physicians. Results: The study cohort (median age: 69.5 years [IQR: 60.1-77.0]; 39% women) generated 25,826 total alerts, with a median of 5 alerts (IQR: 2-14) per patient. Indications for ILR implant included cryptogenic stroke (21.0%), suspected atrial fibrillation (AF) (11.3%), AF management (23.6%), syncope (23.2%), palpitations (14.9%), ventricular tachycardia (2.1%), and unknown (3.8%). Alerts for atrial tachycardia/AF were most frequent (BIO: 70.8%; MDT: 68.0%; ABT: 55.7%; BSX: 35.7%). After adjudication, BSX had the highest positive predictive values (PPVs), followed by ABT and MDT, which had similar and moderate PPVs, whereas BIO had the lowest (BSX: 0.73; MDT: 0.56; ABT: 0.51; BIO: 0.23). False-positive atrial tachycardia/AF alerts with BIO devices were due to ectopy (75.8%), combined ectopy with oversensing/noise (5.3%), isolated oversensing (5.2%), sinus arrhythmias (3.9%), indeterminate reason (3.7%), noise (3.6%), and bradycardia (2.5%). For bradycardia detection, BSX demonstrated the highest PPV (0.96), whereas MDT had the lowest PPV (0.36; primarily due to undersensing). Pause detection accuracy was lowest for ABT (PPV: 0.01; 79.8% due to undersensing) and highest for BSX (PPV: 0.70). Tachycardia alerts had moderate PPVs for BIO (0.62) and BSX (0.61), with substantially lower PPVs for MDT (0.29) and ABT (0.27). Conclusions: ILRs demonstrate variable accuracy in arrhythmia detection, with a substantial burden of false-positive alerts across all vendors despite artificial intelligence–based and other algorithmic enhancements. Further improvement of alert algorithms to reduce clinic burdens is needed.