Cross-vendor performance of ventricular fibrillation detection by implantable cardioverter-defibrillators: insights from a patient-signal injection benchmark model
M Strik, S Ploux, R Dubois, R Tixier, L Van Krimpen, A Thiyagarajah, J Duchateau, P BordacharAbstract
Aims
Reliable sensing and detection of ventricular fibrillation (VF) are fundamental to the life-saving efficacy of implantable cardioverter-defibrillators (ICDs). Although oversensing has been extensively investigated, undersensing of true VF remains largely unexplored because it is rarely observable in clinical practice. We developed a standardized patient-signal injection benchmark to compare VF detection performance across contemporary ICDs from five manufacturers.
Methods
25 VF signals were recorded from endocardial catheters during ablation procedures for ventricular arrhythmias. The first VF episode per patient (≥10 s, unprocessed) was exported from the Bard EP system and injected into ICDs from Abbott, Biotronik, Boston Scientific, Medtronic, and MicroPort using a National Instruments CompactDAQ with NI-9269 analog output and a DF-4 bipolar connector. Each signal was replayed at three amplitudes (100%, 50%, 10%) under two programming schemes: a comparable VF-zone cutoff (185–188 bpm; ABB 187, BIO 188, BSC 185, MDT 188, MP 185) and the recommended settings (ABB 230, BIO 231, BSC 200/250, MDT 188, MP 230). Measured parameters were: (i) VF detection (yes/no), (ii) time-to-diagnosis (TTD, corrected for nominal counter duration), (iii) undersensed events (US and US/s), and (iv) long cycles (LC and LC/s). Data were analysed using linear mixed-effects models with patient as a random factor and Holm-adjusted pairwise vendor comparisons.
Results
Across 25 human VF recordings (mean VF cycle length 172 ± 26 ms), 750 standardized signal injections were performed, All vendors detected 100% of VF at full gain. At 10% gain, detection diverged: MicroPort and Medtronic maintained 96.8%, Biotronik 93.2%, Boston Scientific 96.7%, and Abbott 78.1%. The adjusted time to diagnosis (TTD), corrected for nominal detection window, differed significantly between vendors (p < 0.001). At 100% gain, MicroPort achieved the shortest corrected TTD (25 ± 120 ms), followed by Medtronic (123 ± 120 ms), Abbott (187 ± 120 ms), Biotronik (275 ± 120 ms), and Boston Scientific (400 ± 120 ms). The ranking remained consistent across lower gains and programming modes. MicroPort also showed the fewest undersensed (US/s = 0.04 ± 0.04) and long cycles (LC/s = 0.03 ± 0.03), whereas Boston Scientific exhibited the highest US/s (0.36 ± 0.04) and LC/s (0.20 ± 0.03). Mixed-effects modeling confirmed significant vendor-related differences for all parameters, with MicroPort consistently outperforming others across all gain levels and programming conditions, followed by Medtronic, Abbott, Biotronik, and Boston Scientific.
Conclusion
Using real human VF electrograms injected under identical conditions, this benchmark demonstrates measurable disparities in VF detection performance among contemporary ICD platforms, with sensitivity gaps amplifying at low signal amplitudes. This standardized human-signal injection model enables objective head-to-head evaluation of ICD sensing algorithms.Standardized VF injection protocol