Post-Market Safety Signals for Pulsed Field Ablation Devices: An Artificial Intelligence-Augmented Analysis

Abstract

Background: Pulsed field ablation (PFA) is increasingly used for cardiac arrhythmias, yet uncommon complications and device-specific failure modes may remain undercharacterized after approval. We evaluated recent FDA MAUDE reports involving leading PFA platforms using an artificial intelligence-augmented safety-surveillance workflow.

Methods: MAUDE reports were queried using prespecified manufacturer, trade-name, and component terms for major PFA systems, including Affera/Sphere-9, FARAPULSE/ FARAWAVE, PulseSelect, TRUPULSE/VARIPULSE, Volt/Current, and related catheters, generators, sheaths, and mapping components. Reports were screened, deduplicated, and parsed with ChatGPT-assisted extraction to identify event severity, clinical outcomes, procedural context, and technical failure modes. A standardized taxonomy was applied, and a random sample was manually verified by independent reviewers.

Results: A total of 2,063 unique pulsed field ablation (PFA)-related MAUDE reports were analyzed; malfunction reports predominated (59.8%), followed by injury (38.3%) and death (1.9%) reports. Overall, any device/use problem or malfunction was identified in 74.6% of reports, with frequent issues including power/software/sensing flaws (22.2%), air/gas bubbles (18.7%), and material deformation/kinking (15.6%). Clinical harm was absent or unreported in 61.0% of cases; however, the most common documented patient outcomes were hospitalization or prolonged stay (23.3%), arrhythmias (15.7%), and pericardial effusion or tamponade (10.1%). Distinct device-specific reporting patterns emerged, characterized by high rates of air/gas events in FARADRIVE (67.0%), material deformation in PulseSelect (52.2%), power/software issues in FARASTAR (67.1%), and cerebrovascular events in VARIPULSE (52.5%).

Conclusions: Real-world post-market safety reports for PFA systems are predominantly driven by technical malfunctions rather than direct clinical harm. However, distinct, platform-specific failure modes highlight the unique mechanical and workflow profiles of individual PFA designs. These findings also underscore the promising role of artificial intelligence-assisted surveillance in analyzing unstructured registries, providing targeted insights to guide ongoing operator education and iterative device development.