Real-World Data in Device Safety Evaluation

Abstract

Pre-market clinical trials for medical devices are designed to establish safety and effectiveness under controlled conditions -- but they are not designed to detect rare adverse events, long-term failure modes, or performance differences across the full heterogeneity of real-world patient populations. Real-world data fills that gap, and regulators on both sides of the Atlantic have moved steadily toward making RWD-based post-market safety surveillance a formal component of the device lifecycle. This study introduces the Real-World Device Safety Assessment Framework (RWDSAF), evaluating five RWD methodologies -- electronic health record-based passive surveillance (EHR-PS), medical device registry analysis (Registry-A), insurance claims adverse event mining (Claims-AEM), patient-reported outcome monitoring (PRO-M), and AI-assisted multi-source signal detection (AI-MSD) -- across four device safety contexts: implantable cardiac devices, orthopaedic implants, diagnostic imaging devices, and drug-eluting stents. Performance was scored using the Safety Evidence Quality Score (SEQS), a weighted composite of signal detection sensitivity (0.30), data completeness (0.20), temporal coverage (0.20), regulatory acceptance (0.15), and implementation cost-efficiency (0.15). Medical device registry analysis achieved the highest SEQS (0.903), with the best signal detection sensitivity for implantable devices (0.920) and the highest data completeness (0.940) and regulatory acceptance (0.940) of any methodology. AI-assisted multi-source detection ranked second (0.897), with the study's best signal detection sensitivity overall (0.960) and a demonstrated ability to surface safety signals 4.2 months earlier than single-source surveillance methods. EHR-based surveillance (0.874), claims mining (0.868), and PRO monitoring (0.850) each showed distinct contextual advantages that support a hybrid deployment strategy rather than any single methodology.