Device Safety Under Real-World Conditions

Abstract

Medical device pre-market testing evaluates safety under controlled laboratory and clinical conditions that inevitably underrepresent the range of environmental stresses, patient population diversity, and device interaction scenarios that devices encounter after approval. Patients swim in chlorinated pools with their cardiac monitors. Construction workers use power tools within electromagnetic interference range of their pacemakers. Paediatric patients with adult-labelled devices receive dose exposures that body weight scaling does not fully account for. Elderly patients with cognitive impairment use device interfaces designed for cognitively intact adults. These real-world conditions generate safety signals that pre-market testing cannot detect, and the systematic characterisation of those signals requires post-market surveillance approaches designed specifically to capture real-world performance rather than controlled-condition performance. This study presents the Device Safety Real-World Evaluation Framework (DSRWEF), evaluating five safety evaluation approaches -- post-market clinical follow-up with structured safety monitoring (PMCF-SSM), proactive registry-based safety surveillance (Reg-SS), AI-powered adverse event signal detection (AI-AES), patient-reported safety outcome programmes (PR-SOP), and comparative safety analysis using propensity-matched real-world data cohorts (PS-RWD) -- across four real-world device safety condition contexts: extreme temperature and environmental conditions, physical activity and occupational stress, multi-device electromagnetic interactions, and patient population diversity challenges. Performance was scored using the Real-World Safety Performance Score (RWSPS), a weighted composite of safety signal sensitivity (0.30), real-world applicability (0.20), regulatory compliance (0.20), implementation speed (0.15), and cost effectiveness (0.15). AI-powered adverse event signal detection achieved the highest RWSPS (0.887) through the best safety signal sensitivity (0.940) and strong implementation speed (0.880). Registry-based surveillance ranked second (0.883) with the best regulatory compliance (0.900) and real-world applicability (0.920). Patient-reported safety outcome programmes led on real-world applicability (0.940) and cost effectiveness (0.920), capturing patient-experience safety signals that clinical data sources consistently miss.