Few FDA Approved AI/ML Orthopaedic Devices Have EU MDR Equivalents or Peer-Reviewed Validation.

BACKGROUND: Artificial intelligence (AI) and machine-learning (ML) technologies are increasingly being incorporated into orthopaedic medical devices, offering the potential to enhance surgical planning, intraoperative precision, and personalized care. The incorporation of AI into medical devices introduces unique risks and challenges that traditional regulatory pathways in the United States and European Union (EU) were not designed to address. Although the EU Artificial Intelligence Act is currently being implemented to provide a specific framework for AI medical device assessment, no equivalent directive exists in the United States. QUESTIONS/PURPOSES: In consideration of these evolving regulatory contexts, this study sought to address the following questions related to FDA-approved, AI/ML-enabled orthopaedic devices: (1) How often are FDA-approved, AI/ML-enabled orthopaedic devices also approved under the EU Medical Device Regulation (MDR)? (2) How often are AI/ML-enabled orthopaedic devices approved through the FDA 510(k) pathway using non-AI predicate devices? (3) How often do approved AI/ML-enabled orthopaedic devices have independent, peer-reviewed evidence supporting their use? (4) How often are the details of the training and validation datasets of these devices disclosed in publicly available regulatory documents? METHODS: A total of 37 AI/ML-enabled orthopaedic device approvals were identified from the FDA's list of 1017 AI/ML-enabled medical devices as of May 2025. Regulatory documents were analyzed to identify training and validation data sets and predicate devices. A predicate device is an existing legally marketed device that the FDA has previously approved; a manufacturer may claim that a new device is "substantially equivalent" to this approved device to obtain clearance through the 510(k) pathway. The availability of devices in the EU was assessed through the publicly available European Database on Medical Devices (EUDAMED) database and manufacturer websites. A systematic review of PubMed and manufacturer websites identified peer-reviewed evidence for each device. RESULTS: Our results showed that 38% (14 of 37) of AI/ML-enabled orthopaedic devices were approved for use in the EU, and devices were more frequently available in the United States first. All AI/ML-enabled orthopaedic devices approved by the FDA received regulatory clearance through the 510(k) pathway; most devices (62% [23 of 37]) were approved based on substantial equivalence to non-AI enabled devices. Independent peer-reviewed evidence was identified for 30% (11 of 37) of devices, with 67% (34 of 51) of published studies reporting manufacturer involvement. Information on training and validation datasets was disclosed in 57% (21 of 37) of 510(k) summaries, with dataset sizes ranging from 21 to 150,000 for training and 20 to 29,991 for validation. CONCLUSION: AI/ML-enabled orthopaedic devices have entered the market under existing regulatory frameworks in the United States and, at a slower pace, in the EU. Reliance on non-AI comparators and limited postmarket evaluation raises concerns regarding their long-term clinical reliability. CLINICAL RELEVANCE: As the level of patient risk under the current regulatory pathways remains uncertain, future research should compare AI/ML-enabled device recall rates across jurisdictions to assess whether AI-specific regulatory frameworks improve patient safety and device reliability.