[Artificial intelligence in the preparation of clinical expert opinions : Current evidence, use cases, and limitations in orthopedic assessment].

backgroundGenerative AI, large language models, and ambient AI systems are rapidly being integrated into documentation-related processes in medicine. However, the data available for orthopedic expert reports is significantly less extensive than that or general clinical documentation. RESEARCH QUESTION: What evidence exists by June 2026 for the use of AI in the preparation of clinical and medico-legal reports? Which process steps are suitable for meaningful application in orthopedics, and what limitations arise from legal, data protection, and reimbursement considerations? MATERIALS AND

methodsA narrative, structured review was conducted using four thematic PubMed search blocks on clinical documentation, forensic reporting, orthopedics, and automation bias, supplemented by manual searches of reference lists and official German and European sources. As of June 29, 2026, 486 unique PubMed hits had been identified. A total of 28 sources had been included in the narrative synthesis.

resultsThe most robust evidence pertains to AI-supported clinical documentation and "ambient scribing," which offer efficiency gains, a reduced documentation burden, and heterogeneous, although not error-free, document quality. For expert-level forensic reporting, initial positive studies exist, but their generalizability is limited. In orthopedic expert opinions, file preparation, transcription, linguistic revision, and preliminary quality assurance are particularly suitable. Examination, weighting of contradictory findings, causality assessment, prognosis, and final assumption of responsibility remain non-delegable. Section 407a of the German Code of Civil Procedure (ZPO), Section 8a of the German Act on the Remuneration of Experts and Witnesses (JVEG), Article 9 of the GDPR, and Section 203 of the German Criminal Code (StGB) impose strict limits. The decision of the Darmstadt Regional Court of November 10, 2025, illustrates that a lack of transparency and insufficient usability can lead to the complete loss of remuneration. New considerations include the European Commission's drafts published in May/June 2026 regarding the classification of high-risk AI systems. Currently, the governance, validation, and security costs associated with AI use are not separately reflected in the remuneration regulations.

conclusionAI is currently primarily an assistance tool, not a replacement for orthopedic experts. Responsible use requires clear process boundaries, trained users, data protection-compliant infrastructure, and documented quality control. The additional costs are inadequately represented in the remuneration regulations.