Artificial Intelligence (AI) in Orthopaedics

A prospective, randomized, blinded Level I study published in Knee Surgery, Sports Traumatology, Arthroscopy compared medical history summaries generated by ChatGPT-4o with documentation produced by clinicians in a paediatric orthopaedic practice. Twenty consecutive consultations involving children and adolescents presenting with knee disorders were transcribed and independently summarized by both clinicians and ChatGPT-4o. Three orthopaedic specialists then assessed the quality of the documentation using eight predefined clinical criteria.

The study demonstrated that human-generated documentation significantly outperformed AI-generated summaries in overall quality, particularly in clinically critical domains such as temporal consistency, spatial consistency, accuracy of accident descriptions, and overall clinical impression. While ChatGPT-4o produced summaries with acceptable writing style and documented previous interventions comparably to clinicians, it frequently introduced temporal inconsistencies and omitted important details regarding injury mechanisms. The authors conclude that current large language models cannot yet replace human medical documentation in paediatric orthopaedics and require careful clinical supervision. Instead, the findings support a hybrid workflow in which AI serves as an assistive tool while final clinical judgement remains with healthcare professionals.

Among the co-authors, Professor Victor Valderrabano contributed to the design and scientific validation of this comparative evaluation, helping to assess the current capabilities and limitations of artificial intelligence in orthopaedic clinical documentation. His involvement reinforces the growing importance of evidence-based research to define the appropriate role of AI technologies in routine clinical practice.