Large language models are drawing rapid interest in spine surgery, but their value currently lies in supporting physicians rather than replacing their judgment, according to an article written by Aimen Khan, Maximillian Lee, Noah Pogonitz, Daniel Park, MD, and Kern Singh, MD, in the Summer 2026 issue of Vertebral Columns. Dr. Park and Dr. Singh are orthopedic spine surgeons at Chicago-based Rush University Medical Center.
Eight things to know:
- Patients are wary but already using AI with their health data. In a recent national survey, 77% of U.S. adults said they are concerned about the privacy implications of sharing health information with AI, yet 41% reported having uploaded personal medical information into an AI chatbot, the authors wrote.
- LLMs predict text; they do not reason like a physician. LLMs generate responses by identifying patterns in large volumes of text and predicting the next likely word, the authors explained. That produces convincing, human-like output, but the systems do not understand what they generate and can “hallucinate,” producing fabricated information or citations that sound credible.
- Documentation is the most immediate application. Because spine patients often carry extensive records, LLMs can synthesize notes and surface relevant points efficiently, the authors wrote. But studies of AI-generated documentation have found inaccuracies, including removed findings and incorrect details. Because of that, the authors said LLMs currently work best as a drafting tool requiring physician oversight, alongside tools that organize clinic notes from patient-surgeon conversations.
- Fabricated references are a major limitation in research use. AI-generated abstracts have been hard for reviewers to distinguish from human-written ones, the authors wrote, but a study analyzing ChatGPT-generated citations found a substantial proportion were fabricated or significantly inaccurate, and hallucinated references have turned up in accepted manuscripts, including conference papers.
- LLMs can aid patient communication, with limits. The models can generate simplified explanations of conditions such as lumbar stenosis, cervical myelopathy or spondylolisthesis and help draft responses to common patient questions, the authors wrote. However, they cannot replicate the clinical judgment spine surgery requires.
- LLMs are not ready to make surgical decisions. Enthusiasm for AI currently exceeds the evidence of its reliability, the authors wrote. In a study evaluating LLMs for minimally invasive spine surgery triage, agreement with expert-derived procedural categories was only slight to fair, and weaker still for selecting specific procedures, leading those researchers to conclude procedure selection should remain expert-led.
- Legal and regulatory questions remain unresolved. Because LLMs may process sensitive patient data, HIPAA and HITECH Act compliance are key considerations, the authors wrote. Liability for clinical decisions stays with the treating physician, and the researchers advised against entering protected health information into publicly available AI platforms without appropriate safeguards.
- LLMs are best used as adjunctive tools. LLMs are “neither an infallible process nor an overhyped assistant” and should be understood and controlled by the user, according to the authors. Used appropriately with physician oversight, they can speed documentation, streamline administrative tasks, support research and improve patient health literacy without compromising standards of care.
At the Becker’s 32nd Annual Meeting: The Business and Operations of ASCs, taking place October 29-31 in Chicago, ASC leaders, surgeons and healthcare executives will explore strategies to drive growth, enhance operational performance, navigate reimbursement challenges and prepare for the future of ambulatory surgery. Apply for complimentary registration now.
