Nature and Science of Sleep (Dec 2024)

Patient Support in Obstructive Sleep Apnoea by a Large Language Model – ChatGPT 4o on Answering Frequently Asked Questions on First Line Positive Airway Pressure and Second Line Hypoglossal Nerve Stimulation Therapy: A Pilot Study

  • Pordzik J,
  • Bahr-Hamm K,
  • Huppertz T,
  • Gouveris H,
  • Seifen C,
  • Blaikie A,
  • Matthias C,
  • Kuhn S,
  • Eckrich J,
  • Buhr CR

Journal volume & issue
Vol. Volume 16
pp. 2269 – 2277

Abstract

Read online

Johannes Pordzik,1 Katharina Bahr-Hamm,1 Tilman Huppertz,1 Haralampos Gouveris,1 Christopher Seifen,1 Andrew Blaikie,2 Christoph Matthias,1 Sebastian Kuhn,3 Jonas Eckrich,1 Christoph R Buhr1,2 1Department of Otorhinolaryngology, University Medical Center of the Johannes Gutenberg-University Mainz, Mainz, Germany; 2School of Medicine, University of St Andrews, St Andrews, UK; 3Institute for Digital Medicine Philipps-University Marburg and University Hospital of Giessen and Marburg, Marburg, GermanyCorrespondence: Johannes Pordzik, Department of Otorhinolaryngology, University Medical Center of the Johannes Gutenberg-University Mainz, Mainz, Germany, Email [email protected]: Obstructive sleep apnoea (OSA) is a common disease that benefits from early treatment and patient support in order to prevent secondary illnesses. This study assesses the capability of the large language model (LLM) ChatGPT-4o to offer patient support regarding first line positive airway pressure (PAP) and second line hypoglossal nerve stimulation (HGNS) therapy.Methods: Seventeen questions, each regarding PAP and HGNS therapy, were posed to ChatGPT-4o. Answers were rated by experienced experts in sleep medicine on a 6-point Likert scale in the categories of medical adequacy, conciseness, coherence, and comprehensibility. Completeness of medical information and potential hazard for patients were rated using a binary system.Results: Overall, ChatGPT-4o achieved reasonably high ratings in all categories. In medical adequacy, it performed significantly better on PAP questions (mean 4.9) compared to those on HGNS (mean 4.6) (p < 0.05). Scores for coherence, comprehensibility and conciseness showed similar results for both HGNS and PAP answers. Raters confirmed completeness of responses in 45 of 51 ratings (88.24%) for PAP answers and 28 of 51 ratings (54.9%) for HGNS answers. Potential hazards for patients were stated in 2 of 52 ratings (0.04%) for PAP answers and none for HGNS answers.Conclusion: ChatGPT-4o has potential as a valuable patient-oriented support tool in sleep medicine therapy that can enhance subsequent face-to-face consultations with a sleep specialist. However, some substantial flaws regarding second line HGNS therapy are most likely due to recent advances in HGNS therapy and the consequent limited information available in LLM training data.Keywords: AI, ChatGPT, OSA, PAP, hypoglossal nerve stimulation

Keywords