ChatGPT as a collaborative research assistant in the ICF linking process of the brief version of the Burn Specific Health Scale

GÜL H., Çınar M. A., Bayramlar K.

Burns, cilt.51, sa.7, 2025 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 51 Sayı: 7
  • Basım Tarihi: 2025
  • Doi Numarası: 10.1016/j.burns.2025.107609
  • Dergi Adı: Burns
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, CAB Abstracts, CINAHL, EMBASE, MEDLINE, Veterinary Science Database
  • Anahtar Kelimeler: Burn injuries, ChatGPT-4o, ICF linking, Outcome measurement
  • Akdeniz Üniversitesi Adresli: Evet

Özet

Introduction: Burn injuries profoundly affect multiple aspects of health-related quality of life (HRQoL). The Brief Version of the Burn Specific Health Scale (BSHS-B) is commonly used to assess HRQoL in burn survivors. Linking such tools to the International Classification of Functioning, Disability and Health (ICF) enhances data comparability and standardisation for patients with burn injuries. However, linking process is often complex and time-consuming. Large language models may support linking process and help streamline future linking studies in burn rehabilitation. Objectives: This study evaluated the feasibility of using ChatGPT-4o as a collaborative assistant in the ICF linking process of BSHS-B items. Methods: The study followed the refined ICF linking rules. In the first stage, two physiotherapists independently linked the contents of BSHS-B items to ICF categories. When the two linkers disagreed, a third assigned the item to a category. In the second stage, ChatGPT-4o guided by specialised prompting performed the same task according to linking rules. In the content analysis, Cohen's Kappa coefficient was computed to evaluate the consistency between expert consensus and ChatGPT-4o-based linking. An agreement on item perspective analyses was also conducted. Frequencies of identified ICF categories across major domains were reported descriptively. Results: The agreement between linkers on ICF category assignment was fair (κ = 0.41, p < .001), while ChatGPT and expert consensus agreement was moderate (κ = 0.55, p < .001). In the perspective analysis, agreement between experts was fair (κ = 0.21, p < .01), whereas ChatGPT demonstrated almost perfect agreement with experts (κ = 0.86, p < .001). A total of 25 ICF codes were identified, mainly in Activity Participation (52.11 %) and Body Functions (40.85 %). Conclusion: ChatGPT demonstrated substantial potential in the ICF linking process as a supportive tool. While not replacing human expertise, ChatGPT may be able to reduce workload and facilitate ICF linking process.