Evaluation of ChatGPT-5 responses to patient-centered questions on stromal vascular fraction for knee osteoarthritis: fair to good quality and content

The global burden of symptomatic knee osteoarthritis (KOA) continues to grow, driving clinical interest in biological treatment options. Stromal vascular fraction (SVF), derived from adipose tissue, has gained attention as a potential therapy for KOA. As patients increasingly utilize large language models (LLMs) like ChatGPT for health information, this study aimed to evaluate the patient-information quality and readability of AI-generated responses to a curated set of patient-centered questions on SVF therapy for KOA. Thirty patient-centered questions were developed through literature review by two experts and then posed to ChatGPT-5 after asking it to answer from an orthopaedic specialist perspective. Responses from ChatGPT-5 were evaluated by four orthopaedic specialists using three quality instruments: DISCERN, NLAT-AI (assessing five domains: Accuracy, Safety, Appropriateness, Actionability, and Effectiveness), and the Mika et al. scoring system. Readability was assessed using five standard metrics. Mean scores were as follows: DISCERN 44.61 ± 4.94 (Fair); for NLAT-AI domains, Accuracy 3.92 ± 0.50 (Good), Safety 3.23 ± 0.78 (Fair), Appropriateness 4.14 ± 0.32 (Good), Actionability 3.20 ± 0.67 (Fair), and Effectiveness 4.46 ± 0.35 (Excellent); NLAT-AI (Total/ sum of five domains) 18.95 ± 1.93 (Good); Mika et al. 2.45 ± 0.49 (Fair). Readability metrics indicated an 11th to 12th grade reading level. ChatGPT-5 provides fair-to-good quality responses to patient-centered questions about SVF in KOA. The answers are generally effective and clinically appropriate, with good accuracy, and often require only limited additional clarification. However, safety cautions and practical guidance are less consistently covered, and the reading level is relatively high. Further research is needed to clarify its role as an adjunct tool for patient education in this setting.