BACKGROUND

Patients increasingly turn to large language models (LLMs) and social media platforms for medical advice. The accuracy of these sources, particularly compared to peer-reviewed clinical practice guidelines, remains poorly characterized.

MATERIALS AND METHODS

This cross-sectional study evaluated the perceived accuracy of spine-related medical advice generated by ChatGPT (ChatGPT (OpenAI, powered by GPT-4, San Francisco, CA, USA), TikTok (Los Angeles, CA, USA), and the North American Spine Society (NASS) clinical practice guidelines. Medical advice for four spine pathologies was collected from each source. Sixteen orthopedic surgeons rated the accuracy of excerpted recommendations on a 10-point Likert scale. Descriptive statistics summarized mean ratings and standard deviations.

RESULTS

For lumbar stenosis, mean (±SD) accuracy scores were 7.75 ± 2.11 for ChatGPT, 7.00 ± 1.80 for NASS, and 2.50 ± 1.54 for TikTok. For lumbar spondylolisthesis, scores were 7.56 ± 1.50 for ChatGPT, 5.94 ± 2.63 for NASS, and 5.31 ± 2.49 for TikTok. For lumbar disc herniation with radiculopathy, scores were 7.25 ± 2.13 on ChatGPT, 7.06 ± 1.55 on NASS, and 6.44 ± 2.03 on TikTok. For cervical radiculopathy, scores were 7.13 ± 1.38 for ChatGPT, 4.00 ± 2.44 for NASS, and 6.50 ± 2.12 for TikTok.

CONCLUSIONS

ChatGPT-generated outputs received the highest ratings for perceived accuracy. NASS guidelines, while evidence-based and peer-reviewed, remain inaccessible to most patients. Professional societies may consider adapting guideline content for dissemination via widely used digital platforms to improve public education and reduce misinformation.