GPT-5 outperforms other LLMs in FAI treatment decision-making from Chinese real-world data

In a retrospective study of 26 femoroacetabular impingement patients, GPT-5 achieved 88% accuracy (85% sensitivity, 92% specificity) for treatment decisions when given radiology plus medical record data, outperforming other large language models (Gemini 2.5 Pro 62%, DeepSeek-R1…

• Retrospective single-center study of 26 hospitalized FAI cases (13 surgical, 13 conservative).
• GPT-5 showed significant confidence-accuracy correlation (Group B r=0.55, p<0.01); other models had inconsistent correlations.
• Limitations: very small sample size, retrospective design, no external validation; multi-center prospective evaluation required before clinical use.

RadPigeon summaries are original and for information only. They are not clinical advice.