Scholarlink

Objective <p>This pilot study evaluated ChatGPT’s ability to answer standard emergency room (ER) case questions compared with that of junior residents at a level 1 trauma centre. Since its release in March 2023, ChatGPT‑4 has seen widespread adoption in education, research, and clinical practice, prompting interest in its medical utility. Despite concerns regarding accountability and trust, large language models (LLMs) have shown potential by-passing medical exams and generating differential diagnoses.</p> Materials and methods <p>The study was conducted at a level 1 trauma centre with an interdisciplinary ER from 21st to 27th February 2025. 21 fictional medical cases commonly encountered in the surgical ER were used. Seven junior residents with varying levels of training were included for comparison. ChatGPT and the doctors were prompted to answer these cases under identical conditions. Two board‑certified trauma surgeons independently rated responses for correctness, completeness, and adaptability; results were statistically analyzed.</p> Results <p>ChatGPT achieved mean scores of 4.62 ± 0.79 (correctness), 4.67 ± 0.56 (completeness), and 4.52 ± 0.66 (adaptability). Residents in years 1–3 scored 4.19 ± 0.05, 3.95 ± 0.48, and 4.05 ± 0.1, respectively; years 4–6 scored 3.99 ± 0.19, 3.71 ± 0.3, and 3.96 ± 0.2. ChatGPT and residents both performed best in lower extremity cases (14.5 vs. 12.9 points).</p> Conclusions <p>This pilot study indicates that ChatGPT can generate ER case responses comparable in quality to those of junior residents, particularly in completeness and correctness. Given the small sample size and vignette-based design, these findings do not support replacing physicians but highlight the potential of large language models as supportive or educational tools requiring clinician oversight. Establishing safe and ethical frameworks for AI use in medicine remains essential.<!--Query ID="Q1" Text="Please check and confirm the author names and initials are correct. Also, kindly confirm the details in the metadata are correct." Resolved="yes"--></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ChatGPT-4 in comparison with traumatological junior doctors in emergency room cases at a level 1 trauma centre – a pilot study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月26日

查看全文
  • Sebastian Wegmann,
  • Jannick Leyendecker,
  • Maximilian Lenz,
  • Michael Sarter,
  • Lars-Peter Müller,
  • Maximilian Weber

摘要

Objective

This pilot study evaluated ChatGPT’s ability to answer standard emergency room (ER) case questions compared with that of junior residents at a level 1 trauma centre. Since its release in March 2023, ChatGPT‑4 has seen widespread adoption in education, research, and clinical practice, prompting interest in its medical utility. Despite concerns regarding accountability and trust, large language models (LLMs) have shown potential by-passing medical exams and generating differential diagnoses.

Materials and methods

The study was conducted at a level 1 trauma centre with an interdisciplinary ER from 21st to 27th February 2025. 21 fictional medical cases commonly encountered in the surgical ER were used. Seven junior residents with varying levels of training were included for comparison. ChatGPT and the doctors were prompted to answer these cases under identical conditions. Two board‑certified trauma surgeons independently rated responses for correctness, completeness, and adaptability; results were statistically analyzed.

Results

ChatGPT achieved mean scores of 4.62 ± 0.79 (correctness), 4.67 ± 0.56 (completeness), and 4.52 ± 0.66 (adaptability). Residents in years 1–3 scored 4.19 ± 0.05, 3.95 ± 0.48, and 4.05 ± 0.1, respectively; years 4–6 scored 3.99 ± 0.19, 3.71 ± 0.3, and 3.96 ± 0.2. ChatGPT and residents both performed best in lower extremity cases (14.5 vs. 12.9 points).

Conclusions

This pilot study indicates that ChatGPT can generate ER case responses comparable in quality to those of junior residents, particularly in completeness and correctness. Given the small sample size and vignette-based design, these findings do not support replacing physicians but highlight the potential of large language models as supportive or educational tools requiring clinician oversight. Establishing safe and ethical frameworks for AI use in medicine remains essential.