Assessing Use of ChatGPT to Automate Case Report Creation

Introduction: Case reports contribute to broader medical knowledge by describing unique presentations which may guide larger series or inform clinicians who are investigating similarly rare cases. Increasing patient volumes for physicians may be leading to more unique cases observed per physician with less time available to report them. ChatGPT is a recently released Generative Pre-trained Transformer (GPT) chatbot published by OpenAI which demonstrated utility in the clinical arena, as evidenced by achieving a passing score on the United States Medical Licensing Step 1 examination. Here we aim to assess the utility of using ChatGPT to expedite the creation of a clinical case report.

Methods: Clinical documentation from a previously published case of neurosarcoidosis involving the cervical vagus nerve was collected and de-identified. A case report was constructed by ChatGPT by prompting the chatbot to summarize the clinical documentation. The 2013 CARE checklist for case reports was used to assess the content of the ChatGPT-constructed case report and compare it with the published case report of the same case written by human authors only.

Results: The average score on the CARE checklist of the Chat-GPT constructed case was 15.5, which is less than the average score of the human-authored published case, 21.5. ChatGPT demonstrated utility in summarizing clinical findings, diagnostic findings, and therapeutic interventions. Introducing and discussing the case were mostly inaccurate when performed by ChatGPT.

Conclusions: ChatGPT proficiently summarized clinical findings of a unique case, however was unable to put these findings in context as would be required for a meaningful case report. This project is the first to assess ChatGPT utility in preparing clinical case reports.