Letter to Humans‐written versus <scp>ChatGPT</scp>‐generated case reports

Abstract

We read with great interest Matsubara's article, “Humans-written versus ChatGPT-generated case reports,” in which the potential of large language models (LLMs) in writing case reports is demonstrated, along with their limitations, particularly the lack of a “human touch.”1 Case reports are fundamental to medical literature, yet writing them can be challenging, especially for time-constrained physicians and non-native English speakers.2 These barriers often limit the ability to share valuable clinical insights. In this study, ChatGPT generated three case reports (#1–#3) with varying levels of input. Three researchers compared the readability of the most detailed ChatGPT-generated report (#3) with a human-written report (#4). The findings showed that while LLMs can produce coherent and structured drafts, significant limitations remain, such as a lack of contextual depth and nuanced expression, highlighting the importance of human oversight to ensure quality. Prior to Matsubara's study, a similar investigation comparing case reports authored by humans and LLMs had been conducted. Pinto et al. conducted a comparative study between human- and ChatGPT-generated case reports, including 22 reviewers with diverse medical expertise.3 The reviewers assessed readability, including clarity, logical flow, nuanced writing, and the ability to highlight key diagnostic insights. To complement these subjective evaluations, they used GPTZero, an artificial intelligence tool, to assess perplexity and burstiness as objective measures. The study found that while ChatGPT-produced reports were coherent, human-written reports excelled in quality and depth, emphasizing the value of human authorship. Matsubara also emphasized the importance of a “human touch” in case reports, a subjective quality that remains difficult to define and measure. While Pinto et al. utilized multiple metrics to evaluate readability, none directly measured “human touch.” This suggests the challenge of quantifying subjective qualities in writing. To address these challenges, we propose a collaborative approach; LLMs generate drafts that human authors refine, or vice versa.4 This method combines the efficiency of LLMs with the originality and critical thinking of human authors. In addition, prompt engineering can add empathy and depth, bringing a human touch to LLM drafts. Matsubara's study did not include a case report written by this method, leaving an opportunity for future research. In conclusion, LLMs have a great potential to support physicians with various language backgrounds in academic writing. Combining human expertise with LLMs provides a practical way to increase participation in medical literature and foster a more inclusive academic community. Shunsuke Koga: Contributed to the concept and drafted the manuscript. Wei Du: Reviewed the manuscript. The authors declare no conflicts of interest. Not applicable.

Ähnliche Arbeiten