Comment on ‘How Successful Is <scp>AI</scp> in Developing Postsurgical Wound Care Education Material?’

Abstract

We read with great interest the article by Sürme et al. [1]. The authors aimed to generate postoperative wound-care educational material using ChatGPT-4 and to evaluate the resulting text in terms of readability, understandability, applicability, and quality. In their study, they identified the educational needs of surgical patients, used ChatGPT to produce patient-focused content based on several prompts and supporting literature, and created accompanying visuals with DALL-E. The resulting text was assessed by 10 surgical nursing experts using the Patient Education Materials Assessment Tool (PEMAT-P) and the Global Quality Scale (GQS). Readability was evaluated using both Turkish and English readability indices. Furthermore, it was unclear whether the educational content was generated in Turkish or English. If the outputs were in Turkish, then the use of English-language readability formulas (e.g., Flesch–Kincaid, Flesch Reading Ease) would be inappropriate because these were developed and normed for English texts. Conversely, if the outputs were in English, the application of the Turkish-adapted Ateşman formula may not yield valid results, as it was explicitly designed for Turkish-language material. In the study by Gordejeva et al. [2], the Flesch Reading Ease (FRE) index was calculated using distinct, language-adapted formulas for English, German, and Russian, demonstrating that readability metrics are inherently language-specific and must be adapted to the linguistic structure of each language. The formulas used in this study are as follows. English: FRE = 206.835 − (1.015 × ASL) − (84.6 × ASW). German: FRE = 180 − ASL − (58.5 × ASW). Russian: FRE = 208.7 − (2.6 × ASL) − (39.2 × ASW). Abbreviations: ASL (Average Sentence Length), ASW (Average Syllables per Word). The variations in the formula across English and other languages reflect fundamental structural differences in linguistic composition. Even when identical numerical values are used for sentence and word parameters, applying them to different language-specific formulas can yield markedly different readability levels. According to Durukan [3], because readability formulas developed for other languages may produce inconsistent results when applied together, each readability index should be considered language-specific and used within its own linguistic context. In the reference study, the Turkish Ateşman index indicated a readability level of 9th–10th grade. Previous studies have shown that when large language models are prompted with explicit readability targets such as ‘generate at a 6th-grade reading level’ or ‘explain to a medical layperson’, the resulting outputs demonstrate significantly improved readability scores [4, 5]. Further research should clearly specify the language of artificial intelligence (AI) generated materials, as readability indices are language-dependent and must correspond to the linguistic structure of the text. Additionally, prompting AI models with explicit readability targets may help generate content that is more comprehensible and appropriately tailored for the intended audience. All authors wrote the main manuscript text and prepared it. All authors reviewed the manuscript. Advanced AI systems such as ChatGPT-5 were used solely to assist with grammatical refinement and linguistic fluency, ensuring clarity and consistency in the final version of the manuscript. All content was carefully reviewed and verified by the authors before submission. The authors have nothing to report. The authors declare no conflicts of interest. No datasets were generated or analyzed during the current study.

Ähnliche Arbeiten