Eur Arch Otorhinolaryngol. 2025 Nov 17. doi: 10.1007/s00405-025-09760-8. Online ahead of print.
ABSTRACT
PURPOSE: Artificial intelligence (AI) has emerged as a potential tool in postoperative care, particularly for complex procedures like Endoscopic Transnasal Skull Base Surgery (ETSBS), where patient comprehension of recovery instructions is critical. This study aimed to compare the readability, understandability, and actionability of postoperative instructions generated by three AI platforms (ChatGPT, DeepSeek, and Gemini).
METHODS: Each platform was prompted to create ETSBS postoperative instructions. Readability was assessed using Flesch Kincaid Grade Level (FKGL) and Reading Ease (FKRE). The Patient Education Materials Assessment Tool for printable materials (PEMAT-P) was used to evaluate understandability and actionability. Two outputs per platform were analyzed. Statistical comparisons were conducted using Kruskal-Wallis tests and Pearson correlation coefficients.
RESULTS: Gemini had the highest FKRE score (52.18), followed by DeepSeek (46.46) and ChatGPT (39.85), though differences were not significant (p = 0.458). FKGL was lowest in Gemini (9.07), compared to DeepSeek (9.82) and ChatGPT (10.87) (p = 0.469). Understandability scores were highest in ChatGPT and DeepSeek (76.45%), while Gemini scored lower (63.30%, p = 0.005). ChatGPT showed the highest actionability (58.5%), followed by DeepSeek (51.0%) and Gemini (45.15%), with no significant difference (p = 0.645). A strong inverse correlation was found between FKRE and FKGL (r = -0.998, p = 0.000). Correlations with understandability and actionability were moderate and non-significant (p > 0.1).
CONCLUSION: While AI platforms generated similarly readable content, significant differences emerged in usability. None met optimal standards for patient education, highlighting the need for clinician review before clinical application.
PMID:41249683 | DOI:10.1007/s00405-025-09760-8
