xRead - Incorporating Artificial Intelligence into Clinical Practice (March 2026)

FIGURE 3 | Number of studies by specific applications of NLP for Clinical and Research Applications. A single study may contain multiple appli cations. [Color figure can be viewed in the online issue, which is available at www.laryngoscope.com.]

the remainder ranged from 94% to 100% [38, 55, 87, 142, 145, 147, 155, 170, 175]. Three studies implemented NLP into databases, including searching EMRs to identify patients who meet in clusion criteria, extracting data from a public database, and indexing biospecimens in a biobank [17, 29, 54]. Five studies used NLP to extract or analyze information from literature [13–15, 112, 172]. Seven studies utilized BERT for extraction [88, 107, 109, 115, 124, 144, 155]. Seventeen studies used NLP for thematic anal ysis [15, 22, 24, 43, 86, 88, 89, 107, 109, 112, 119, 120, 133, 138, 143, 144, 164]. Of these, nine studies uti lized NLP to analyze the sentiment of OHNS patient posts on forums and social media platforms [22, 24, 43, 86, 88, 107, 138, 143, 164]. Three studies used NLP to analyze letters of recommendation for fellowship applications, revealing gender-­ based variations [119, 120, 133].

3.4.2 | Manuscript Preparation Four studies explored NLP assistance in scientific writing [26, 41, 71, 130]. Three studies evaluated ChatGPT's provision of scientific references with accuracies of 73%–87%, compared to earlier versions (ChatGPT 3 and 3.5) with accuracies ranging from 10% to 60% [41, 71, 130]. Lechien et al. utilized ChatGPT 4 for writing assistance where it detected 50% of errors and pro posed acceptable corrections for 84% [26]. 3.4.3 | Evaluation of Scientific Reporting Five studies examined the use of NLP tools for the evaluation of scientific reporting within otolaryngology [18–20, 30, 69]. Dang and Hanba evaluated ChatGPT 3.5's ability to assess method ological quality, while Alohali et al. developed an NLP model

3054

The Laryngoscope, 2025