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

4.9 | Limitations Due to the rapidly evolving research in this domain, this study may have missed articles published after our search date, August 30 2024. Additionally, the Oxford CEBM levels of evidence do not apply to a significant portion of these studies, limiting our assessment of potential bias. Most studies are of low evidence, and methodologies were very heterogeneous; therefore, limiting the conclusions about NLP performance that can be drawn from this review. The ethical implications of NLP in OHNS were not reviewed in this study. 4.10 | Future Implications The use of NLP in otolaryngology without a governance frame work poses significant ethical risks, including patient safety concerns, data integrity issues, and potential privacy violations, particularly with the use of LLM's [189]. Without proper over sight, AI-generated outputs may contain inaccuracies or biases that could mislead clinicians and patients, compromise research findings, and exacerbate healthcare disparities. Additionally, the misuse of patient data for AI training without transparent consent processes raises concerns about privacy and intellec tual property. To ensure ethical compliance, AI tools in OHNS must be rigorously validated, used alongside clinical expertise, and governed by clear protocols that prioritize patient welfare, privacy, and scientific integrity. Future research should validate the use of AI scribes in the otolar yngology context, and more work is needed for surgical documen tation. Guidelines around NLP in scientific reporting, particularly manuscript preparation, are required. Further research is needed to validate NLP's role in manuscript preparation, including areas like text, table, figure, and citation generation. Research on NLP in trainee education is sparse, particularly regarding its trust worthiness and relevance to learning objectives. Future work to implement NLP tools for search and data extraction in databases would be beneficial as there is evidence of impressive accuracies for these tasks. Finally, NLP's potential in resource-limited set tings should be a focus of future work, taking into consideration the ramifications on performance if a non-English language is used. Considering ChatGPT's declining market share compared to LLMs such as Claude, Llama, and Gemini, it is crucial that more efforts are made to incorporate these alternatives [190]. 5 | Conclusions NLP in otolaryngology is rapidly growing, with a broad range of applications, but most research is nascent. Although LLMs can pass the OHNS board exam simulations, they are not yet reliable for clinical use. NLP shows promise in enhancing pa tient education, streamlining documentation, improving data management for efficiency and care, and triaging. It also offers potential in early complication detection for better outcomes. NLP is currently accurate and efficient in data extraction from diverse sources and aiding in qualitative analysis. Guidelines for NLP use in research and its application in administration and trainee education warrant further study.

and biobanks, to streamline searches and data extraction [13– 15, 17, 29, 54, 112, 172]. NLP can enhance data analysis through tasks such as data cleaning, statistical analysis, and thematic analysis. Thematic analysis uses topic modeling and sentiment analysis to create themes among unstructured data, making it a valuable tool for qualitative studies that traditionally require extensive human effort. Two main research applications of thematic analysis seen in the literature include analyzing patient posts on forums and social media, and examining recommendation letters for demo graphic biases. NLP has the ability to reveal subtle insights that might be missed by human reviewers on a large scale, saving resources. 4.6 | Manuscript Preparation NLP can aid in many facets of scientific dissemination, in cluding writing support. Using NLP for writing assistance has demonstrated significant benefits, particularly for non-native English speakers [130]. Its perceived value among scholars is evidenced by the rising prevalence of suspected AI-generated content in peer-reviewed otolaryngology literature [30]. While there is also growing enthusiasm for using NLP to assist with citation creation, our findings highlight the limitations of NLP in both writing support and citation generation. Despite the po tential for NLP to aid in the communication of scientific results, developing guidelines in the use of AI platforms is imperative to ensure that academic content is generated by authors as opposed to chatbots. 4.7 | Evaluation of Scientific Reporting NLP-based evaluation of scientific reporting tools can supple ment conventional review processes to uncover insights that might elude human reviewers and improve objectivity. In par ticular, NLP-driven algorithms capable of predicting impact may provide data-driven insights that could ensure robust methodology, results, and guide editorial decisions. Moreover, NLP can provide an automated safeguard for ensuring the in tegrity of scientific communication. While current research has evaluated it for the use of identifying misleading find ings, ensuring correct conflicts of interest, and assessing for AI use in published works, other potential applications may include identifying unusual patterns in data, citation verifica tion, and fact checking. More research is required to evaluate performance as well as to apply to non-manuscript scientific communication. 4.8 | Level of Evidence Most studies did not fit the Oxford CEBM 2011 evidence hier archy ( n = 134), mainly consisting of performance evaluation ( n = 119), qualitative ( n = 9), bibliometric analysis ( n = 4), and methods ( n = 3). Among classifiable studies, 26 were lower ev idence levels (4 or 5) and 5 higher (1–3), highlighting the field's emerging nature.

3056

The Laryngoscope, 2025