Innovations in otolaryngology using LLM for early detection of sleep-disordered breathing

Sleep Disordered Breathing (SDB), including conditions like Obstructive Sleep Apnea (OSA), represents a major health concern, characterized by irregular airflow during sleep due to airway obstruction. SDB can result in serious health problems. Implementation of early intervention is vital whenever patient outcomes are to be considered. This research aims to advance research on otolaryngology using Machine Learning (ML) models, and Large Language Models (LLM) for identification of SDB using Electronic Health Record (HER). The approach proposes a hybrid ML framework combining the Dynamic Seagull Search algorithm-driven Large Language model (DSS-LLM). The extensive clinical dataset is used to train the model. It includes patient demographics, medical history, sleep habits, comorbidities, and physical measurements. Data pre-processing involves handling missing values, applying NLP techniques, and normalization. Feature extraction is done using Principal Component Analysis (PCA) to reduce the dimensionality of the hyperparameters and finally for selecting the best set of predictors. The extracted features are then used to train the proposed DSS-LLM model, which incorporates the DSS algorithm to optimize the LLM classifier, improving classification accuracy and model robustness. Subsequently, the idea of LLM is introduced for its application on textual clinical records comprising physicians' reports and patients' symptoms. The findings from an experiment suggest that the proposed model enhances the classification accuracy achieved to 98.91 %, precision attained by 98.9 %, recall achieved to 98.92 % and F-1 score attained by 98.58 % as compared to the models developed earlier. This research provides a novel solution to the screening of OSA at the pre-clinical level which involves hybrid machine learning models integrated with LLMs. This proposed framework is expected to boost clinical judgment and thereby increase better ophthalmology outcomes for patients.