•Retrospective analysis of 50 anonymized speech sample notes authored by two SLPs at a single-institution

•Processed using an in-house adaptation of the GPT-4 model (VERSA) to extract key speech and language concepts

•Prompts: patient's diagnosis, any surgical history along with the date of surgery, resonance, nasal airflow evaluations, grimace, and speech acceptability

•Resonance evaluation included hypernasality (0-4 scale) and hyponasality on a (0–2 scale)

•Nasal airflow evaluations assessed audible nasal emission or nasal turbulence (0-2 scale)

•Grimace (0 or 1) and speech acceptability (0-3) 

•The LLM had an overall accuracy of 92%, with the lowest performance in the "grimace" category

•These preliminary findings highlight the potential of LLMs in enhancing the efficiency and reliability of clinical assessments in speech-language pathology by streamlining the extraction of diagnostic and evaluative information from pediatric speech pathology notes

•This model reduced the time spent manually retrieving information and opens the potential to aggregate data to see trends and comprehensively extract data for research

•Small cohort of 50 anonymized speech notes from two speech language pathologist at a single-institution

•Limited generalizability of LLM at other institutions

•Future studies of this LLM model will focus on expanding the dataset and refining model parameters to further improve accuracy and applicability in diverse clinical scenarios