Identifying Screening-Relevant Context in an OSA Study Using Clinical Note Metadata and LLM-Extracted Signals

Author: Ashley Batugo

(ashley[dot]batugo[at]pennmedicine[dot]upenn[dot]edu)

Overview

This project examined which note-level metadata features are associated with clinical research coordinator (CRC) exclusion decisions during screening for an NIH-funded Obstructive Sleep Apnea (OSA) clinical study. Large Language Models were used to extract note level exclusion flags, and multivariable logistic regression models used these exclusion flags (the outcome veriable) to identify the most important metadata features.

Repository Structure

1. final_presentation: Subdir containing files related to the project presentation
   • ./css: Relevant css for xarigan slides
   • ./img: Main image for presentation (Generated by ChatGPT)
   • BMIN5030_PRES_BATUGO.Rmd: Rmd file of the presentation
   • BMIN4030_PRES_BATUGO.html: Html file of the presentation
2. final_report: Subdir containing files relation to the project report
   • final_report_BATUGO.html: Html file of the report
   • final_report_BATUGO.qmd: qmd file of the report
   • osa-report-theme.css: css styling for the project report
3. llm_prompt: Subdir containing file with final LLM prompt
   • final_prompt.txt: Text file with final LLM prompt