Skip to main content
AAO-HNSF 2025 Annual Meeting & OTO EXPO Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Scientific Oral Presentations

Facial Plastic and Reconstructive Surgery

Session: Scientific Oral Presentation 1 - Facial Plastics and Reconstructive Surgery

Artificial Intelligence Assisted Chart Review of CMF Trauma Patients

Sunday, October 12, 2025
1:18 PM - 1:24 PM EDT
Location: 137
Disclosure(s):
Aaron L. Zebolsky, MD:

faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Anas A. Eid, MD:

faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

Introduction: Artificial intelligence and natural language processing has been used to augment and optimize text review in a wide array of different fields. Chart review in retrospective studies is often time and labor intensive which creates a need for further optimization by artificial intelligence.

Methods: Patients that underwent craniomaxillofacial trauma at our large regional trauma center from July 2022 to March 2023 were selected for our study. Notes from the patient visits were obtained and a team of reviewers extracted relevant information from these notes pertaining to the nature of the injury, demographic information, and adverse complications. This information was then used to train a natural language model to automatically extract this information from patient notes. LLAMA 3.1, GEMMA, and MISTRAL large language models created by Meta, Google and MISTRAL AI respectively, were then also used to perform automatic chart review.

Results: The large language models demonstrated high accuracy and specificity in identifying most fracture types, particularly mandibular fractures (accuracy: 96-97%) and nasal bone fractures (accuracy: 94%). While negative predictive values were consistently high, ensuring reliability in ruling out fractures, sensitivity and positive predictive values varied. Orbital floor fractures, for instance, had low sensitivity (35%) despite high specificity (98%), highlighting areas for improvement. These results suggest that LLMs can automate chart reviews efficiently and accurately, though further optimization is needed for complex fracture types.

Conclusions: Large language models can be used to automate chart review in retrospective studies. This can help to save time, increase efficiency and reduce human bias in retrospective studies.

    Presenting Author(s)

  • Travis Jonathan Clarke, MD

    Resident Physician
    University of Tennessee
    Memphis, TN, United States

    Disclosure information not submitted.

    • Co-Author(s)

  • AZ

    Aaron L. Zebolsky, MD

    Resident Physician
    University of Tennessee
    Memphis, TN, United States

    Disclosure(s):

    faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.

  • Mohammad Almomani, MD

    Resident physician
    SSM/ SLU
    Memphis, Tennessee, United States

    Disclosure information not submitted.

    • Presenting Author(s)

  • JK

    Jillian Krebs, N/A

    Student
    The Ohio State University, Dept. of Otolaryngology--Head and Neck Surgery
    Columbus, Ohio, United States

    Disclosure information not submitted.

    • Co-Author(s)

  • Emily R. Baker, MD

    Resident Physician
    Vanderbilt University Medical Center
    Memphis, Tennessee, United States

    Disclosure information not submitted.

    • Senior Author(s)

  • Anas A. Eid, MD

    Assistant Professor
    University of Tennessee, Department of Otolaryngology--Head and Neck Surgery
    Memphis, TN, United States

    Disclosure(s):

    faculty for this accredited education activity has no relationship(s) with ineligible companies to disclose.