EP1036 - Can Artificial Intelligence (AI) Platforms Improve the Readability of Surgical Consent Forms?

ePoster Presenter(s)

• 

  Ann Pham

  Medical Student
  University of Miami
  University of Miami Miller School of Medicine
  Miami, FL, United States

  Disclosure information not submitted.

Author(s)

• GH

  Gianina C. Hernandez Marquez, MD

  T32 Postdoctoral Research Fellow
  UNIVERSITY OF ALABAMA AT BIRMINGHAM
  Birmingham, Alabama, United States

  Disclosure information not submitted.
• CH

  Connor Hwang, n/a

  Undergraduate
  University of Alabama at Birmingham
  Goleta, California, United States

  Disclosure information not submitted.
• 

  Alizeh Abbas, MD (she/her/hers)

  General Surgery Resident
  University of Alabama at Birmingham
  University of Alabama at Birmingham (UAB)
  Birmingham, Alabama, United States

  Disclosure information not submitted.
• CA

  Chandler A. Annesi, MD (she/her/hers)

  General Surgery Resident
  University of Alabama at Birmingham
  University of Alabama at Birmingham
  Birmingham, AL, United States

  Disclosure information not submitted.
• LW

  Lauren N. Wood, MSPH

  Statistician
  UAB
  Birmingham, Alabama, United States

  Disclosure information not submitted.
• AH

  Alfonsus Adrian H. Harsono, MD, MSPH

  Postdoctoral Fellow
  University of Alabama at Birmingham
  Birmingham, Alabama, United States

  Disclosure information not submitted.
• MR

  Michael Rubyan, PhD, MPH

  Assistant Professor
  University of Michigan School of Public Health
  Ann Arbor, Michigan, United States

  Disclosure information not submitted.
• 

  Robert Hollis, MD, MSPH

  Assistant Professor
  University of Alabama at Birmingham
  University of Alabama at Birmingham
  Birmingham, AL, United States

  Disclosure information not submitted.

Senior Author(s)

• 

  Daniel I. Chu, MD, MSPH (he/him/his)

  Professor and Division Director
  University of Alabama at Birmingham
  University of Alabama at Birmingham
  Birmingham, AL, United States

  All of the relevant financial relationships listed below have been mitigated.
  This individual has no financial relationships with ineligible companies.

Purpose/Background: Understanding surgical consent forms empowers patients to make informed decisions about care, including weighing risks, benefits, and alternatives. Inadequate comprehension can result in poor risk recall, procedural misunderstanding, and increased vulnerability to harm. Standard consent forms can be difficult to comprehend, especially for patients with limited health literacy. Artificial Intelligence (AI) platforms may provide a scalable solution to improve readability and understandability of surgical consent forms. We aim to evaluate publicly accessible, free AI platforms' ability to adapt surgical consent forms with improved readability and understandability.

Methods/Interventions: Surgical consent forms (n=5) from a tertiary health center were collected and uploaded to four AI platforms (Chat GPT 4.0, Gemini, Copilot, and DeepSeek) as templates to generate more health literate forms. A prompt detailing the writing standards and readability criteria meeting 6th grade reading level or lower was utilized. The primary outcome assessed was readability, using Flesch-Kincaid Grade Level (FKGL) and Simple Measure of Gobbledygook (SMOG) scores. Understandability and Actionability were measured with Patient Education Materials Assessment Tool (PEMAT) and graded by three reviewers. Results for AI-generated consent forms were compared to the template consent form via paired t-tests for readability (FKGL and SMOG), understandability (PEMAT), and actionability (PEMAT).

Results/Outcomes: All five original templates scored 16-17 on the SMOG scale (higher score indicates higher level of education required for text comprehension) and achieved college graduate reading level by FKGL criteria. AI-generated forms demonstrated lower reading levels overall, both with SMOG and FKLG (p < 0.001). ChatGPT 4.0 forms scored 9.7-12.1 (SMOG) and 10th-12th grade (FKGL). Gemini forms scored 10.5-11.9 (SMOG), with two at 8th grade and three at 10th grade (FKGL) out of the five. Copilot forms scored 9.6-12.0 (SMOG), with three at 8th grade and two at 10th grade (FKGL); and DeepSeek forms scored 10.1-11.6 (SMOG), with two at 8th grade, two at 10th grade, and one at college level (FKGL) (Figure 1). PEMAT evaluation revealed no significant differences in understandability and actionability between templates and AI-generated forms across all four platforms.

Conclusion/Discussion: Despite AI platform's ability to generate surgical consent forms with improved reading levels measured by SMOG and FKGL scales, no platform achieved the targeted 6th grade reading level specified in the prompt. They also failed to enhance understandability or actionability beyond template performance. While AI shows promise for improving patient consent form accessibility with Copilot generating the highest proportion of ideal reading-level among tested platforms, further research is needed to optimize AI applications for generating better patient-friendly consent materials.

Comparison of (A) SMOG scores and (B) FKGL levels between the template and AI-generated forms.