AI Simplification of Dermatopathology Reports for Patients: Basic Versus Prompt-Engineered Approaches

Area of Science:

• Dermatology
• Medical Informatics
• Artificial Intelligence

Background:

• Patients face challenges understanding complex dermatopathology reports.
• Increasing accessibility of AI tools prompts exploration of their use in patient report interpretation.
• Optimal AI strategies for simplifying medical reports remain uninvestigated.

Purpose of the Study:

• To assess if prompt-engineered AI simplification enhances factualness and completeness of dermatopathology reports.
• To determine if advanced AI prompts reduce potential harm compared to basic AI usage.
• To compare the efficacy of different AI prompting techniques for patient-facing medical information.

Main Methods:

• A survey-based study involving 52 US dermatology and dermatopathology professionals.
• Six fictitious dermatopathology reports were simplified using basic AI (ChatGPT-4.0) and a custom AI (DermDecoder GPT).
• Participants rated simplified reports on factualness, completeness, and potential harm using Likert scales.

Main Results:

• Both AI methods yielded ratings indicating general agreement with factualness/completeness and low perceived harm.
• The custom AI (DermDecoder) performed significantly worse in completeness for psoriasis reports.
• The custom AI also showed increased harmfulness ratings for molluscum contagiosum and melanoma in situ reports.

Conclusions:

• Prompt engineering did not offer a significant advantage over basic AI for simplifying dermatopathology reports for patients.
• AI-generated explanations require human-in-the-loop oversight to ensure accuracy and patient understanding.
• Limitations include the use of fictitious reports, a small professional sample, and evolving AI technology.