Related Experiment Video
Updated: Aug 14, 2026
Dermoscopy Aids in the Diagnosis of Discoid Lupus Erythematosus
Published on: May 16, 2025
AI Simplification of Dermatopathology Reports for Patients: Basic Versus Prompt-Engineered Approaches
William J Nahm1, Arlene M Ruiz de Luzuriaga2, Goranit Sakunchotpanit3
1Dr. Phillip Frost Department of Dermatology & Cutaneous Surgery, University of Miami Miller School of Medicine, Miami, Florida, USA.
Background:
Patients struggle to comprehend dermatopathology reports. As artificial intelligence (AI) tools become more accessible, patients may use them to interpret reports; however, optimal approaches remain unexplored.
Objective:
Evaluate whether prompt-engineered AI simplification of dermatopathology reports improves factualness, completeness, and reduces potential harm compared to basic AI usage.
Methods:
Survey-based study (January-April 2025) of 52 US dermatology and dermatopathology professionals (70.3% response rate). Six fictitious dermatopathology reports were simplified using: (1) Basic ChatGPT-4.0 with simple prompt and (2) Custom "DermDecoder" GPT with structured 489-word prompt. Participants rated reports on 3-point Likert scales for factualness, completeness, and potential harm, with free-text responses analyzed thematically.
Results:
Mean ratings ranged from 1.27 to 1.63 (factualness/completeness) and 1.31-1.83 (harmfulness), indicating "Agree" to "Mostly Agree" or "Completely Harmless" to "Mostly Harmless." DermDecoder performed significantly worse for completeness in psoriasis (t = -2.79, p = 0.007) and harmfulness in molluscum contagiosum (p = 0.049) and melanoma in situ (p = 0.048). Free-text analysis revealed Basic Prompt preserved details but lacked clinical context, while DermDecoder provided generic education disconnected from pathological findings.
Limitations:
Fictitious reports, small sample, evolving AI capabilities, and absence of patient perspectives.
Conclusion:
Prompt engineering offered no advantage over basic AI usage in balancing professional accuracy with patient accessibility, necessitating human-in-the-loop oversight for AI-generated explanations.
