Abstract:Abstract Patient engagement is evolving with the advent of digital tools such as interactive websites and artificial intelligence (AI), yet significant barriers remain in the technical expertise required to create tools that benefit patients. We explore how large language models (LLMs), a type of AI that can facilitate and accelerate programming tasks, can enable clinicians and researchers to create interactive web-based patient engagement tools without the need for a dedicated web developer. We explore this possibility by developing, with the help of an LLM, an interactive skin self-assessment web tool. We used GPT4, a state-of-the-art LLM, as a web development assistant, and created an interactive JavaScript web page that guides patients through the Fitzpatrick skin phototype questionnaire. This involved building a minimal version first, and iteratively improving it. The web page initially asked the user the first two questions of the assessment, calculated a score, and returned this to the user. Once the code written by GPT4 was tested and shown to work, the tool was gradually improved to include all questions in the Fitzpatrick scale, and the result was modified to provide not only the total score, but also the result category. GPT4 allowed us, with no experience writing interactive JavaScript web pages, to develop and launch a web page that guides patients through a Fitzpatrick scale assessment. When GPT4 was asked to expand the questionnaire from two questions to the whole test, it gladly did so; however, the questions were not accurate. It then had to be given the specific wording of the questions and answers for the test, which it incorporated into functioning JavaScript code. This had to be performed in stages, adding two questions at a time, due to context limitations at the time. When this introduced issues, such as the code breaking or the website style looking inconsistent, GPT4 was able to fix these issues when they were described to it. Since the web page’s launch in March 2023, the web page has been viewed 1875 times by 1419 unique users. This study demonstrates one way that LLMs can be used to enhance patient engagement in dermatology, by facilitating the creation of interactive skin assessments with utility in many dermatology settings. Hallucination remains a problem, and any created tools must be thoroughly verified for factuality. Although no authors had any experience writing interactive JavaScript, they have coding experience elsewhere, and future work should assess whether similar results can be achieved by a broader user base. Two authors received funding from a commercial organization for this work (A & Z Dermatology Clinic Ltd).

OIPFST: AI-based Fitzpatrick skin tone labelling utilizing skin lesions

Revisiting Skin Tone Fairness in Dermatological Lesion Classification

Analysis of Manual and Automated Skin Tone Assignments for Face Recognition Applications

Estimating Skin Tone and Effects on Classification Performance in Dermatology Datasets

PatchAlign:Fair and Accurate Skin Disease Image Classification by Alignment with Clinical Labels

Identification of Skin Lesions by Using Single-Step Multiframe Detector

Skin Tone Analysis for Representation in Educational Materials (STAR-ED) using machine learning

Reliability and Validity of Image-Based and Self-Reported Skin Phenotype Metrics

Image‐based remote evaluation of PASI scores with psoriasis by the YOLO‐v4 algorithm

Validity of facial skin analysis pore detection: A comparative analysis

Optimizing Skin Lesion Classification via Multimodal Data and Auxiliary Task Integration

A comparative study of an advanced skin imaging system in diagnosing facial pigmentary and inflammatory conditions

Skin Tone Analysis Through Skin Tone Map Generation With Optical Approach and Deep Learning

A New Approach for Automatic Segmentation and Evaluation of Pigmentation Lesion by using Active Contour Model and Speeded Up Robust Features

A protocol for annotation of total body photography for machine learning to analyze skin phenotype and lesion classification

A Pilot Study to Assess the Reliability of Digital Image-Based PASI Scores Across Patient Skin Tones and Provider Training Levels.

SkinLiTE: Lightweight Supervised Contrastive Learning Model for Enhanced Skin Lesion Detection and Disease Typification in Dermoscopic Images

Artificial Intelligence–Based Psoriasis Severity Assessment: Real-world Study and Application

P034 Using artificial intelligence to bridge the technical gap in dermatology: the role of large language models in creating an interactive Fitzpatrick skin phototype assessment

Identification of Skin Lesions by Snapshot Hyperspectral Imaging