ChatGPT for healthcare providers and patients: Practical implications within dermatology

The launch of ChatGPT-3 (OpenAI) has spurred discussion surrounding use of language learning models (LLMs) as potential clinical tools. ChatGPT is a free online chatbot that generates complex text responses to user-inputted queries using natural language-based artificial intelligence. Recent studies highlight ChatGPT's potential: one demonstrated that cardiologists graded ChatGPT responses to fundamental cardiovascular disease prevention questions as largely appropriate,1Sarraju A. Bruemmer D. Van Iterson E. Cho L. Rodriguez F. Laffin L. Appropriateness of cardiovascular disease prevention recommendations obtained from a popular online chat-based artificial intelligence model.JAMA. 2023; 329: 842-844https://doi.org/10.1001/JAMA.2023.1044Crossref PubMed Scopus (0) Google Scholar while a second determined ChatGPT could pass the standardized United States Medical Licensing exam.2Kung T.H. Cheatham M. Medenilla A. et al.Performance of ChatGPT on USMLE: potential for AI-assisted medical education using large language models.PLoS Digit Heal. 2023; 2e0000198https://doi.org/10.1371/JOURNAL.PDIG.0000198Crossref PubMed Google Scholar These proof-of-concept findings showcase the promising capabilities of LLMs; however, the potential range of ChatGPT's practical applications and limitations are not well-characterized. We aim to summarize ChatGPT technology, suggest 5 future domains of specific use cases pertinent to dermatologists and patients, and discuss limitations to consider prior to clinical deployment. ChatGPT is powered by an LLM based upon a large corpus of online text, which generates novel word sequences (eg, sentences) that read plausibly in English (dermatology use cases in Table I; representative input/output combinations in Supplementary Table I, available via Mendeley at https://doi.org/10.17632/dgrfs639f8.1).Table IFive domains of ChatGPT use cases for healthcare providers and patients in the dermatology settingDomain and ExamplesRepresentative ChatGPT Input∗Due to space limitations, the corresponding ChatGPT output to each inputted prompt listed in this table can be found in Supplementary Table I.For the healthcare providerStreamline clinical and administrative tasksDraft an in-basket response to the following: A 48yo F with psoriasis sent the following message: “Hi doctor, I noticed that my recent labs came back with elevated LFTs. I am wondering what this means and whether I should continue taking my methotrexate?”Enhance patient educationIn one paragraph, explain squamous cell carcinoma and write after-visit care instructions for an excision with 5 mm margins on the arm, tailored to an 8th grade reading level.Supplement medical educationSummarize for a 2nd-year medical student audience: Insert textbook section describing the types of topical steroidsAid clinical researchInput clinical trial protocol. If my patient experienced itchy eyes due to dupilumab, is this adverse event categorized as serious, moderate, or unrelated? Do they need to stop participating in the study?For the patientIncrease healthcare literacyIf I have psoriasis and require biologic medication for my disease, what should I consider when choosing my insurance plan?∗ Due to space limitations, the corresponding ChatGPT output to each inputted prompt listed in this table can be found in Supplementary Table I. Open table in a new tab First, ChatGPT has the potential to automate administrative tasks (eg, drafting prior authorizations) that otherwise increase inefficiency and financial burden, while negatively impacting patient outcomes.3Carlisle R.P. Flint N.D. Hopkins Z.H. Eliason M.J. Duffin K.C. Secrest A.M. Administrative burden and costs of prior authorizations in a dermatology department.JAMA Dermatology. 2020; 156: 1https://doi.org/10.1001/JAMADERMATOL.2020.1852Crossref PubMed Google Scholar Clinically, providers can draft in-basket responses or query de-identified patient notes with specific questions (eg, result of previously tested immunostaining markers).3Carlisle R.P. Flint N.D. Hopkins Z.H. Eliason M.J. Duffin K.C. Secrest A.M. Administrative burden and costs of prior authorizations in a dermatology department.JAMA Dermatology. 2020; 156: 1https://doi.org/10.1001/JAMADERMATOL.2020.1852Crossref PubMed Google Scholar Second, ChatGPT can enhance patient education by assisting with de novo generation of after-visit summaries or simplifying complex diagnoses. Beyond supporting clinical functions and patient education, ChatGPT can enhance both medical education and research. ChatGPT may be used in parallel with search engines, including summarizing textbook chapters for students, synthesizing peer-reviewed articles for researchers, and interpreting text-heavy study protocols for clinical research coordinators. Finally, ChatGPT can improve health literacy by assisting patients with questions about insurance coverage of specific services, selection of an insurance plan, or clinical research participation. However, practical and ethical challenges require consideration prior to widespread clinical adoption, the greatest of which is realistic-sounding presentation of false information (a phenomenon termed “artificial hallucination”), leading to patient safety concerns. For example, ChatGPT can generate artificial hallucinations of nonexistent reference citations with broken hyperlinks that appear plausible based on the title and co-authors. ChatGPT output is highly dependent on user input and may change with repeat questioning. Within specialist fields, ChatGPT can appear both confident and wrong. For instance, one publication found deficits in situational awareness, inference, and consistency when applying ChatGPT to complex infectious disease scenarios.4Howard A. Hope W. Gerada A. ChatGPT and antimicrobial advice: the end of the consulting infection doctor?.Lancet Infect Dis. 2023; 23: 405-406https://doi.org/10.1016/S1473-3099(23)00113-5Abstract Full Text Full Text PDF PubMed Scopus (39) Google Scholar Thus, clinicians must closely proofread any output. Furthermore, the ChatGPT dataset exclude events after 2021 and may include harmful social biases toward disadvantaged groups present within the training text.5Weidinger L. Mellor J. Rauh M. et al.Ethical and social risks of harm from language models.arXiv. 2021; https://doi.org/10.48550/arxiv.2112.04359Crossref Google Scholar Qualitative assessment models for LLMs have been proposed and the use of ChatGPT in dermatology-specific situations should be validated systematically,4Howard A. Hope W. Gerada A. ChatGPT and antimicrobial advice: the end of the consulting infection doctor?.Lancet Infect Dis. 2023; 23: 405-406https://doi.org/10.1016/S1473-3099(23)00113-5Abstract Full Text Full Text PDF PubMed Scopus (39) Google Scholar with eventual publication of guidelines surrounding appropriate usage for dermatologists. ChatGPT is poised to increasingly permeate clinical practice and teaching. Future ChatGPT versions and improved understanding of validated use cases in the clinical setting will hopefully enable dermatologists to maximize their time, expand their capabilities, and improve patient care. None disclosed.