Toxic Epidermal Necrolysis in a Critically Ill African American Woman: A Case Report Written With ChatGPT Assistance

Abstract

Stevens-Johnson syndrome (SJS) and toxic epidermal necrolysis (TEN) are life-threatening spectrum diseases in which a medication triggers a mucocutaneous reaction associated with severe necrosis and loss of epidermal integrity. The disease has a high mortality rate that can be assessed by dermatology scoring scales based on an affected total body surface area (TBSA). Sloughing of <10% TBSA is considered SJS, with a mortality of 10%. Sloughing of >30% TBSA is termed TEN, with an increased mortality rate of 25% to 35%. We present a case and management of TEN that involved >30% TBSA in a critically ill African American woman. Identification of the offending agent was difficult due to complicated medication exposure throughout her multi-facility care management. This case conveys the importance of close monitoring of a critically ill patient during a clinical course involving SJS-/TEN-inducing drugs. We also discuss the potential increased risks for SJS/TEN in the African American population due to genetic or epigenetic predispositions to skin conditions. This case report also contributes to increasing skin of color representation in the current literature. Additionally, we discuss the use of Chat Generative Pre-trained Transformer (

Assessing the research landscape and clinical utility of large language models: a scoping review

“…Six articles, including a research letter, two brief reports, a commentary, and two case reports, discussed the benefits of using LLMs like ChatGPT for creating patient and/or discharge notes as well as case reports [ 12 – 17 ]. These benefits included improved efficiency and organization, the potential for standardizing notes, and the identification of grammar and/or lab value errors during extraction [ 12 – 17 ]. However, ChatGPT’s documentation abilities were limited in accuracy based on medical condition complexity [ 12 ].…”

“…For instance, Ali et al found that the accuracy of the generated patient letters differed significantly based on different types of skin cancer [ 12 ]. Additionally, ChatGPT may “hallucinate” false references and can be easily influenced by the rhetoric and assumptions of its users [ 15 , 17 ]. Further research is needed to explore integrating voice-to-text dictation software to ChatGPT documentation workflows, evaluating ChatGPT’s documentation ability in non-English languages, and whether it can be reconfigured to increase accuracy and enhance privacy [ 12 – 17 ].…”

Assessing the research landscape and clinical utility of large language models: A scoping review

“…Six articles, including a research letter, two brief reports, a commentary, and two case reports, discussed the bene ts of using LLMs like ChatGPT for creating patient and/or discharge notes as well as case reports [12][13][14][15][16][17]. These bene ts included improved e ciency and organization, the potential for standardizing notes, and the identi cation of grammar and/or lab value errors during extraction [12][13][14][15][16][17].…”

“…For instance, Ali et al found that the accuracy of the generated patient letters differed signi cantly based on different types of skin cancer [12]. Additionally, ChatGPT may "hallucinate" false references and can be easily in uenced by the rhetoric and assumptions of its users [15,17]. Further research is needed to explore integrating voice-to-text dictation software to ChatGPT documentation work ows, evaluating ChatGPT's documentation ability in non-English languages, and whether it can be recon gured to increase accuracy and enhance privacy [12][13][14][15][16][17].…”

“…Table 1) [15,52]. Only several studies, such as Ali et al's, included results from both a formal, carefully-worded prompt vs results from a less formal, more conversational prompt that may be more realistically seen in a busy clinical setting [12][13][14][15][16][17]. Overall, all studies included human reviewers to evaluate characteristics such as the accuracy and humanness of ChatGPT's responses, sometimes in combination with online tools, to assess the readability of text [4,20,.…”