ChatGPT may potentially improve patch testing efficiency for allergic contact dermatitis

Key takeaways:

• Clinicians and ChatGPT-4 disagreed in 17.9% of cases.
• ChatGPT-4 identified whether a list of products contained a particular allergen but named fewer potential sources for each allergen.

SAN DIEGO — While it did not perform at the level of clinicians, ChatGPT could be used as a tool to improve patch testing efficiency for allergic contact dermatitis, according to a poster presentation here.

“Artificial intelligence (AI) could serve to enhance the patient experience in patch testing,” Lauren Passby, BMBCH, PGCert, a specialist dermatology registrar at University Hospitals Birmingham NHS Foundation Trust, and colleagues wrote in a poster presented at the American Academy of Dermatology Annual Meeting. “This study assessed whether ChatGPT-4 can correctly interpret patch testing results and provide appropriate patient education and counseling.”

In their study, 67 patients (47 women; mean age, 44 years) were patch tested for allergic contact dermatitis. At 96 hours, 38 patients had positive patch test reactions, with the most common positive haptens being methylchloroisothiazolinone, methylisothiazolinone and nickel.

Following testing, the researchers instructed ChatGPT to counsel patients on the relevance of their results. ChatGPT was also asked to identify alternative chemical names and potential sources for the 80 haptens that comprise the North American 80 Comprehensive Series (NAC-80), a list that contains the top allergens that may cause an individual to develop contact dermatitis.

Results showed that clinicians and ChatGPT-4 disagreed in 17.9% of cases. While clinicians had the ability to extract detailed information about patients’ past encounters with allergens, ChatGPT-4 occasionally attributed patients’ reactions to allergens that they did not encounter.

When compared with the Patient Information Sheets for NAC-80, the AI software listed fewer potential sources for each allergen (P = .003) as well as fewer alternative chemical names (P < .00001).

“Whilst ChatGPT-4 can analyze written recorded patch testing results, it has neither the comprehensiveness of the manufacturer’s patient information sheet, nor the clinical expertise to accurately ascribe past, current or future relevance in all cases,” the authors wrote.

“It can, however, identify whether a list of products contain a particular hapten, and can identify the importance of patient immunosuppression impacting upon patch testing results, showing how artificial intelligence may one day assist efficiency in the counseling of patients undergoing patch testing,” the authors concluded.