ChatGPT and DeepSeek in Dermatologic Practice: Do Artificial Intelligence Models Adhere to Psoriasis Treatment Guidelines?

BACKGROUND

The integration of artificial intelligence (AI), particularly large language models (LLMs), into healthcare has rapidly expanded. In dermatology, machine learning is increasingly employed for diagnosis, prognosis, and treatment planning. However, the reliability of AI-generated recommendations, especially in chronic conditions such as psoriasis, remains insufficiently explored.

OBJECTIVE

This study aims to evaluate the alignment of two prominent LLMs—ChatGPT and DeepSeek—with the clinical guidelines for psoriasis treatment issued by the American Academy of Dermatology (AAD).

METHODS

Thirty-one guideline-based questions were formulated using the 2021 AAD-NPF guidelines. Each question was presented to ChatGPT-4 and DeepSeek. Responses were evaluated by two board-certified dermatologists blinded to the model source and rated as either concordant or discordant with the guidelines. Discrepancies were resolved by a third reviewer. Concordance rates were calculated according to the strength of recommendations, and inter-model agreement was assessed using Cohen’s kappa coefficient.

RESULTS

Both ChatGPT and DeepSeek demonstrated an overall concordance rate of 87.1%. While both models fully aligned with strong recommendations, ChatGPT showed higher concordance with moderate recommendations (87.5% vs 81.3%), whereas DeepSeek outperformed ChatGPT in scenarios involving limited recommendations (66.7% vs 33.3%). Inter-model agreement was moderate (κ = 0.43).

CONCLUSION

Although ChatGPT and DeepSeek show promise in aligning with evidence-based dermatologic care, inconsistencies, particularly in cases with limited evidence, highlight the necessity of clinical oversight. AI may serve as a valuable adjunct in psoriasis management; however, its safe integration into practice requires careful validation and context-sensitive application. &nbsp.