Comparing ChatGPT with healthcare provider responses to home parenteral nutrition questions.
Journal:
Nutrition in clinical practice : official publication of the American Society for Parenteral and Enteral Nutrition
Published Date:
Nov 9, 2025
Abstract
BACKGROUND: Patients receiving home parenteral nutrition (HPN) face complex challenges and increasingly seek support online, including generative artificial intelligence (AI) platforms like ChatGPT. This survey study compared ChatGPT with clinician responses to common HPN-related questions. METHODS: Responses to 20 HPN-related questions spanning five content themes were generated by ChatGPT and provided by HPN expert clinicians. In a blinded online survey, practicing clinicians (study participants) rated each response on a five-point scale (1 = excellent; 5 = very poor) for accuracy, appropriateness, and empathy and selected their overall preferred response. RESULTS: Among 23 participants (73.9% registered dietitians; mean HPN experience: 14.0 years), ChatGPT's responses were rated more favorably for accuracy (median [IQR] = 1.80 [0.79] vs 2.15 [0.62], P = 0.003), appropriateness (1.80 [0.70] vs 2.15 [0.53], P = 0.013), and empathy (1.95 [0.66] vs 2.25 [0.65], P = 0.007). Participants preferred ChatGPT responses in 48.5% of cases, clinician responses in 33.9%, and had no clear preference in 17.6%. ChatGPT outperformed clinicians across content themes for "best practices, care, and safety of HPN use/infection risk" and scored more favorably for empathy in "symptoms" and for accuracy and appropriateness in "lifestyle stressors." Clinicians scored more favorably for appropriateness in "biochemical test concerns." CONCLUSION: ChatGPT may support HPN care and patient education, particularly for broad medical and lifestyle topics. However, complex clinical issues require medical expertise. Further research is needed to guide the safe integration of AI into clinical practice and patient care.
Authors
Keywords
No keywords available for this article.
