P686 - PEGASSIST: A CHATGPT-BASED CHATBOT FOR POST-PEG EDUCATION IN OLDER ADULTS

P686

PEGASSIST: A CHATGPT-BASED CHATBOT FOR POST-PEG EDUCATION IN OLDER ADULTS

D. OZATA^1,*, K. CINGAR ALPAY¹, G. K. AVLAGI¹, S. BILGIN¹, U. DURAK¹, I. H. OZATA², S. AVCI¹, A. DOVENTAS¹, U. D. ERDINCLER¹

¹GERIATRICS, ISTANBUL UNIVERSITY CERRAHPASA-CERRAHPAŞA FACULTY OF MEDICINE, ²GENERAL SURGERY, KOC UNIVERSITY SCHOOL OF MEDICINE, ISTANBUL, Türkiye

Rationale: Percutaneous endoscopic gastrostomy (PEG) is widely used for long-term enteral nutrition in older patients with neurological or structural impairments. Post-discharge, patients and caregivers often face uncertainties about PEG care. Artificial intelligence-based chatbots may offer accessible, consistent information to support patient education. This study aimed to evaluate the reliability and quality of PEGAssist, a ChatGPT-4o-based chatbot developed to address frequently asked questions (FAQs) following PEG placement.

Methods: PEGAssist was designed using the ChatGPT-4o model (OpenAI), informed by updated PEG guidelines and relevant clinical literature. Twenty representative FAQs were selected based on Google search trends and expert opinion. Responses generated by the chatbot were evaluated by a multidisciplinary expert panel (two geriatricians, one general surgeon, one geriatric nurse) using a 5-point Likert scale on three dimensions: depth, clarity, and scientific accuracy. Inter-rater agreement was assessed using the Intraclass Correlation Coefficient (ICC(2,k)).

Results: Average scores for the chatbot responses were: depth 4.44 (SD=0.50), clarity 4.85 (SD=0.36), and scientific accuracy 4.78 (SD=0.42). Clarity was rated highest, while depth received relatively lower scores. The ICC analysis demonstrated strong inter-rater agreement (ICC=0.80), indicating high reliability in expert ratings.

Conclusion: PEGAssist demonstrated high performance in providing clear and accurate information on PEG-related topics. However, the relative underperformance in content depth suggests the need for more personalized question formats. AI-driven chatbots may serve as effective tools in post-PEG care, supporting both patients and caregivers. Further studies are required to validate its usability and clinical impact in real-world settings.

Disclosure of Interest: None declared