Artificial intelligence for TB education and counselling: a modified Delphi consensus

Résumé fourni par la source

BACKGROUND Optimal TB care and control requires improved health education and effective patient counselling. Effective counselling is a lengthy process that requires addressing all questions related to TB transmission, risk, the disease process, prevention, diagnosis, and treatment. We evaluated the quality of artificial intelligence (AI) chatbot responses, compared to global expert opinion. METHODS We configured an AI chatbot (TB counselling assistant [TBCA]) based on GPT-4. It was designed to draw information from reputable TB guidelines and tasked to provide educational responses. We tested the TBCA on 39 questions frequently asked by people with TB and their caregivers. Responses were appraised for quality by global TB experts using a modified Delphi consensus approach. RESULTS Ninety-four experts were invited to participate, of whom 91 (96.8%) participated. Overall, the TBCA provided accurate answers to questions in all relevant domains, including epidemiology, clinical presentation, prevention, diagnosis, and treatment, as well as special considerations in vulnerable populations, while citing its information sources. When appropriate, it consistently directed the individual to seek advice from an appropriate health care provider. At times, the information provided was out of date (e.g., the definition of multidrug-resistant TB). It also struggled to distinguish diagnostic tests for TB infection and disease. CONCLUSION Large language models can provide accurate responses to general TB questions, but the information provided may be out of date, or lack context. Although the advice provided was generally safe and helpful, health care professionals remain crucial to ensure advice is up to date and appropriate to the individual's unique circumstances. .

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.