Evaluation of Artificial Intelligence Models for Nutritional Symptom Management in Breast Cancer Patients Undergoing Chemotherapy.

Abstract

The purpose of the study is to determine if artificial intelligence (AI) models could provide dietary recommendations to manage chemotherapy-induced nutritional symptoms in breast cancer (BC) patients that are on comparable levels with the American Cancer Society, National Cancer Institute and World Cancer Research Fund guidelines which were used as evidence-based recommendation. The AI models-ChatGPT, ChatGPT 4.0, Gemini, Gemini Advanced, Copilot, and Copilot Pro-were evaluated based on their adherence to the guidelines. Specific queries were posed to each model, and the generated responses were rated by two experienced dietitians using a 5-point likert scale. Gemini provided the most adherent recommendations for metallic taste in mouth, while Gemini Advanced excelled in managing dehydration. ChatGPT and Gemini Advanced were the most effective in addressing heartburn, and Copilot Pro consistently showed the lowest performance across most symptoms. Overall, ChatGPT 4.0 attained the highest total score, followed by Copilot and Gemini, indicating a general trend where certain models were better suited for specific symptoms. Various AI models (e.g. ChatGPT 4.0) show potential in addressing certain chemotherapy-induced symptoms, but they do not consistently align with evidence-based recommendations. Improvement of AI models is necessary for their adherence to evidence-based recommendations.