A Comparative Study of Large Language Model Approaches in the Nutrition Therapy of Ghanaians With Sickle Cell Disease.

Abstract

Modern Digital Diet Managers offer various artificial intelligence-driven services to help people with special dietary needs or diseases make smarter decisions regarding their food intake and overall nutrition. Computer Vision and Recommendation systems are among the popular artificial intelligence methodologies used in managing the diet of Diabetics, Hypertensives, and Cancer patients; however, this technology is yet to be adopted in the nutrition therapy of Sickle Cell Disease(SCD), especially in Africa, where 80% of SCD cases are located, and 1 in 4 people suffer from the disease. [1] This study conducted an experiment aimed at comparing and contrasting sickle cell dietary advice reports generated by two distinct artificial intelligence systems, System B: A Large Language Model with prompt data obtained from a Knowledge Graph, and System A: A Large Language model augmented with a vector database. Speed was measured using the time difference between sending a request and receiving a response, while the accuracy of the reports was validated by assessing the number of factually accurate responses and mathematically accurate responses. Although System A performed much faster than System B by an average of thirty seconds, accuracy is of greater importance than speed in non-emergency healthcare. The research found that System B, the system that incorporates a knowledge graph for information retrieval, performed much better at giving factually accurate and mathematically accurate responses, than System A, which made use of a vector database.