Advances, challenges, and future directions in precision nutrition by machine learning-based prediction of glycemic responses to foods: a narrative review.

Recent advances in multiomics technologies, continuous glucose monitoring, and machine learning (ML) enable precision nutrition by prediction of glycemic responses to foods. To date, no narrative reviews have focused on advances, challenges, and future directions in ML-based prediction of glycemic responses. This narrative review aimed to summarize the existing research on ML-based prediction of postprandial glycemic responses, highlight the challenges in prediction of glycemic responses, and offer directions for future research. We searched PubMed, Embase, Cumulative Index to Nursing and Allied Health Literature (CINAHL), Cochrane Library, Institute of Electrical and Electronics Engineers (IEEE) Xplore, and Association for Computing Machinery (ACM) Digital Library, from inception to 21 November, 2025. Studies that developed/updated/conducted external validation of an ML model and predicted the outcomes of interest (i.e., glycemic responses, peak glucose, and glucose excursions) were included. A total of 13,471 records and 2 reports identified 26 studies for inclusion. Although some studies used traditional linear or deep learning models, most existing prediction models were tree-based ML models and developed for healthy people and those with prediabetes or diabetes. Most existing models included anthropometry, biomarkers, and microbiota as personal features and nutrients/food groups as food features. However, some key food features such as food processing were overlooked. Some clinical trials reported that ML-based personalized dietary recommendations failed to yield favorable and clinically significant glycemic outcomes compared with dietary guidelines. There are challenges in ethics and compliance of algorithms, reliability of measurement instruments, prediction of multiple metabolic outcomes, trade-offs in raw data collection and feature selection, and the balance of performance metrics and clinical benefits. Most existing ML-based prediction models for postprandial glycemic responses are tree-based and incorporate personal and food features. Research on prediction of glycemic responses faces challenges that require careful consideration during study design. To improve transparency, reproducibility, and quality of future research, this review provides a minimum reporting framework with checklist items and recommendations.

Copyright © 2026 American Society for Nutrition. Published by Elsevier Inc. All rights reserved.