Personalizing dietary interventions by predicting individual vulnerability to glucose excursions

Elevated postprandial glucose levels pose a global epidemic and are crucial in cardiometabolic disease management and prevention. A major challenge is inter-individual variability, which limits the effectiveness of population-wide dietary interventions. To develop personalized interventions, it is critical to first predict a person's vulnerability to postprandial glucose excursions, or elevated post-meal glucose relative to a personal baseline, with minimal burden. We examined the feasibility of personalized models to predict future glucose excursions in the daily lives of 69 Chinese adults with type-2 diabetes (M age=61.5; 50% women; 2595 glucose observations). We developed machine learning models, trained on past individual context and meal-based observations, employing low-burden (continuous glucose monitoring) or additional high-burden (manual meal tracking) approaches. Personalized models predicted glucose excursions (F1-score: M=74%; median=78%), with some individuals being more predictable than others. The low burden-models performed better for those with consistent meal patterns and healthier glycemic profiles. Notably, no two individuals shared the same meal and context-based vulnerability predictors. This study is the first to predict individual vulnerability to glucose excursions among a sample of Chinese adults with type-2 diabetes. Findings can help personalize just-in-time-adaptive dietary interventions to unique vulnerability to glucose excursions in daily live, thereby helping improve diabetes management.