Quantifying the Contributions of Food, Glucose, Sleep, and Microbiome Data to Personalized Glycemic Response Prediction

Background Individual glycemic responses to foods vary, and can be predicted using microbiome, activity, and dietary data. However, these data are expensive and invasive to collect, and it is not known how much each modality contributes to accuracy. Objective We aim to quantify the contributions of dietary, sleep, continuous glucose monitor (CGM), and microbiome features for glycemic response prediction; understand how much personal data is required for training; and evaluate how microbiome sample timing impacts model accuracy. Methods We used data from 8,334 participants in The Human Phenotype Project cohort study who provided demographic, anthropometric, dietary, and CGM data. Participants self-reported meals in a dietary tracking app for mean 10.78 days during which they wore CGMs. We trained CatBoost models to predict postprandial glycemic response using 2-hour incremental area under the curve (PPGR), and peak 2-hour postprandial glucose rise (Glu max ). We conducted ablation studies with varied feature combinations to assess the contribution of each data modality. We used three train/test splits (split-by-meal, 5-day personal training, split-by-person) to assess the impact of personal training data. Lastly, we evaluated accuracy as a function of microbiome sample timing (from before meal logs to up to 60 days after). Results The model combining all features performed best, and CGM was the most informative feature. Models trained with more personal data had the best performance (PPGR split-by-meal R=0.731; split-by-person R=0.590) and personal training data had a larger effect on accuracy than microbiome. Microbiome features improved predictions most when collected within 7 days of meal logs, and did not improve performance without personal training data or for samples collected beyond 14 days after meal logs. Conclusions While CGM was the most important feature group, combining it with personal training data and timely microbiome samples led to the most accurate models in our analysis. These findings can help researchers understand the tradeoffs between the time and effort of data collection and how data types impact model performance.