A multivariate mixed-effects machine-learning model applied to 50,463 free-living meals from 992 non-diabetic adults successfully isolated stable, individual-specific postprandial glucose response traits amid substantial meal-to-meal noise. The analysis identified two dominant axes of between-person variation: one reflecting overall magnitude of glucose excursions and a second capturing carbohydrate responsiveness with lower late-phase glucose elevations. Critically, the dominant-axis score derived from just 3 days of continuous glucose monitoring closely approximated scores from 14-day monitoring windows, and both correlated with responses to held-out standardized meals independent of age, sex, BMI, and baseline glucose metrics.
This finding matters considerably for precision nutrition research, where CGM-based metabolic phenotyping has been theoretically promising but practically expensive and burdensome. If a 3-day monitoring window is sufficient to reliably fingerprint an individual's glucose response profile, clinical and research protocols become dramatically more scalable. The two identified axes also hint at mechanistically distinct physiologies — glycemic amplitude versus carbohydrate-specific sensitivity — potentially informing stratified dietary interventions beyond simple glycemic index approaches.
Important caveats apply. The cohort excludes diagnosed diabetics, limiting generalizability to metabolically compromised populations. The framework is observational and cannot establish causal dietary recommendations. The 'candidate coordinate for stratification' language remains speculative without prospective intervention trials. As a preprint posted on medRxiv and not yet peer-reviewed, these findings require independent validation before influencing clinical practice. Nonetheless, the analytical approach represents a meaningful methodological advance for population-scale metabolic phenotyping.
