Predicting Postprandial Glucose Rise From Macronutrients, Clinical Features, and Gut Microbiome Indices

Abstract

Postprandial glucose depends on the meal and the person consuming it. This exploratory secondary analysis used 1,047 isolated meals from 32 CGMacros participants with complete nutrition, clinical, microbiome, and Fitbit records to test whether information beyond macronutrients improved prediction for people excluded from model training. Abbott FreeStyle Libre Pro measurements supplied the 120-minute peak-rise outcome. Under leave-one-subject-out cross-validation, macronutrients produced R² = .162 and a mean absolute error of 27.5 mg/dL; adding premeal glucose, A1c, age, and body mass index raised R² to .350 and reduced the error to 24.0 mg/dL. The clinical-plus-gut model reached R² = .389. A 50-resample participant bootstrap yielded a mean gut increment of .008 with an exploratory interval of [-.038, .054], while activity yielded .002 [-.019, .024]. A1c, carbohydrate, and baseline glucose produced the largest out-of-fold permutation losses. A researcher-defined Bite Balance Score was associated with lower responses in random-intercept models, but replacing carbohydrate, protein, fat, and fiber with it reduced R² from .384 to .295. Clinical measurements contributed the clearest predictive gain. The data did not establish stable incremental value from the eight commercial gut indices or passively measured activity.