Zeevi · Cell 2015 · Prospective cohort model derivation and validation with a blinded randomized controlled trial · n=900

Personalized Nutrition by Prediction of Glycemic Responses.

Cited 2901 times in the scientific literature.

Level 2 - randomized trial

Blinded randomized controlled trial component alongside prospective cohort derivation and validation

PubMed 26590418 · doi:10.1016/j.cell.2015.11.001 · record verified 2026-08-30

What was done

Researchers tracked week-long continuous glucose levels in an 800-person cohort across 46,898 meals. Using blood parameters, dietary habits, anthropometrics, physical activity, and gut microbiota, they trained a machine-learning algorithm to predict personalized postprandial glycemic responses. The model was validated in an independent 100-person cohort. Finally, a blinded randomized controlled dietary intervention based on the algorithm was conducted to evaluate postprandial glucose responses and gut microbiota changes.

What was found

The authors observed high interpersonal variability in glycemic response to identical meals. The machine-learning algorithm accurately predicted personalized postprandial glycemic responses in both the discovery cohort and the 100-person validation cohort. In the blinded randomized controlled trial, the algorithm-based personalized diet significantly lowered postprandial glycemic responses and induced consistent alterations in gut microbiota configuration. The abstract reports no numerical values, effect sizes, or p-values.

Why it matters

This work demonstrates that universal dietary advice overlooks substantial individual glycemic variation, providing proof-of-concept for personalized, algorithm-driven nutrition to manage blood glucose.

Limits

The abstract provides no numerical data, effect sizes, or statistical confidence intervals. The primary monitoring period was short (one week), the specific sample size of the randomized intervention sub-group was not stated, and long-term metabolic outcomes were not evaluated.

Cited by