Two people, the same bread, two different blood sugar responses
800 adults. 46,000 meals. Continuous glucose monitors. The result dismantled the idea that a food has one effect, and explained why the diet that works for your friend does not work for you.
For decades, nutrition ran on a table. Each food had a glycemic index, that index was a property of the food, and the recommendation followed from there: this bread raises your blood sugar, that fruit raises it less, arrange your plate accordingly.
In 2015 a group from the Weizmann Institute published a study in Cell that turned that logic inside out.
The design
The Personalized Nutrition Project fitted continuous glucose monitors to 800 adults without diabetes and recorded what they ate over a week: food, exercise and sleep logged. In total, 46,898 meals. On top of the free-choice meals, every participant ate some identical standardized meals.
The finding
Postprandial glycemic responses — what happens to your blood sugar after you eat — turned out to be highly variable between individuals, even when they ate exactly the same standardized meal.
Read that again, because it is the sentence that matters: the same meal, in two people, produces two different curves. Not similar with noise. Different.
That means the glycemic index is not a property of the food. It is a property of the relationship between a food and one specific person. And a table cannot contain that, by definition.
The glycemic index does not describe a food. It describes an encounter.
The second part, which is the one people forget
The study did not stop at the observation. The authors built an algorithm integrating clinical and gut microbiome data, and with it they predicted personalized glycemic responses to real, complex meals. The prediction was also validated in an independently collected cohort.
And then they did what really closes the loop: they designed personalized dietary interventions based on that algorithm. Those interventions produced lower glycemic responses, accompanied by consistent changes in the gut microbiota.
That is, not only "each person responds differently" — which would be interesting but useless. Also: that difference can be predicted, and by predicting it you can intervene better.
What follows from this for a nutrition plan
One very concrete and rather uncomfortable thing for the diet industry follows: a plan that does not incorporate your data is guessing. It may guess well — many general recommendations are reasonable — but it is still a bet on a population average you may not be part of.
- Removing a food "because it raises blood sugar" assumes it raises yours, which this study shows cannot be assumed.
- Two people with the same weight, the same age and the same goal may need different plans for reasons that only show up when you measure.
- The microbiome enters the equation: it was one of the variables with which the algorithm improved its prediction.
- And the operational conclusion: the correct sequence is measure, then predict, then indicate. Not the other way around.
And what the study does not say
It does not say that lowering the blood sugar peak after a meal is, in itself, the goal of a healthy life. It is an intermediate marker: it moved, and it moved in the expected direction, with a predicted intervention. How much of that translates into less disease twenty years out is another question, with other evidence and other timescales.
But as an argument against the template diet, it is hard to refute. And it is exactly why at PROFIT7 the nutrition plan does not exist before the Precision Assessment: there would be nothing to build it with.
Where the evidence stands
What the evidence supports
- That postprandial glycemic responses are highly variable between people, even in response to identical standardized meals.
- That the study was done with 800 adults without diabetes and 46,898 meals recorded with continuous glucose monitoring.
- That an algorithm integrating clinical and microbiome data predicted those personalized responses, and that the prediction was validated in an independent cohort.
- That personalized dietary interventions based on that algorithm produced lower glycemic responses, with consistent changes in the microbiota.
What it does not yet
- The postprandial glycemic response is an intermediate marker: lowering it does not, on its own, equal less disease in the long run.
- Follow-up was one week; it says nothing about adherence or about effects sustained over time.
- Prediction algorithms are specific to the cohorts they are trained on and are not interchangeable between populations.
- It does not make continuous glucose monitoring necessary for everyone who wants to eat better.
Sources
- Zeevi et al. — Personalized Nutrition by Prediction of Glycemic Responses The original study in Cell (2015), full text.
- Personalized Nutrition by Prediction of Glycemic Responses — PDF PDF version hosted by the research group itself.
- Microbiome-based approaches to personalized nutrition: from gut health to disease prevention Later review placing this work as a turning point for the field.
This article is general information. It does not replace a medical assessment, and none of its statements should be read as an indication for treatment.
Your data, not an average.
Everything explained here only means something when it is applied to one specific person. That is the starting point.
See the Precision Assessment
