The Glycemic Index of Food Varies Among Individuals

The Glycemic Index of Food Varies Among Individuals

From Lennerz et al., 2013

After eating a meal, blood sugar (glucose) levels rise. Through the interaction of hormones (i.e., insulin), this sugar is then shuttled into cells for energy or, if in excess, converted to fat for long-term storage. Excessively high levels of blood glucose are now thought to be a major contributory factor towards type 2 diabetes and obesity. Consequently, there has been more interest in understanding the extent to which foods raise blood glucose levels. A common approach is the “Glycemic Index” which is a scale that ranks individual foods based on how much they raise blood glucose levels relative to the standard of pure glucose (glucose is defined as 100). Thus a high glycemic food should raise blood glucose more than a low glycemic food (see figure to the right). But is this always the case? In a study by Zeevi and coauthors (2015) the authors proposed that identical meals will not necessarily raise blood glucose the same amount when given to different individuals.

The Takehome: This is a very detailed study with multiple experiments performed (as is typically required of studies published in Cell), so my summary is somewhat abridged. The major finding of the study was that postprandial (post-meal) glucose levels (PPGR) are highly variable across individuals, even if they consume the same standardized meal (i.e, bread). Thus, although on average a meal may be middle of the road in how much it raises blood glucose levels, for some participants it was found to raise blood sugar very little, while in others it was found to raise blood sugar a great deal. Similar variability was found in more complex foods such as rice and potatoes. Although these foods raised glucose levels fairly high on average, for some individuals the elevation was minimal. Similar trends were found for meals where the expected increase in blood sugar was expected to be less (i.e., if fat was a significant part of the meal). Here again, different individuals were found to have very low and very high glucose levels after ingestion. In this study the mechanism for variation is attributed to variation in gut microbes from person to person. Data was collected by the authors from the patients and a model was built which could predict glucose levels after meals. The model could could also identify meals that would give lower blood glucose levels for specific individuals. Thus, we have further evidence that nutrition is not one-size-fits-all, and that our gut bacteria may be a significant regulatory factor in our dietary health.

Experimental Design:

  • 800 individuals 18-70.
  • Continuous glucose monitoring every 5 min for 7 days (blinded to participants).
  • During a 7 day period one of 4 standardized meals was provided (meals could include one or more of glucose, bread, bread and butter, bread and chocolate, and fructose), each with 50g of carbohydrates.
  • Other meals were at participants’ discretion.
  • Real-life meals were compared. They all had 20-40g of carbohydrates and a single dominant food component where carbohydrates were never more than 50%.
  • All participants had similar stool microbiome profiles.
  • Postprandial (post-meal) glycemic response (PPGR) was the specific measurement compared among groups.
  • A variety of statistical methods were used to predict glucose levels after eating and to quantify gut microbe changes.
  • Reproducibility of glucose levels (PPGR) after eating a specific meal was found to be fairly high (R = 0.70-0.77).


  • For all standardized meals except fructose, glucose levels after eating (PPGR) were found to differ among different individuals. As an example, the average glucose level after eating bread was 44+/-31mg/dl*h (mean+/-SD) with the bottom 10% having a value below 15 and the top 10% having a value above 79 mg/dl*h.
  • The high variability in PPGR was still present even after normalization of each participant’s PPGR to pure glucose.
  • Real-life meals had average PPGR values that matched reported values for glycemic index (i.e, rice and potatoes were relatively high). However, the variability in PPGR for these meals, like standardized meals, varied across individuals.
  • Standardized meal PPGR values were significantly associated with clinical and gut microbiome data.
  • Type 2 Diabetes/Metabolic syndrome risk factors such as systolic blood pressure, high body mass index, and alanine aminotransferase activity were all positively associated with PPGR levels.
  • A model was developed that predicts PPGRs. The model gives a modest, yet significant prediction of PPGR using only meal carbohydrate content. The model was improved in predictive power by incorporating person-specific factors from their microbiome.
  • The final model was independently validated using a separate cohort of 100 participants and was similar in its predictive power (R = 0.68 and R = 0.70 for the original and validation group, respectively).
  • The model was used to generate data on how how food combinations affected PPGR. For example, although the addition of fat to meals reduced PPGR on average (an expected result based on our understanding of the glycemic index of mixed meals), this effect was found to very person to person.
  • Using the data from the model, predictions were made for foods that should lower PPGR values and this was tested in another group of 12 participants. The “good” diet group had significantly lower PPGRs than the “bad” diet group as expected by the model.


  • Participant data merges males and females which makes sex-effects unclear.
  • The age range of participants is substantial, making age-effects unclear.
  • The final clinical study aiming to lower PPGR based on model predictions has a very small sample size and the duration of the study is short. Confidence in this result will require a larger study of longer duration.
  • Microbiome data, specifically, which bacteria were more or less prevalent in high and low PPGR situations does not always agree with previous microbiome studies, so the specific bacteria that may be “good” or “bad” for PPGR remains unclear.