Glycemic index is as reliable as macronutrients on food labels

Dear Editor:

In a recent article, Matthan et al. (1) measured glucose and insulin responses elicited by 50-g available carbohydrate portions of white bread and glucose in 63 subjects. They concluded that the substantial variability in individual responses to glycemic index (GI) value determinations makes GI an unreliable approach to guiding food choices. This conclusion is false and based on a common misunderstanding.

Matthan et al. confuse the terms “glycemic response” (a variable characteristic of an individual) and “glycemic index” (a property of a food that is assessed with the use of human subjects). They state in the Introduction that “The objective of the present study was to determine the intraindividual and interindividual variability in glycemic response to a single food challenge and potential methodological and biological factors that could mediate responses among healthy adults” (our italics). However, the results and conclusions do not refer to glycemic responses, but rather to GI. The distinction between “glycemic response” and “GI” is more than a matter of semantics: GI is widely recognized to have clinical (2–4) and public health (5) significance and it would be irresponsible to discard this evidence for inadequate reasons. People have glycemic responses; foods have GI values. The former is well known to vary substantially due to intra- and interindividual variations in carbohydrate metabolism; the latter is reliable and reproducible with the use of standardized methodology.

Matthan et al. (1) reported that the mean ± SD GI of Pepperidge Farm Original White Bread was 62.4 ± 15.3, and they consider this amount of variation to be substantial enough to show that GI should not be used to guide food choices because of the potential to misclassify foods as low (GI ≤55), medium (56–69), or high (GI ≥70) GI. However, this interpretation is incorrect. The GI of a food is the mean in n = 10 subjects (5). Thus, with an SD = 15.3, the SEM for n = 10 subjects would be 4.8 and the 95% margin of error, by using the t-distribution, would be 10.9. This margin of error, 10.9, is much smaller than the difference between low- and high-GI categories (i.e., 70–55 = 15), and there would be a chance of <1% of misclassifying a high-GI food as low GI. The margin of error is ±17% of the mean, which is less than the ±20% difference that is legally allowed between the measured macronutrient or fiber content of foods and the values on the Nutrition Facts table (6), and thus is within the range accepted by regulatory authorities for labeling macronutrients. Furthermore, the GI values of common foods vary across a range from 35 to 90, implying that there are highly significant differences in GI between common foods. Therefore, contrary to the authors’ contention, the variability in GI values they observed is sufficiently small to support the utility of GI for distinguishing between high- and low-GI foods.

Furthermore, it is illogical to reject GI as an approach to guiding food choices on the basis of high intraindividual variations in glycemic responses. On this basis, therefore, we should also discard glucose-tolerance testing for the diagnosis of diabetes and carbohydrate counting in the management of type 1 diabetes. The purpose of the GI is not to indicate what an individual’s glycemic response will be on any one eating occasion but rather to indicate which carbohydrate-containing foods will produce, on average, relatively different responses. The utility of GI in this respect has been shown by studies showing that GI is a significant determinant of the mean glycemic response elicited by a range of test meals of varying nutritional composition tested in a group of subjects (7) and a significant determinant of individual glycemic responses elicited by self-selected breakfast meals in a large number of subjects (8).

Matthan et al. (1) indicate that subject glycated hemoglobin explained 16% of the interindividual variability in mean GI without indicating if this correlation was significant nor whether there were any outliers driving the relation. This finding conflicts with the results of other studies that showed a similar ranking of foods according to mean GI among individuals with glycated hemoglobin values varying from normal to poorly controlled diabetes (9, 10).

One of the major objectives of the study by Matthan et al. (1) was to investigate methodologic factors that could influence the results; they found that the mean GI value depended on the length of time that blood was sampled and the method of calculating the AUC, that intraindividual variation tended to become smaller as the number of reference tests increased, and that increasing the number of subjects from 10 to 63 had little or no effect on the mean or SD value. All of these findings support current GI methodology, which dictates that GI must be determined by using a standardized protocol: a defined method for calculating AUC, a defined blood sampling schedule, at least 2 tests of the reference food, and with the use of at least n = 10 subjects.

In summary, Matthan et al. confirm the well-established observation that individual glycemic responses vary substantially. However, their results do not impeach the validity of GI, nor do they undermine evidence from clinical trials and observational research that show important public health benefits of reducing GI.