We are pleased to respond to the comments of Ludwig and Roberts (1) as well as Wolever and Brand-Miller (2). We will address the points raised by the former first.
Ludwig and Roberts state that we “conclude that the glycemic index values of individual foods do not predict glycemic response to mixed meals.” Actually, we go beyond that and demonstrate that the glycemic index value of individual foods do not even reliably predict the glycemic response to that food alone. Indeed, Jenkins et al. (3) showed 15 years ago that the glycemic response to the gold-standard stimulus, glucose in water, depends on the timing of ingestion.
Second, Ludwig and Roberts state that “[b]ecause the observed glycemic response did not differ between diets, the lack of effect on appetite is not surprising.” This assumes glucose or insulin is a key determinant of appetite. While both are correlated with hunger after meals, this is not evidence for causality. Euglycemic clamp studies demonstrate that independent manipulation of plasma glucose or insulin does not alter reported hunger (4).
Third, a question is raised about the adequacy of the methods used to select study foods. This concern was surprising because we considered this a study strength. We selected potential foods from the 2002 International Table of Glycemic Index and Glycemic Load Values (5). This table includes values verified as being determined by methods proposed by the Food and Agriculture Organization of the United Nations and the World Health Organization, as Ludwig and Roberts recommend. However, we then conducted a second round of testing, albeit less vigorous, to verify the values. Thus, the foods were more vigorously tested than in nearly any other published study. In addition, each food, not the combined mean, was comprised of comparable macronutrient composition, energy density, and palatability. This reduced several additional common confounds to study interpretation.
Interestingly, Ludwig and Roberts note that “the composition or manufacturing procedures of individual products may change over time, and shelf life and preparatory methods may also affect glycemic index.” This is the very reason we question the utility of expected glycemic index influences on outcome measures. Given this agreed-upon fact, the concern with our test foods leads to an untenable argument that this variability does not negate the predicted responses of glycemic index diets in free-living consumers but does in more controlled clinical trials.
Fourth, Ludwig and Roberts state, “There are many studies demonstrating that the glycemic index of individual foods predicts response to mixed meals when appropriate methodology is utilized.” We recognize there are studies finding associations, but to be fair to the literature, it should be acknowledged that there are also those that do not (6), and the latter are likely under-represented due to publication bias. It is in part this reason that glycemic index diets have not been endorsed for weight management by most biomedical societies and governmental agencies.
Wolever and Brand-Miller raise three points. The first reflects the same misunderstanding expressed by Ludwig and Roberts regarding the criteria we used for food selection. Their power analysis assumes the foods were only tested by 3 individuals when, in effect, they were tested in 13 (10 before their publication [5] and 3 more by us for verification), a level that exceeds any research recommendations, including their own (7).
Next, they state, “Discarding means with inconsistent values is questionable[…].” It is difficult to imagine that contrasting responses to foods with reliable glycemic index/glycemic load values is a methodological flaw. The fact that predictive power improves with inclusion of foods eliciting unreliable responses suggests that either a property other than glycemic index is responsible or that only a subset of individuals are responsive to the property. Both options undermine the utility of the glycemic index concept.
Finally, Wolever and Brand-Miller recognize that “large between-subject variation of glycemic responses exists.” This recognition is completely consistent with our findings and undermines the predictive value of the glycemic index classification of foods. The glycemic index rating is a property of a food, not a response of an individual. There would be no point in testing foods and publishing their glycemic index values, as these authors have done, nor creating diets based on this property if individual responses to their ingestion are highly variable.
In the larger picture, our findings do not argue against the potential health benefits of a low–glycemic index diet. The balanced inclusion of fruits, vegetables, nuts, legumes, and whole grains that comprise such a diet are wholesome and may aid weight management through mechanisms independent of their glycemic index rating.