The article by Liese et al. (1) investigated the association of glycemic index, glycemic load, and total carbohydrate and fiber intake with direct measures of insulin sensitivity, insulin secretion, and adiposity in a cohort of 979 adults with normal and impaired glucose tolerance. The authors concluded that there were no significant associations of glycemic index, glycemic load, and carbohydrate intake with any measure of insulin sensitivity or secretion.
In their study, usual dietary intake was assessed via a 114-item food frequency questionnaire (FFQ) that had been previously “validated” in a subsample of the Insulin Resistance Atherosclerosis Study (IRAS) population. Unfortunately, the validation study showed that the FFQ did not confidently predict total carbohydrate intake as assessed by repeated 24-h recall of food intake. The Pearson correlation coefficient between the two methods was only 0.37 after adjustment for energy. Furthermore, the correlation coefficients for starch and sucrose were similarly low at r = 0.33 and 0.46, respectively (2).
Is a correlation coefficient between 0.3 and 0.4 adequate for the purposes of validation? Brunner et al. (3) have suggested that a value “of about 0.5 for most nutrients and 0.8 for alcohol between methods is good evidence that the FFQ has the ability to rank individuals … according to nutrient intake.” Indeed, correlation coefficients in the order of 0.6–0.7 are more typical for energy-adjusted nutrients in FFQs (4). In the Nurses Health Study (5) and Health Professionals Follow-up Study (6), the correlation coefficient for energy-adjusted total carbohydrate was r = 0.69 (7).
Like most current FFQs, the IRAS questionnaire was not originally constructed for the purpose of measuring glycemic index, and glycemic index was not included in the original validation study. Therefore, the validity of the calculated glycemic index values is essentially unknown. However, given the relatively low correlation coefficients for quantity of total carbohydrate, starches, and sucrose, it would seem unlikely that subtle differences in the quality (i.e., glycemic index) of that carbohydrate can be accurately assessed. The authors note that correlations for carbohydrate were higher in urban non-Hispanic white subjects (<20% of the sample) and that stratification for ethnicity did not alter their conclusions. However, due to the relatively small sample size, subanalysis may not have had sufficient statistical power to detect associations.
Interestingly, observational studies that found no association between glycemic index, glycemic load, and incidence of type 2 diabetes (8,9) also had validation correlation coefficients for total carbohydrate that were <0.5. In contrast, those studies finding glycemic index, glycemic load, or both to be predictive of diabetes (5,6,10) had r values >0.6.
Given the limitations of the IRAS study in relation to its ability to accurately assess total carbohydrate intake, starches, and sucrose, we believe that the conclusions drawn from the study of Liese et al., i.e., that glycemic index and glycemic load are not related to measures of insulin sensitivity, insulin secretion, and adiposity, should be interpreted with caution.
References
J.C.B.-M. serves on the board of directors for Glycemic Index Limited.
