Recently, Alfenas and Mattes (1) concluded that the differential glycemic responses of foods tested in isolation are not preserved under conditions of chronic ad libitum consumption of mixed meals (1). This conclusion is unwarranted because of serious methodological problems that undermine the validity of their results.
Foods were classified as low–or high–glycemic index by the investigators; the glycemic index of each food was determined in three subjects by measuring glucose four times with a glucose meter and discarding means with inconsistent values. Since white bread was used as the reference, all glycemic index values discussed here are adjusted accordingly. We commend the authors for wanting to measure glycemic index; however, nonstandard methods were used (2). Discarding means with inconsistent values is questionable; bootstrap analysis of our data (2) suggests, paradoxically, that the discarded means may be more reliable estimates of the true mean than the remaining ones.
Different blood sampling schedules influence the mean and variation of glycemic index values (3). Using our data (five foods tested by 47 subjects) (2), we found that the average SD of glycemic index values calculated from glucose results for the blood sampling times used by Alfenas and Mattes was 35, compared with 29 for the recommended seven blood samples. If a glucose meter is used to measure glycemic index the SD is increased by ∼15% (4); thus, we estimate the SD of glycemic index values determined using Alfenas-Mattes methodology to be 35 × 1.15 = 40. With SD = 40 and n = 3, the 95% CI of a mean glycemic index value is ±99, and the chance of obtaining a mean within ±10 of the true mean is only ∼33%. Thus, it is likely that the glycemic index category (high or low glycemic index) of many of the foods was misclassified. This is consistent with the failure to detect a difference in glucose response on day 1 of the period when subjects consumed only one food for breakfast.
Also, Alfenas and Mattes compared glycemic responses elicited by low–and high–glycemic index foods in different groups of subjects. Since large between-subject variation of glycemic responses exists, groups of normal subjects can have different means; e.g., the mean response after 50 g glucose in different groups of 10 subjects of similar ethnicity varied from 153 to 210 (2). Between-subject variation is a confounding variable the authors have not accounted for.
The combination of these several methodological problems seriously undermines the reliability of the results.
References
T.M.S.W. is president and part owner of Glycaemic Index Testing, Inc., and Glycemic Index Laboratories, Inc.
