| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
LETTER TO THE EDITOR |
Human Nutrition Unit
University of Sydney
Sydney
NSW, 2006
Australia
Department of Statistics
Macquarie University
NSW, 2109
Australia
E-mail: j.brandmiller{at}mmb.usyd.edu.au
Dear Sir:
Mulholland et al have suggested that our meta-analysis of observational studies of glycemic index (GI) and glycemic load (GL) (1) lacks consistency in the presentation of data for breast and endometrial cancer. At the time of writing, we and many others believed that the etiology of breast cancer in premenopausal and postmenopausal women might have common elements. Although the weight of evidence now suggests that insulin-like growth factor-1 (IGF-1) is unlikely to play a role in the etiology of postmenopausal breast cancer, it was still considered a plausible risk factor at the time the meta-analysis was conceived and was mentioned as the likely mechanism in the most recent article included in the meta-analysis (2). The studies by Giles et al (2) and Nielsen et al (3) were excluded in the final analyses because they did not meet our a priori validity criteria.
To address the concerns raised by Mulholland et al, a reanalysis of the breast and endometrial cancer data are presented in Table 1. When pre- and postmenopausal breast cancers were analyzed separately, there were no statistically significant associations for either GI or GL. When analyzed together, the results were close to what was originally reported. Also, when all valid cancer studies were combined, the rate ratio (RR) remained essentially unchanged from the RRs originally reported for both GI [RR = 1.07; 95% CI: 1.01, 1.14; Cochran's heterogeneity statistic for fixed-effects model (Q) = 11.78; P = 0.76] and GL (RR = 1.01; 95% CI: 0.94, 1.09; Q = 21.522; P = 0.16). Even if the postmenopausal breast cancer studies were removed, the results for GI (RR = 1.07; 95% CI: 1.00, 1.15; Q = 8.331; P = 0.82) and GL (RR = 1.04, 95% CI: 0.95, 1.15; Q = 19.254, P = 0.12) were not substantially different from those originally reported. Finally, when all valid studies for all diseases were combined, the results for GI (RR = 1.13; 95% CI: 1.07, 1.19; Q = 47.235; P = 0.031) were essentially the same as originally reported, whereas those for GL were marginally improved (RR = 1.13; 95% CI: 1.05, 1.22; Q = 79.487; P = 0.00). If only the postmenopausal breast cancer studies were removed from all diseases combined, the results for GI were, again, essentially the same (RR = 1.14; 95% CI: 1.08, 1.21; Q = 42.649; P = 0.038), whereas those for GL were, again, marginally improved (RR = 1.16; 95% CI: 1.07, 1.26; Q = 69.863; P = 0.000).
|
Mullholland et al also questioned whether the use of correlation coefficients between total carbohydrate intakes from food-frequency questionnaires (FFQs) and other dietary assessment methods (eg, weighed food records or a series of 24-h dietary recalls) is the best means of stratifying the studies according to their validity: frequency of dietary assessment and/or FFQ length are suggested as alternative approaches. Analysis of the Blue Mountains Eye Study data (10) showed that correlations between an FFQ and weighed food records for total carbohydrates and GI were very similar (carbohydrate = 0.55; GI = 0.57) and combined with general recommendations for ascertaining the validity of FFQs (11) was the basis for our hypothesis that a study with a correlation for total carbohydrate <0.5 would be unlikely to reliably rank individuals according to their GI. Although repeated measures with a valid FFQ may have improved the overall accuracy and reliability of the results, repeated measures with an invalid tool would not. More than 95% of the 20 studies that measured participants dietary intakes more than once had correlations for carbohydrate intakes measured by their FFQs and an alternate method that were >0.5 and were included in the final meta-analysis. Removal of the additional 7 studies that used a valid FFQ only at baseline would not likely have changed the overall conclusions of the meta-analysis, but would have further limited its generalizability to the broader community. At the time the meta-analysis was conceived and subsequently accepted for publication, it was generally believed that FFQs with longer food lists were not better than shorter FFQs at ranking subjects for most nutrients (12).
Because of the rigorous exclusion criteria used, the final analysis included only 2 studies for each category of heart disease, pancreatic cancer, endometrial cancer, and gall bladder disease and only 1 study for stroke and gastric cancer. Whereas this was not ideal, it did provide greater insight into the effect of GI/GL on the overall risk of chronic disease than did single studies viewed in isolation. Furthermore, these data were aggregated to provide RRs for all cardiovascular disease and all cancers.
Space limitations permitted only a brief overview of the hypothesized mechanisms linking high glycemic diets to the risk of developing certain chronic diseases. We agree that the etiology of these different diseases is complex and multifactorial, but we argue that the main pathophysiological mechanism underlying development is the combined metabolic effects of hyperglycemia, hyperinsulinemia, and alterations in insulin sensitivity in different target tissues and organs (13).
ACKNOWLEDGMENTS
JCB-M is a coauthor of The New Glucose Revolution book series (Marlowe and Co, New York, NY), is the Director of a not-for-profit GI-based food-endorsement program in Australia, and manages the University of Sydney GI testing service. AWB is a coauthor of one of these books, Diabetes & Pre-diabetes Handbook, and is a consultant to a not-for-profit GI-based food-endorsement program in Australia. No other conflicts of interest were disclosed.
REFERENCES
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
