| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
LETTER TO THE EDITOR |
Cancer Epidemiology & Prevention Research Group
Centre for Clinical & Population Sciences
Queen's University Belfast
Room 02040, Mulhouse Building
RVH Site
Grosvenor Road
Belfast
Northern Ireland
E-mail: hmulholland04{at}qub.ac.uk
Dear Sir:
We read with interest the recent meta-analysis published by Barclay et al (1) on glycemic index (GI), glycemic load (GL), and chronic disease risk. Given the recent surge in research in this area, the meta-analysis is undoubtedly addressing a timely issue; however, we have a number of concerns about the way in which this systematic review and meta-analyses were conducted.
First, data extraction from individual publications does not appear to have been done systematically. For example, 3 studies included in the analysis of GI, GL, and breast cancer risk incorporated both premenopausal and postmenopausal women (2-4). However, the reviewers used different methods of extracting data from each of these 3 articles. In one study, information was extracted on risk in all women rather than on risk in premenopausal and postmenopausal women separately (2). In another study, only results for postmenopausal women were considered and risk in premenopausal women was ignored (4), whereas risk in all women and postmenopausal women was considered from the third study (3), thereby excluding results for premenopausal women but including results for postmenopausal women twice. Furthermore, the authors stated that they combined data describing risk in the highest versus the lowest category of GI or GL intake; however, they included results from 2 studies of breast cancer risk that examined GI and GL intake as continuous variables (5, 6) rather than as categories of intake. Thus, the data are not comparable and should not have been included in the meta-analysis. All of these factors may have contributed to a miscalculation in the overall estimates given for GI, GL, and breast cancer risk.
Similarly, for studies of GI, GL, and endometrial cancer risk, the authors extracted data from a subgroup analysis of physically inactive and overweight women in one study (7) rather than from the cohort as a whole, as they did for the other studies of GI, GL, and endometrial cancer risk. In doing so, the results may be biased toward statistical significance, and the positive risk estimate presented may be a misleading exaggeration.
Second, the article also included a stratified analysis of GI and GL intake and chronic disease type based on the validity of the dietary assessment tool used in each study. Although, in theory, this is a commendable objective, the authors acknowledged that no study actually validated their dietary assessment tool for GI or GL intake and therefore they assumed that the correlation coefficients quoted for total carbohydrate were applicable to GI and GL intake. Given that dietary assessment is often problematic and some degree of misclassification is almost certainly involved in the assignment of GI and GL values to foods, it is necessary to apply quality criteria to studies included in a meta-analysis. Despite this, it is questionable whether the proxy correlation measure for total carbohydrate intake should have been used as criteria for a stratified analysis. An alternative method to consider study quality could have perhaps considered the number of times that diet was assessed in individual study populations, the length of the food-frequency questionnaire used, and the validity of the assessment tool used. Furthermore, an appropriate quality assessment scale could have been applied to all studies (8).
Third, the rate ratios (RRs) quoted in the stratified analysis of GI, GL, and chronic disease risk are based on only 1 or 2 studies of heart disease, stroke, pancreatic cancer, gastric cancer, gallbladder disease, and endometrial cancer. To reiterate, endometrial cancer included estimates from a small subgroup analysis of physically inactive, overweight women. Is it really useful to combine data from 2 studies or even use data from 1 study to give a summary risk estimate of GI, GL, and chronic disease?
Fourth, and of more general concern, is a problem with the hypothesis offered to justify the combined RR estimates given for GI, GL and all diseases combined (GI RR: 1.14, 95% CI: 1.09, 1.19; GL RR: 1.09, 95% CI: 1.04, 1.15). The proposed mechanisms by which GI and GL may be associated with chronic disease risk are suggested to be due to effects on insulin secretion and postprandial glycemia following ingestion of high-GI or high-GL carbohydrates. It cannot be assumed that GI and GL contribute to the etiology of different chronic diseases and even different cancers via the same pathways and the authors fail to offer biologically plausible mechanisms for each disease and GI and GL intake. For example, growth factors related to insulin have been linked to premenopausal but not to postmenopausal breast cancer risk (9); however, menopausal status has not been accounted for in the risk estimate of GI, GL, and breast cancer. It is highly likely that the RRs quoted for GI, GL, and all chronic disease risk are being influenced by positive associations seen for some conditions, for example type 2 diabetes, rather than by an increased risk of all chronic diseases. Unfortunately, the article does not indicate the weighting given to individual studies in each meta-analysis.
Finally, the authors also state that heterogeneity between studies in their meta-analysis is allowed for by using a random-effects meta-analysis. Although this is an appropriate method of combining data from observational studies, it should be noted that it is difficult to assess the degree of heterogeneity when data from such a small number of studies are combined. The inclusion of an appropriate chi-square or I2 statistic (10) would have been useful to quantify the heterogeneity observed in each meta-analysis, particularly when risk estimates for all chronic diseases were combined.
All of the highlighted methodologic concerns cast doubt about the risk estimates reported in the Journal regarding GI and GL intake in relation to chronic disease, and we believe that these doubts ought to be addressed by the authors to ensure that the readers are not misled.
ACKNOWLEDGMENTS
None of the authors declared a conflict of interest.
REFERENCES
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
