点击显示 收起
1 From the Steno Diabetes Center, Gentofte, Denmark (CL, KB-J, and CG); the Research Centre for Prevention and Health, Glostrup, Denmark (UT and TJ); the Department of Nutrition, The Danish Institute for Food and Veterinary Research, Søborg, Denmark (IT); and the Department of Endocrinology C, University of Aarhus, Aarhus, Denmark (BR)
2 Supported by the Danish Medical Research Council, the Danish Centre for Evaluation and Health Technology Assessment, Novo Nordisk A/S, Copenhagen County, the Danish Heart Foundation, the Danish Diabetes Association, the Danish Pharmaceutical Association, the Augustinus Foundation, the Ib Henriksen Foundation, the Becket Foundation, and the Danish Research Counselling System. 3 Address reprint requests to C Lau, Steno Diabetes Center, Niels Steensens Vej 2, DK-2820 Gentofte, Denmark. E-mail: cala{at}steno.dk.
ABSTRACT
Background: The few studies examining the potential associations between glycemic index (GI), glycemic load (GL), and body mass index (BMI) have provided no clear pictures. Underreporting of energy intake may be one explanation for this.
Objective: We examined the associations between GI, GL, and BMI by focusing on the confounding factor of total energy intake and the effect of exclusion of low energy reporters (LERs).
Design: This was a cross-sectional study of 6334 subjects aged 30–60 y. Dietary intake was estimated from a food-frequency questionnaire. GI and GL were estimated by using white bread as the reference food. Underreporting of energy intake was assessed as reported energy intake divided by basal metabolic rate (EI/BMR); LERs were defined as those having an EI/BMR < 1.14. Univariate and multiple linear regression models were used to test for associations between GI, GL, and BMI. The confounders were sex, age, smoking, physical activity, alcohol intake, and energy intake. All analyses were conducted on 1) the entire population and 2) a subsample excluding LERs.
Results: In the univariate analyses of the entire population, GL was inversely associated with BMI. No association was observed for GI. After full adjustment (including energy intake), both GI and GL were positively associated with BMI. When LERs were excluded, GL was positively associated with BMI in all analyses, and GI was positively associated with BMI in the multiple analyses.
Conclusions: We showed a positive association between GI, GL, and BMI. Energy adjustment and the exclusion of LERs significantly affected the results of the analysis; thus, we stress the importance of energy adjustment.
Key Words: Glycemic index • glycemic load • BMI • underreporting • population-based study
INTRODUCTION
Few studies have examined the association between glycemic index (GI), glycemic load (GL), and body mass index (BMI), and no clear relation has been identified (1, 2). The reasons for this may be numerous. Inclusion of underreporters may be one of them. Nevertheless, it is rarely discussed whether underreporting of dietary intake (especially total energy intake) is an issue in studies examining associations between diet and disease.
GI provides a measure of carbohydrate quality but not quantity, whereas GL provides a measure of both quality and quantity (3). Therefore, it is assumed that underreporting and adjustment for energy intake will affect estimates of GL more so than estimates of GI. Other studies have not examined the effect of underreporting on the association between GI, GL, and BMI (1, 2).
Food intake is often biased by underreporting, which obscures interpretation of the results; however, adjustment for energy intake provides some correction (4, 5). Hence, most studies within nutritional epidemiology focus on results after adjustment for energy intake. Adjustment for total energy intake is, however, only meaningful if it is assumed that underreporting occurs at a whole-diet level (6). The purpose of the present study was to examine the association between GI, GL, and BMI, focusing on the confounding factor of total energy intake and the effect of exclusion of low energy reporters (LERs).
SUBJECTS AND METHODS
Study population
We used cross-sectional data from 30- to 60-y-old women and men from the Danish population-based Inter99 study. The overall aim, data collection methods, and nondietary baseline results of the study are reported elsewhere (7, 8).
In 1999 the study population comprised 61 301 individuals born in 1939–1940, 1944–1945, 1949–1950, 1954–1955, 1959–1960, 1964–1965, and 1969–1970 who were resident in 11 municipalities in the southwestern part of Copenhagen County. All individuals were drawn from the Civil Registration System. An age- and sex-stratified random sample of 13 016 persons was drawn from the study population, and 12 934 persons were eligible for further examination. All these individuals were invited for a health survey at the Research Centre for Prevention and Health in Glostrup. Baseline data were collected in 1999–2001, and 6784 (52.5%) persons agreed to participate.
Data collection
Weight was measured to the nearest 0.1 kg and height was measured to the nearest 0.5 cm. BMI was calculated as weight in kilograms divided by squared height in meters. Information on physical activity and smoking was obtained from a self-administered general questionnaire completed in advance of the first visit. On the basis of answers about physical activity level during work and leisure time, all individuals were categorized into 1 of 5 groups: physical passivity, light physical activity, intermediate physical activity, stronger physical activity, and heavy physical activity. The participants scored themselves into 1 of 4 categories of smoking status: daily smokers, occasional smokers, exsmokers, and never-smokers.
The participants completed a self-administered 198-item food-frequency questionnaire (FFQ) on which they were asked to report their dietary intake during the previous month. The GI for carbohydrate-containing food items was estimated by using average GI values from the GI table by Foster-Powell et al (3) with white bread as the reference food. The calculation of daily GI and daily GL, the latter including available carbohydrates, was based on 78 different carbohydrate-rich food items with GI values ranging from 10 to 147. A detailed description of the FFQ and estimation of the dietary intake, daily GI, and daily GL of the population is published elsewhere (9, 10). In the present article, calculation of dietary intake was based on an updated version of the Danish Food Composition Data Bank (11).
We used Goldberg's equation, which compares dietary energy intake with energy expenditure, to estimate underreporting of energy intake. Here, the reported energy intake for each individual was divided by the estimated basal metabolic rate (EI/BMR) (12). The calculation of EI/BMR in Inter99 was previously described (9). The fixed cutoff used to identify LERs was an EI/BMR < 1.14, which identifies the minimum plausible level of energy expenditure at the individual level when the dietary method covers intake for >28 d (12). Participants with an EI/BMR 1.14 were classified as being adequate energy reporters (AERs) or high energy reporters (HERs).
We excluded individuals who had not filled in the FFQ, who had not answered any questions on 5 pages out of 14, or who had misunderstood the FFQ. Five pages of missing information were defined as the cutoff because missing information on <5 pages was assumed to give an acceptable estimate of habitual diet. Misunderstanding included questionnaires in which 2 answers per question line recurrently were ticked or questionnaires in which only the highest response frequency was ticked for one or more questionnaire boxes. Those known to have diabetes and those with missing information on smoking status and physical activity were also excluded. Thus, 6334 individuals qualified for the present analyses.
All participants gave written consent before taking part in the study. The protocol was in accordance with the Helsinki declaration and was approved by the local ethical committee.
Statistical analysis
Three linear regression models were used to test for linear trend between each of the continuous explanatory variables GI and GL and the continuous response variable BMI. The residuals of the log-transformed residuals of BMI approximated a normal distribution a little better than the residuals of the nontransformed residuals, but the difference was not dramatic. Thus, for reasons of interpretation, we have chosen to present the results of the nontransformed BMI data.
First, univariate analyses were conducted (model 1). Second, the confounding factors of sex, age, smoking, physical activity at work and during leisure time (categorical), and alcohol intake as a percentage of energy (continuos) were included. We tested for interactions and found that sex in the entire population modified the effect of alcohol (P < 0.05). Hence, the interaction term between sex and alcohol intake was included in this multivariate model (model 2). Third, energy intake as a continuous variable was included in the model together with the interaction term sex x energy intake (model 3), because sex modified the effect of energy intake (P < 0.05).
Finally, we tested whether the categorical variable EI/BMR (including the LER and AER-HER groups) modified the effect of GI or GL on BMI. Significant interactions were observed for each of these univariate models (P < 0.001). Hence, stratified analyses of the categorical variable of EI/BMR were conducted together with analyses on the entire population by using SAS 8.2 (SAS Institute, Cary, NC). A P value of 0.05 was considered significant.
RESULTS
The subjects' mean (±SD) age was 46.1 ± 7.8 y, their BMI (in kg/m2) was 26.2 ± 4.6, their basal metabolic rate was 1645 ± 255 kcal/d, their energy intake was 2331 ± 823 kcal/d, and their EI/BMR was 1.48 ± 0.51. A total of 24.7% (n = 1565) of the population was classified as LERs, 35.5% (n = 2247) were classified as daily smokers, and 34.8% (n = 2202) were classified as physically inactive. The subjects' mean (±SD) GI and GL were 79 ± 6 and 210 ± 93, respectively. The LER group had higher BMIs and were older but had lower energy intakes, lower GI and GL, and lower EI/BMR than did the AER-HER group (P < 0.0001).
The associations between GI, GL, and BMI are presented in Table 1. Negative parameter estimates were observed for the unadjusted analyses (model 1) for the entire population, but only the association between GL and BMI was significant. Adjustment for age, sex, smoking, physical activity, and alcohol intake (model 2) changed the direction of the estimate for GI. Further adjustment for energy intake (model 3) resulted in significant positive estimates for both GI and GL.
View this table:
TABLE 1. Associations between daily glycemic index, glycemic load, and BMI in the Inter99 study
To explain the difference between the observed estimates of model 1, model 2, and model 3 (on the entire population), univariate regression analyses between energy intake and BMI were conducted. An inverse association (Figure 1) was observed between energy intake (kcal) and BMI (kg/m2) when all individuals were included (P < 0.0001). When the LER group was excluded, the association was positive (P < 0.0001), and a very positive association was observed when only the LER group was included in the analyses (P < 0.0001). Because the LER group had a higher BMI but a lower energy intake than did the AER-HER group, an inverse association was observed when the entire population was analyzed. However, positive associations were observed for each subsample, because those with higher BMIs within each subsample, despite underreporting of their energy intake, were reporting a higher energy intake than were those with lower BMIs. Hence, underreporting could explain some of the observed difference between model 1, model 2, and model 3 on the entire population.
FIGURE 1.. Univariate association between energy intake and BMI in low energy reporters (+) and adequate and high energy reporters (). The illustration is based on a random sample of n = 1000. The association was similar for the entire population. The lines may exceed the most extreme points of the subsample, because they are based on data from the entire population (n = 6634). The dark solid line is for low energy reporters, the light solid line is for adequate and high energy reporters, and the dashed line is for the entire population.
Therefore, as also shown in Table 1, when the LER group was excluded from the analyses, all models (1, 2, and 3) resulted in positive estimates for GI and GL, respectively. However, the unadjusted association between GI and BMI was nonsignificant (model 1).
In analyses in which only the LER group was included, a relatively large and significant estimate was observed for the association between GL and BMI in models 1 and 2. However, the association was nonsignificant when we adjusted for energy intake (model 3). The associations with GI as explanatory variable were all nonsignificant.
DISCUSSION
Our main results indicate that both GI and GL were positively associated with BMI when either energy adjustment or exclusion of LERs was considered. To our knowledge, apart from 2 very recent publications (1, 2), no other epidemiologic data on GI, GL, and BMI have been published. Liese et al (2) observed no association between GI and BMI in an unadjusted analysis or after adjustment for confounding factors (age, sex, ethnicity, family history of diabetes, current smoking status, and total energy expenditure) and energy intake. However, GL was positively associated with BMI both in unadjusted analyses and after adjustment for confounding factors, but not after further adjustment for energy intake. It is unclear whether underreporting of energy intake was considered in the study by Liese et al; they concluded that their results showed a consistent lack of association between GI, GL, and BMI (2). Ma et al (1) observed a positive association between GI and BMI but no significant association between GL and BMI in a multivariate model that included energy intake. From this study, it is also unclear whether underreporting of energy intake was considered. Ma et al concluded that the type of carbohydrate (expressed as glycemic index) may be related to BMI.
Despite difficulties in the interpretation of studies in which associations between either GI or GL and different disease outcomes (including body weight) have been analyzed, authors generally emphasize the importance of GL compared with GI (13). We observed that the P values for the associations between GL and BMI generally were stronger than the P values for GI. Our results are therefore not in direct conflict with the existing body of knowledge. However, our data do suggest that GI may be of importance for the prevalence of obesity.
The stratified analyses highlight the problem of underreporting, which can be difficult to account for by any statistical method. But, when comparing the stratified analyses with the analyses of the entire cohort, it becomes clear that energy adjustment is very important. Adjustment for energy intake (model 3) had a great effect on the direction of the associations, the size of the estimates, and the level of significance, especially in analyses of the entire population and as expected for GL. Because GL to a large extent is a measure of carbohydrate intake, adjustment for energy in analysis of GL will take on an isocaloric replacement interpretation of carbohydrate quantity for energy-equivalent quantity of the 3 other macronutrients. Other studies have also observed a great effect of energy adjustment (2). We realize that underreporting of energy was a problem in the Inter99 study. Additionally, it is likely that LERs selectively underreported high-GI foods. We were, however, unable to analyze whether selective underreporting was present. Hence, if we assume that underreporting occurred at a whole-diet level, the results observed after energy adjustment should be more reliable than those from models without adjustment for energy intake (4, 6).
No major difference in the associations was observed when LERs were included or excluded from the energy-adjusted models (model 3) in the present study. The associations observed for the LER group after energy adjustment indicate, however, that inaccurate data combined with a reduction in sample size decrease the probability of significant and true associations. Others have found that the bias introduced by LERs may both create and remove associations between dietary intakes and BMI. For instance, the inclusion of LERs reversed the association between consumption of high-fat sweet foods and BMI, because obese women underreported these products (14). Others have found that the association between diet and the metabolic syndrome is different in LERs and AERs (15). Our results confirm that different associations can be observed for LERs compared with AERs and HERs. This is presumably explained by the fact that overweight and obese subjects, more so than normal-weight subjects, underreport their dietary intake, both in this study (data not shown) and in many other studies (6, 16, 17).
We found that almost 25% of the population were LERs (EI/BMR < 1.14). Another Danish population-based survey (18) showed that 23% of the study sample were categorized as LERs (EI/BMR < 1.1). Direct comparison of the fraction of LERs across studies is difficult because different criteria are used to identify LERs. Comparison of mean EI/BMR is therefore another possibility. The study from Denmark (18) and another large study from Sweden (17) showed that mean EI/BMR is < 1.50. Hence, the mean EI/BMR of 1.48 in Inter99 is acceptable, and underreporting does not seem to be a greater problem in the Inter99 population than in other populations.
We calculated a rough estimate for the degree of underreporting in Inter99 by using Goldberg's cutoff, because the information needed for calculation of a more precise estimate was not available. Other studies showed that around 35% of those classified as AERs by the use of individual physical activity levels are classified as LERs when a fixed cutoff for EI/BMR is used (17). The estimated fraction of LERs in the present study may therefore be slightly greater than the fraction that would have been identified if estimates of energy expenditure had been based on individual physical activity level values.
We showed a positive association between GI, GL, and BMI and addressed the effect of energy adjustment with and without the inclusion of LERs, which radically affected the results of the analysis. Thus, we stress the importance of energy adjustment if the exclusion of LERs is inappropriate. Generally, studies reporting associations between GI, GL, and BMI should be interpreted carefully.
ACKNOWLEDGMENTS
We thank all the participants who took part in the survey and the staff for their serious efforts that made this study possible.
KB-J, TJ, and CG were responsible for the study design, data collection, and obtaining funding. CL was responsible for data analysis and drafting the manuscript. CL, UT, and BR were responsible for interpretation of the results. CG and IT provided significant advice and consultation. All authors edited and reviewed the manuscript. The authors had no conflicts of interest.
REFERENCES