Results from our study support the hypothesis that type of carbohydrate may be related to body weight. However, percentage of calories from carbohydrates, daily total carbohydrate intake, and glycemic load were not found to be related to BMI, and glycemic index was also found not to be related to daily caloric intake.
Validity and reliability of glycemic index and glycemic load are of vital importance to our investigation and to some extent depend on the characteristics of the dietary assessment method. The 7DDR has been shown to be extremely competent in terms of both cross-sectional and longitudinal assessment of diet, especially macronutrients (21
). To minimize the influence of missing values, we identified and assigned substitute values of glycemic index for the foods with a higher carbohydrate content, for which glycemic index values were not available. The glycemic index, glycemic load, and total carbohydrate values in our study are also comparable to those reported in several studies that used an FFQ as a dietary assessment tool (35
). For example, for the 131-item FFQ used in the Women’s Health Study (40
), the reported overall dietary glycemic index was 75 (SD, 5.0), glycemic load was 166 (SD, 32), and total carbohydrate intake was 221 g/day (SD, 36) (40
The hypothetical association between glycemic index and body weight is supported by evidence from short-term animal studies as well as from epidemiologic studies (8
). Short-term studies in humans and animals have provided evidence that a high-glycemic-index diet affects hunger and nutrient partitioning in a way that promotes body fat storage. Compared with rats fed amylose (a low-glycemic-index starch), those fed amylopectin (a high-glycemic-index starch), under nutrient- and energy-controlled conditions for 3–5 weeks, exhibited physiologic changes that favored fat deposition, including larger adipocyte diameter, increased glucose incorporation into lipids, greater fatty acid synthase, and Glut 4
gene expression in fat tissue (41
). By 7 weeks, animals fed a high-glycemic-index diet developed increased epididymal fat mass (9
); by 32 weeks, they developed marked obesity compared with animals fed a low-glycemic-index diet (8
Furthermore, several short-term experimental studies in humans suggest that body weight is positively associated with dietary glycemic index (13
). Obese, hyperinsulinemic women appear to lose more weight after 12 weeks of consuming an energy-restricted, low-glycemic-index diet compared with women who consume an energy-restricted, high-glycemic-index diet (11
). Similarly, it has been observed that obese men have reduced adiposity, as assessed by dual-energy x-ray absorptiometry, after 5 weeks on an energy- and nutrient-controlled, low-glycemic-index diet compared with a high-glycemic-index diet (13
). BMI decreased significantly during an average of 4 months in obese children prescribed an ad libitum low-glycemic-index diet compared with those prescribed an energy-restricted, low-fat diet (12
Some epidemiologic studies provide further evidence that the glycemic effect of the diet might influence weight control. For example, a lower-glycemic-index diet has been found to predict lower waist-to-hip ratio and waist circumference independent of carbohydrate, fat, and fiber intakes (14
), and, in a study of young adults, low fiber consumption (glycemic index was not assessed) predicted higher 10-year weight gain, higher waist-to-hip ratio, and higher 2-hour postprandial glucose insulin concentrations (a measure of insulin resistance) to a greater extent than did total or saturated fat consumption (15
). Although fiber and glycemic index are not equivalent, they tend to be related because viscous dietary fibers and foods in which the natural cell wall architecture remains intact (e.g., legumes) generally have lower glycemic indexes.
Results from analysis of both cross-sectional and longitudinal data in our study support findings from a retrospective cohort study of 107 children attending an outpatient obesity program, in which a low-glycemic-index diet was associated with greater weight loss within 4 months of follow-up (12
). The average height of subjects in our study was 5 feet 7 inches (170 cm). Given this height, the cross-sectional effect of glycemic index on BMI from our study is calculated to be the equivalent of a difference of 4.8 pounds of weight (2.2 kg of mass) for a five-unit change in glycemic index across subjects, or 9.5 pounds (4.3 kg) for a 10-unit change. This effect is clinically relevant, since it is comparable to that achieved in clinical trials involving changes in dietary intake of fat and saturated fat (43
). With knowledge of how foods containing carbohydrate may be classified (e.g., the glycemic indexes for Kellogg’s Corn Flakes (Kellogg’s, Battle Creek, Michigan) and Cheerios (General Mills, Golden Valley, Minnesota) are 92 and 74, respectively), persons are able to make appropriate substitutions acceptable to their palate and lifestyle. Therefore, a change of 5–10 units in glycemic index is possible and of practical significance in terms of their BMI (45
Because there were only small changes in glycemic index and BMI longitudinally over the 1-year observational period of this study, the association between glycemic index and BMI that was found comes from the cross-sectional analysis in which we observed a larger association between glycemic index and BMI than in the longitudinal analysis. This observation is essentially the converse of what is seen in large-scale intervention trials, where large, relatively predictable changes in serum cholesterol values occur with large changes in dietary fatty acid intake, but there is no observable difference cross-sectionally (46
). Here, the spontaneous change in glycemic index or BMI was small, so the variance became very large in relation to the change score because of the simple fact that the variance of a difference is the sum of the variance of the two measures from which the difference is computed (e.g., here, mean weight change = −0.01 pounds (variance, 83.36) and mean change in glycemic index = −0.14 (variance, 38.81)). Therefore, we were unable to detect a stronger association in the longitudinal analyses from this observational study.
Diets low in carbohydrates have been purported to enhance weight loss (2
). Studies have reported greater weight loss in those who consumed a low-carbohydrate diet compared with a diet higher in carbohydrate at 6 months (48
), but weight loss was not sustained at 12 months (48
). Therefore, the long-term effect of a low-carbohydrate diet on weight loss remains controversial (52
). Our data, derived from a free-living population, did not support the hypothesis that low total carbohydrate intake is related to lower BMI and are consistent with 1-year follow-up data from clinical trials (48
The glycemic load is the glycemic index of a food multiplied by its carbohydrate content in grams. It has been suggested that glycemic load better reflects a food’s physiologic effect than either the amount of carbohydrate or the glycemic index alone (5
). Ebbeling et al. (53
) tested a reduced-glycemic-load diet for treating obesity in 14 pediatric subjects aged 13–21 years. Results showed that a decrease of 17 units in glycemic load led to a 1.3-unit decrease in BMI, which was significant in comparison with the control group that consumed a conventional low-fat, low-calorie diet. However, our data did not support this hypothesis. The possible explanation is that daily total carbohydrate intake appeared to be negatively associated with BMI, and glycemic index was positively associated with BMI. Therefore, the combination of glycemic index and carbohydrates diminished the association between glycemic load and BMI. Further investigation is warranted.
It has also been suggested that glycemic index influences hunger (16
) and therefore may be related to overeating (18
). Ludwig (54
) summarized 16 studies that examined the effects of glycemic index on hunger, and all but one demonstrated that low-glycemic-index foods increased satiety, delayed return of hunger, and decreased ad libitum food intake compared with high-glycemic-index foods in humans. For example, a study by Leathwood and Pollet (10
) reported lower blood glucose levels and slower return of hunger after meals containing bean puree (a low-glycemic-index starch) compared with meals containing potato (a high-glycemic-index starch). Holt et al. (55
) showed that glycemic and insulinemic responses to various breakfast cereals are inversely related to satiety score. Ludwig et al. (56
) found that high-glycemic-index meals induce a sequence of hormonal and metabolic changes that promote excessive food intake in obese subjects. However, our data indicated that high glycemic index is unrelated to total caloric intake. Again, further research is warranted.
Strengths of this study include the following: 1) Longitudinal analysis was conducted, and two to five measures for dietary factors and BMI from each subject provided more reliable values for the analyses and therefore more stable results; 2) recall bias was less likely in our study because the 7DDR is a food assessment tool designed to recall food intake during 7 days, in contrast to an FFQ requiring estimation of average exposure over a much longer period, typically 3 months to 1 year; 3) information on physical activity also was collected and controlled for in the analyses; and 4) the data analysis method used in this study enabled us to examine the presence of cross-sectional and longitudinal effects in the same model. This method has been widely used and accepted in the statistical field (33
), although it has not been widely used in most epidemiologic analyses of longitudinal data.
There are some potential limitations to this study. Unlike in an intervention study, we were not able to create large contrasts over time, thus obscuring a possible causal relation between high-glycemic-index foods and weight. Published glycemic index values are also subject to error. The average coefficient of variation for glycemic responses from 11 normal subjects who consumed 50 g of glucose an average of eight times each was 25 percent. The mean coefficient of variation for the glycemic index for 22 normal subjects who consumed 50 g of carbohydrate from white bread was 22 percent (58
). The coefficients of variation were 13 percent for the glycemic index value of Kellogg’s Corn Flakes as tested in four laboratories and 26 percent for potato as tested in 10 laboratories (28
), although, if a standardized method is used, the results for glycemic index agree about as well as data on carbohydrate, protein, fat, and fiber from 24-hour dietary recalls (i.e., with a coefficient of variation of ±5 percent) (59
). Although the 7DDR may have minimized recall bias compared with the FFQ, it was designed to focus on dietary fat. As with most FFQs, it underestimates total carbohydrate (by an average of 33 g; SD, 68) compared with multiple 24-hour recalls (21
). Still, to estimate effect, the absolute amount of bias is relatively unimportant if the degree of linear agreement is good.
Dietary fiber, particularly soluble fiber, is a very important factor affecting food digestion and absorption rate, which in turn affects the glycemic index value (62
). We attempted to break food items down into individual components to calculate glycemic index and glycemic load, but the process is not as precise as it is with the detail available from 24-hour dietary recall data. However, to our knowledge, there are no published glycemic index data based on 24-hour recalls.
Other potential limitations to our study include the fact that our study population consisted largely of White, middle-class subjects who were members of a health maintenance organization. In addition, because the study protocol involved a lengthy series of clinic visits and diet assessments, participants who stayed in the study were highly motivated. Selection factors relating to the participants’ interest in their own health and time availability for participation may have created a fairly homogeneous and health-conscious study group. For these reasons, our findings might not be generalizable to other socioeconomic strata or to other cultural and ethnic groups.
In conclusion, results from our study suggest that daily dietary glycemic index is independently and positively associated with BMI. This finding is consistent with the hypothesis that with increased glycemic index, more insulin is produced and more fat is stored, suggesting that type of carbohydrate may be related to body weight. Our data did not support the current public trend of lowering total carbohydrate intake for weight loss or of lowering glycemic load for weight loss, as suggested by other researchers (5
). These results add valuable information about the role of glycemic index, glycemic load, carbohydrates, and management of body weight and about the potential prevention of obesity. However, further research is needed to deepen understanding of the relation between body weight and dietary carbohydrates.