点击显示 收起
1 From the Department of Nutritional Sciences, Faculty of Medicine, University of Toronto, Toronto, Canada (TMSW and MY), and The Human Nutrition Unit, School of Molecular and Microbial Biosciences, The University of Sydney, Sydney, Australia (XYZ, FA, and JCB-M)
2 Supported by Sydney University GI Research Services and Glycemic Index Testing, Inc, Toronto. 3 Address reprint requests to TMS Wolever, Department of Nutritional Sciences, University of Toronto, Toronto, Ontario, Canada M5S 3E2. E-mail: thomas.wolever{at}utoronto.ca.
ABSTRACT
Background: Recent studies have concluded that the carbohydrate content and glycemic index (GI) of individual foods do not predict the glycemic and insulinemic effects of mixed meals. We hypothesized that these conclusions may be unwarranted because of methodologic considerations.
Objective: The aim was to ascertain whether the GI and carbohydrate content of individual foods influence glucose and insulin responses elicited by realistic mixed meals in normal subjects.
Design: With the use of a crossover design, we determined the glucose and insulin responses of 6 test meals in 16 subjects in Sydney and the glucose responses of 8 test meals in 10 subjects in Toronto and then the results were pooled. The 14 different test meals varied in energy (220–450 kcal), protein (0–18 g), fat (0–18 g), and available carbohydrate (16–79 g) content and in GI (35–100; values were rounded).
Results:The glucose and insulin responses of the Sydney test meals varied over a 3-fold range (P < 0.001), and the glucose responses of the Toronto test meals varied over a 2.4-fold range (P < 0.001). The glucose responses were not related to the fat or protein content of the test meal. Carbohydrate content (P = 0.002) and GI (P = 0.022) alone were related to glucose responses; together they accounted for 88% of the variation in the glycemic response (P < 0.0001). The insulin response was significantly related to the glucose response (r = 0.94, P = 0.005).
Conclusions: When properly applied in realistic settings, GI is a significant determinant of the glycemic effect of mixed meals in normal subjects. For mixed meals within the broad range of nutrient composition that we tested, carbohydrate content and GI together explained 90% of the variation in the mean glycemic response, with protein and fat having negligible effects.
Key Words: Humans • dietary carbohydrates • glycemic index • mixed meals • glucose • insulin
INTRODUCTION
The clinical relevance of classifying carbohydrate foods by using the glycemic index (GI) has been questioned for >20 y (1-5). The most recurrent and, perhaps, most critical objection is that the GI of individual foods are not maintained in mixed meals because of the confounding effects of fat and protein (1-5). However, in many instances, the conclusion that GI has no clinical utility was based on errors in the way the area under the curve (AUC) or predicted glycemic effect of mixed meals was calculated, use of incorrect GI values, or a lack of statistical power (6). Several well-designed studies showed that the GI predicts the relative glycemic effect of mixed meals containing equivalent amounts of protein, fat, and carbohydrate (7-9).
Recently, there has been interest in exploring the role of GI in determining the glycemic effect of meals that vary in fat, protein, and carbohydrate content. The glycemic response elicited by a meal, which is expressed as a percentage of the glycemic response elicited after ingestion of 50 g available carbohydrate from the GI reference food (glucose or white bread), is termed the "relative glycemic response" (RGR). Wolever and Bolognesi (10) developed the following equation to estimate the RGR of a meal from its GI and the grams of available carbohydrate content (GAvCHO) it contains: RGR = 1.5 x GI x (1 – e–0.018 x GAvCHO) + 13. RGR was found to explain 90% of the variation in mean glucose responses elicited by mixed meals in normal subjects (11). The concept of glycemic load (GL), which is defined as GI times grams of carbohydrate, was introduced as a measure of the glycemic effect of diets and accounts for both carbohydrate source and amount (12). Brand-Miller et al (13) found that the GL of test meals was closely related to the glucose responses they elicited. However, 2 recent studies, both of which tested a large number of composite meals, concluded that GI, RGR, and GL of individual foods do not predict the glycemic responses elicited by mixed meals (14, 15). We disagreed with this because of major concerns about the validity of the methods used (16, 17). Nevertheless, the design of these studies highlighted limitations of our previous work in the area; namely, that only 5 different mixed meals were tested (12) and that the validity of GL was examined by using only individual foods (13).
Therefore, we conducted new experiments to ascertain whether the GI and carbohydrate content of individual foods were determinants of the glycemic effect of normal composite breakfast meals that varied in energy, fat, protein, and carbohydrate content and in GI. We also measured the relations between the mean glycemic effect of the test meals and their calculated GL and RGR values. These relations could differ because GL increases linearly with the amount of carbohydrate consumed, whereas RGR is derived from an exponential model to account for the fact that glycemic responses increase in a nonlinear fashion as carbohydrate intake increases.
SUBJECTS AND METHODS
Postprandial responses elicited by 14 test meals were measured in 2 groups of healthy subjects: 16 subjects (n = 8 men and 8 women) in Sydney tested 6 meals (S1–S6) and 10 subjects (n = 7 men and 3 women) in Toronto tested 8 meals (T1–T8). The subjects in Sydney had a mean (±SD) age of 23 ± 1 y and body mass index (BMI: in kg/m2) of 21.3 ± 0.4. The subjects in Toronto had a mean age of 26 ± 4 y and BMI of 22.4 ± 0.8. Because we previously showed that our results for glucose testing were reproducible using centrally provided test meals (18), we decided, prospectively, to use different test meals in Sydney and Toronto and pool the results so as to increase the number of different test meals for which results were available. At each center, the study had a crossover design with each subject testing every test meal in a randomized order. The procedures used were in accordance with the ethical standards of the institution or regional committee on human experimentation and approval was obtained from the relevant committee on human subjects.
The subjects were studied after a 10–14 h overnight fast. After a fasting blood sample was obtained, the subjects consumed a test meal and additional blood samples were taken 15, 30, 45, 60, 90, and 120 min after the subjects started eating. The 2 participating centers used their normal procedures to collect blood samples and analyze glucose concentrations. In Sydney, finger-prick capillary blood (0.7 mL) was collected from warmed hands into heparinized tubes, the plasma was removed and kept at –20 °C before analysis for glucose (Model HITACHI 911; Hitachi, Tokyo, Japan) and insulin (Coat-A-Count Insulin; Diagnostic Products Corporation, Los Angeles, CA) concentrations. In Toronto, finger-prick capillary blood (2–3 drops) was collected into fluorooxalate tubes, which were mixed and stored at –20 °C before analysis of whole blood glucose (Model 2300 STAT; YSI Inc, Yellow Springs, OH). Although the blood collection and analysis differed in Sydney and Toronto, we previously showed that these specific differences have no significant effect on the GI values obtained for centrally provided test foods (18). Insulin was measured in Sydney but not in Toronto for 2 reasons. The primary objective of the study was to determine the predictability of glucose responses, which, for this purpose, are best measured in capillary blood (18). Routine ability to measure insulin economically in small volumes of plasma was developed in Sydney. The methods used in Toronto require venous blood samples to measure plasma insulin responses. In addition, the Toronto group already showed a very close correlation between mean plasma glucose and insulin responses elicited by 5 mixed meals that varied in energy, carbohydrate, fat, protein, and GI (r = 0.99, P = 0.002) (11).
Thirteen of the test meals consisted of normal breakfast meals and 1 was 50 g oral glucose. The 14 different test meals varied in energy (223–451 kcal), protein (0–17.5 g; 0–31% of energy), fat (0–18.2 g; 0–52% of energy), available carbohydrate (15.5–79.4 g; 28–100% of energy), and GI (35–100) (Table 1 and Table 2). The nutrient composition of the foods used was based on information found on the food label or from the manufacturer's information, with available carbohydrate defined as total carbohydrate minus dietary fiber. The GI values of the individual foods were taken from published tables (21).
View this table:
TABLE 1. Test meals used in Sydney, Australia1
View this table:
TABLE 2. Test meals used in Toronto, Canada1
The GI of each test meal was calculated as follows:
RESULTS
The glucose and insulin responses elicited by the different test meals are shown in Figure 1. In a comparison of the 6 test meals from Sydney, there was significant heterogeneity in the mean glucose (P < 0.001) and insulin (P < 0.001) AUCs (Table 3). Similarly, there was significant heterogeneity in the mean glucose AUCs (P < 0.0001) for the 8 test meals from Toronto (Table 3).
FIGURE 1.. Mean (±SEM) glucose and insulin responses in normal subjects elicited by 14 different test meals. Test meals S1–S6 (top and middle) were tested by 16 subjects in Sydney; test meals T1–T8 (bottom) were tested by 10 subjects in Toronto.
View this table:
TABLE 3. Observed glucose and insulin responses and calculated glycemic load (GL), relative glycemic response (RGR), and relative insulinemic response (RIR) for each test meal1
The fat and protein contents of the test meals did not correlate significantly with mean glucose AUC (Figure 2). However, the amount of available carbohydrate (P = 0.0018) and GI (P = 0.022) alone were significantly related to mean glucose AUC. Although variation in GL explained 90% of the variation in mean glucose AUC (P < 0.0001), the y axis intercept of the regression line, 28 ± 8.5, was significantly greater than zero (95% CI: 10, 47). RGR explained 88% of the variation in mean glucose AUC, and the y axis intercept of the regression line, –11 ± 13, was not significantly different from zero.
FIGURE 2.. Univariate correlations between the mean area under the glucose response curves (AUC) and the fat, protein, and available carbohydrate (AvCHO) contents; glycemic index (GI); glycemic load (GL); and calculated relative glycemic response (RGR) (10) for the 14 different test meals. •, results for test meals S1–S6 (means for n = 16 subjects); , results for test meals T1–T8 (means for n = 10 subjects).
When center, fat, protein, available carbohydrate, and GI were entered into a multiple linear regression model, 93% of the variation in mean glucose AUC was explained; the coefficients for fat (P = 0.44), protein (P = 0.58), and center (P = 0.35) were not significant, whereas the coefficients for available carbohydrate (P = 0.0004) and GI (P = 0.0008) were significant. When center, protein, and fat were dropped out of the model, the remaining 2 variables explained 88% (P < 0.0001) of the variation of mean glucose AUC as follows:
DISCUSSION
These findings showed that, when properly applied in realistic settings, the GI of individual foods is a significant determinant of the glycemic effect of composite mixed meals. More generally, the results indicated that for mixed meals containing 0–18 g fat, 0–18 g protein, 220–450 kcal, and 16–79 g of available carbohydrate, available carbohydrate content and the calculated meal GI together explained 90% of the variation in the mean glycemic response; the effects of protein and fat were negligible. This is inconsistent with the findings of Flint et al (14) and Alfenas and Mattes (15). We believe 2 methodologic differences account for the disparity: correct classification of the GI of foods and the range of carbohydrate intakes and GIs used.
Correct classification of food GI is essential for a valid test of whether the GI predicts the glycemic responses elicited by mixed meals. Flint et al (14) classified food GI by using published tables of GI values (21). However, there is a large range of GI values for many foods, and important misclassification errors can occur if the wrong value is chosen; we contend that Flint et al (16) made such errors. For example, whole-grain rye bread was a carbohydrate source in 2 test meals used by Flint et al (14) and in 2 of our test meals. Twenty-one entries exist for rye and specialty rye breads in the GI tables (20), with a range of 41–86 and a mean (±SD) of 59 ± 12. Flint et al (14) ascribed a GI of 65 to rye breads, whereas we used a value of 49 or 50 (Table 2). We were able to select values from the GI tables that accurately represented the foods used because we had previously determined the GI values of most of them.
Alfenas and Mattes (15) classified the GI of their test foods based on their measurements of GI. However, the methods used were imprecise and each food was tested in only 3 subjects. We estimated that the 95% CI of the Alfenas-Mattes GI values (adjusted to the glucose scale) was ± 70 (17). Because the difference in GI between high- and low-GI foods is only 15–20, a CI of ± 70 means there is a large chance of misclassification error. The GI values we used were based on tests done with the use of recommended methods in 10 subjects, which yielded a CI of ±15 (22).
Fat and protein influence glucose and insulin responses (4, 23). However, few dose-response studies have been performed (24, 25). In addition, it is not clear that the results of studies examining the effect on glycemic responses of adding a source of fat or protein to a fixed carbohydrate intake can be extrapolated to mixed meals. To test whether fat and protein disrupt the ability of carbohydrate content and GI to predict the glycemic response of mixed meals, the carbohydrate content and GI of the different test meals must vary over a reasonably large range. In addition, for a fair and clinically relevant comparison, the variability of the fat and protein contents of the different test meals should be similar to that of carbohydrate. Large variations in one factor may make it impossible to detect the effect of small variations in another. The CVs of the protein, fat, and carbohydrate contents and of the calculated GI of our test meals were 55%, 61%, 41%, and 28%, respectively. In contrast, Flint et al (14) fixed the carbohydrate content of all test meals at 50 g (CV: 0%), whereas the CV of the fat and protein contents were 51% and 89%, respectively. With this design, there is no possibility of finding any effect of carbohydrate content on the glycemic response of the meal.
Similarly, to ascertain whether RGR predicted glycemic responses, which was one of the stated objectives of Flint et al (14), the predicted difference in RGR must be large enough to be detected. The CV of RGR of our test meals was 32%, which was >2.5-times that found by Flint et al (12%; 14); thus, our study had more power to detect the predicted effect of carbohydrate content and GI on glycemic responses. We believe this to be a major factor explaining the disparity in conclusions.
Variation in the fat and protein content of our test meals had a negligible effect on glycemic responses. An excessively large range of carbohydrate content and GI cannot account for this; indeed, the variation in carbohydrate content and GI of our meals (CVs were 41% and 28%, respectively) was less than the variation in protein and fat (CVs were 55% and 61%, respectively). Although the absolute range of carbohydrate content, ie, 16–79 g, was greater than that for protein (0–18 g) and fat (0–18 g), when expressed as a percentage of energy, carbohydrate (28–100%), fat (0–52%), and protein (0–31%) varied over a large range, which likely encompasses most normal meals. Adding 5–15 g protein or fat to carbohydrate has been shown to reduce glycemic responses in normal subjects (24-27); this suggests that the range of protein and fat in our meals, 0–18 g, was large enough to influence glycemic responses.
The utility of GI has been criticized on the grounds that the glycemic responses of foods are not closely related to their insulinemic effects (4). Our previous (11) and current results do not support this hypothesis, because the insulin responses of test meals S1–S6 were closely correlated with their glycemic responses. However, because insulin responses were only measured for 6 of the test meals, there was insufficient power for detailed comparisons of the determinants of the insulin response.
GL and RGR were virtually identical in the degree to which they correlated with mean glucose AUC (r = 0.95 and 0.94, respectively). GL may be a more clinically useful index of the glycemic effect than RGR because it is easier to calculate. However, GL does not account for the fact that the relation between dose of carbohydrate and glucose AUC is not linear (10). Thus, GL was not proportional to glucose AUC, whereas RGR was. This does not affect the ability of GL to qualitatively rank glycemic responses, but proportional extrapolation of small GL values over a large range would tend to overestimate the expected glycemic response.
Because of large between- and within-person variation of glycemic responses, the carbohydrate source and amount cannot be used to predict the absolute glycemic response of a person on a single occasion. However, they can provide a guide to the relative glycemic effect or the rank order of the average glycemic responses elicited by different mixed meals eaten on multiple occasions over a period of time. Indeed, the present results suggest that, when properly applied, the GI and carbohydrate content are remarkably good predictors of the average glycemic effect of mixed meals with a range of nutrient contents. However, accurate prediction of the relative glycemic effect of different test meals is only possible in clinical settings if accurate values for both the carbohydrate content and GI of the specific foods consumed are known. The carbohydrate content of most foods has been measured with the use of reliable methods, and the results are readily available on food labels. Similarly, for the GI to become clinically useful, the GI values of specific foods must be measured with reliable methods and the results made widely available.
In conclusion, we found that, when properly applied in realistic settings, GI is a significant determinant of the glycemic responses of mixed meals. For mixed meals containing 0–18 g fat, 0–18 g protein, 220–450 kcal, and 16–79 g available carbohydrate, carbohydrate content and meal GI together explain 90% of the variation in the mean glycemic response, with the effects of protein and fat being negligible.
ACKNOWLEDGMENTS
TMSW and JCB-M designed the study and obtained funding for the study. MY, XYZ, and FA recruited the subjects and managed the testing sessions. TMSW performed the statistical analysis and drafted the manuscript. All authors reviewed and commented on the manuscript. TMSW is president of Glycaemic Index Testing Inc, Toronto (www.gitesting.com), a privately owned corporation involved in research, training, and education related to GI and GI methodology. He is also president of Glycemic Index Laboratories Inc, Toronto (www.gilabs.com), a privately owned contract research organization. JCBM serves on the board of directors of Glycemic Index Limited, a not-for-profit company that administers in the Glycemic Index Symbol food labeling program in Australia (www.gisymbol.com.au). She is also the director of a not-for-profit glycemic index testing service at the University of Sydney (). TMSW and JCB-M are coauthors of a series of books under the general title ‘The New Glucose Revolution’ (published by Marlowe and Co in North America), which explains the theory and practice of the glycemic index to the lay public. None of the other authors declares a conflict of interest.
REFERENCES