The glycemic index (GI) is a classification of foods based on their blood glucose-raising potential. The American Diabetes Association (ADA) has questioned the clinical utility of the GI and recommends that priority should be given to the amount rather than the source of carbohydrate. Some have interpreted this to mean that all carbohydrates have a nearly equal impact on blood sugar, and some feel that the GI is now a dead issue.
‘The horse is alive and well.’
Nevertheless, the reasons for questioning the clinical utility of the GI are unfounded because of the following: 1) they are based on studies of single test meals, which provide insufficient evidence on which to base dietary recommendations; 2) they are based on a faulty interpretation of the studies actually cited as evidence; 3) they take no account of better designed studies showing that the GI does apply in mixed meals; and 4) they take no account of studies showing that a low-GI diet improves overall blood glucose control in persons with diabetes. The GI is a valid and potentially useful concept, but is also deceptively complex. There are a number of unresolved problems and unanswered questions, and the appropriate place for the GI in patient education is not known. However, progress cannot be made without balance and objectivity.
The American Diabetes Association (ADA) position on the glycemic index (GI) is that priority should be given to the amount rather than the source of carbohydrate (1). At least some individuals have interpreted this to mean that all carbohydrate foods produce the same glycemic response; for example, an advertisement for booklets on carbohydrate counting published by the ADA gives the following reason for why carbohydrate counting should be taught: "Studies have proven that . . . all carbohydrates have nearly equal impact on blood sugar." I know from anonymous comments received from manuscript reviewers that some individuals feel that the GI is no longer an issue and that continued interest is "flogging a dead horse." However, the horse is not dead, and those who suggest that it is, are misinterpreting data, misquoting the literature, and misusing statistics.
One reason for the ADA not recommending the GI is the belief that it would make life more difficult for persons with diabetes by severely limiting food choices (3). There is no evidence for this. Following any kind of therapeutic diet requires discipline. Choices have to be made resulting in the use of smaller amounts of some foods and more of others. Being on a low-GI diet does not require elimination of all high-GI foods. Indeed, there are situations where high-GI foods may be appropriate or even desirable. The primary emphasis of low-GI diet advice is using more low-GI foods. Evidence from clinical trials suggests that on a low-GI diet, the diet variety actually increases. In our studies in subjects with diabetes, the only foods avoided on the low-GI diet, which were used on the high-GI diet, were ready-to-eat breakfast cereals, instant mashed potatoes, and polished rice. By contrast, the low-GI diet contained pumpernickel bread, beans, peas, lentils, bulgur, parboiled rice, spaghetti, barley, and oat bran, all of which were absent from the high-GI diet (4).
The other reason the ADA does not endorse the GI is because of doubts about its clinical utility, which were raised by the conclusion of a small group of studies that measured glycemic responses to a single meal or, in some two meals. Early on, it was pointed out by Coulston et al. (5) that it is not valid to make dietary recommendations on the basis of the results of studies with a single test meal, an assertion with which I agree (6). Therefore, it could be argued that the ADA's doubts about the GI should be discounted because they are not based on sound evidence. However, I have been drawn into the debate about the effect of the GI on glycemic responses to mixed meals because I feel the results of many studies have been interpreted wrongly. Indeed, there is good evidence that source of carbohydrate is one of the factors that influences the glycemic response of mixed meals in subjects with diabetes and also that low-GI diets improve overall glycemic control.
The Stanford group was the first to disparage the GI, their major contention being that the expected differences between the glycemic responses of foods do not persist in the context of mixed meals (7). This was based on their studies of the glycemic responses of mixed meals containing potato, rice, spaghetti, and lentils from which it was concluded that the results were "totally disparate" from what would have been expected from the GI (7). We explained in 1986 (8) and 1991 (9) why we felt this conclusion was wrong; because it was based on the use of total rather than incremental area under the curve, did not take into account all the carbohydrate foods in the meal when calculating the expected difference in glycemic response, ascribed an inappropriate GI value to one of the foods, and did not consider the possibility of a type II error if the expected difference is small (10). Reanalysis of the 1984 Stanford data lead us to the completely opposite conclusion that the results strongly supported the ability of the GI to predict glycemic responses to mixed meals (8).
Since then, much new data about the GI have appeared. However, only a small and select portion of this literature is considered in the recent technical review of nutrition principles for managing diabetes on which the current ADA position is based (3). The GI concept is dismissed in the second sentence under the heading "Carbohydrates/Sweeteners" (p. 494), which states: "...questions have arisen as to the clinical utility of these [glycemic index] data." Three studies are cited to support this statement: Laine et al. (11), Hollenbeck et al. (12), and Nuttall et al. (13). The Nuttall study does not mention the GI. Despite its conclusion that the ADA food exchange list is useful for meal planning, the data show that there were significant differences among the glycemic responses of the four meals designed to be similar according to exchange lists. Nuttall was a co-author with me in a 1985 study, where we showed that the glycemic responses of these meals were significantly related to the calculated meal GI (14). Hollenbeck et al. fed three different diets to diabetic subjects and found that the diet with the lowest GI had statistically significantly lower glucose and insulin responses than the diet with the highest GI. However, in the abstract there is no indication that the differences between the glucose and insulin responses were significant, and the conclusion is the following: "...daylong plasma glucose did not vary substantially ... providing further evidence that there is relatively little clinical benefit [of the glycemic index]." To me, the data in this study actually support the GI because a difference was expected and a difference was observed. Laine et al. conclude that "...the diabetic exchange lists more accurately predict postprandial responses to carbohydrate-containing foods eaten as part of a mixed meal than does the GI of foods." This is the most challenging study to the GI concept because the glucose and insulin responses to the three meals in the diabetic subjects did not differ. However, in the normal subjects the relative glycemic effects of the three meals, as indicated by Laine et al. (11) (100, 89, and 56%) are similar to the relative meal GI values given (100, 71, and 58%), and the differences in insulin were statistically significant.
These three studies were quoted in the technical review (3) as evidence that the GI does not apply in mixed meals. However, I do not agree with this interpretation; to me, these data tend to support the GI, although not strongly. In these studies, the approach taken to determine whether the GI applies in mixed meals is not stated, but it appears that for the GI to be considered valid, the mean response to each meal has to be statistically significantly different from the mean response for every other meal, regardless of the expected difference. If a significant difference is not observed, the responses are considered to be the same.
This is an inappropriate approach because the fact that no difference is detected does not necessarily mean that no difference really exists. Glycemic responses are quite variable from day to day, within subjects, thus, when the number of subjects studied is not large and the expected difference is small, there is a large possibility of not detecting a difference that really exists (type II error). For example, Laine et al. explain that in their study, a difference of 1 honestly significant difference (HSD) was statistically significant. However, a difference of 1.35 HSDs was needed to have an 80% chance of being significant (11). Thus, there was possibly only about a 50% chance of detecting a difference of 1 HSD.
What approach can be taken to establish whether the data support the GI? We have suggested both quantitative and qualitative approaches. The former is based on the correlation between the observed and predicted responses (8,9). Quantitatively, the GI would apply perfectly in mixed meals if the correlation between the observed and predicted responses was 1 and the y-intercept of the regression line was zero. Such a relationship would indicate that the observed response is directly proportional to the predicted response. In practice, one would not expect perfect correlations because of the variability inherent in determining glycemic responses. The qualitative approach is based on whether the ranking of the observed mean responses corresponds to that predicted by the GI, and the results are amenable to probability analysis using the binomial distribution or X2 test. The Laine, Hollenbeck, and Nuttall studies are not designed for correlation analysis because the number of test meals is too small to have a good chance of detecting a significant correlation or to have confidence that no correlation exists. Nevertheless, when all the data are plotted together, the correlation between the observed and expected responses has a probability <0.1, which is often considered as indicating a trend (Fig. IA). Consideration of ranking may be a more appropriate way to assess these data. Here, the null hypothesis is that the GI does not apply in mixed meals. If the null hypothesis were true, when comparing the glycemic responses of two meals, the chance that the meal with the lower GI would result in a lower glycemic response is 50%. In these three studies, there are a total of 15 comparisons of two different meals taken by the same group of subjects, and in 13 of these cases the meal with the lower GI produced the lower glycemic response. With 13 out of 15 correct predictions, the probability that the null hypothesis is true is only P = 0.003. This suggests that the Hollenbeck, Laine, and Nuttall data, taken together, support the hypothesis that the GI is clinically useful, at least qualitatively, for determining the ranking of the glycemic responses of mixed meals.
Strong quantitative support for the GI concept, as it applies to mixed meals, is provided by three studies (15-17) that were not cited in the ADA technical review. These studies were better designed than those discussed above because they tested 5 or 6 meals in each group of subjects, thus allowing a correlation analysis to be performed. In every group of subjects, there was a significant correlation between the observed and expected glycemic responses. Taken together, the data cluster tightly around the regression line, which, when extrapolated, goes virtually through the origin (Fig. IB).
Further on in the same paragraph in the technical review (3), it is stated as follows: "When equal grams of complex and simple naturally occurring carbohydrates are compared, no significant differences are noted in glycemic response." The citation in support of this statement is from a study by Hollenbeck et al. (18) in which three different diets were fed. This statement is incorrect because the diet with the lowest amount of simple sugars produced a significantly lower mean plasma glucose concentration than either of the other two diets. We have pointed out that this difference is in line with what would be expected from the GI (19).
In 1984 the Stanford group insisted that nutrition recommendations must be based on long-term studies (5). It is therefore surprising that no long-term studies of low-GI diets were cited in the ADA technical review, despite the fact that at least nine such studies in subjects with diabetes have been published (4,20-27), and two of them were published by the ADA in Diabetes Care (24,25). Most of these studies were well-designed, controlled, randomized trials in which diet GI was altered for periods of 2 weeks to 3 months (Table 1). They include a total of 127 subjects with type I or II diabetes. In eight of the nine studies, a significant improvement in blood glucose control was observed, and the results of the negative study (23) were not surprising because there was only a small difference in GI between the two study diets (28). With a weighted mean reduction of diet GI of 15 (e.g., 85-70), the weighted average percentage reduction of glycosylated albumin or glycosylated hemoglobin in the nine studies was 10.2%. These results cannot necessarily be extrapolated to the long-term; nevertheless, a 10% improvement in blood glucose control is about half that achieved in subjects with type I diabetes using intensified insulin therapy in the Diabetes Control and Complications Trial (29) and is equivalent to the effect of oral hypoglycemic agents or insulin in the treatment of type II diabetes (30).
Thus, I cannot accept the conclusion that the GI has no clinical utility. The horse is alive and well. Much experimental evidence supporting the clinical utility of the GI is available, both in terms of its application to single mixed meals and also its effects on long-term glucose control. The principle behind the GI is exactly the same as that of the -glucosidase inhibitors, namely, reducing postprandial glucose responses by reducing the rate of absorption of dietary carbohydrate. The clinical utility of reducing postprandial glucose responses using acarbose in the treatment of type II diabetes is clearly established (31,32). Nevertheless, there are several reasons why I feel caution is required in applying the GI in clinical practice. The GI is not the only, nor the most important, criterion by which to judge a food. Some low-GI foods need to be used in moderation because of high fat content (e.g., chocolate, peanuts), and some high-GI foods may be good choices because they are low energy and of a high nutrient content (e.g., carrots) or, in some situations, may be an appropriate choice because some high-GI foods are low in fat and are convenient (e.g., breakfast cereals, bread). As originally intended (33), the GI was meant to supplement the information in foods tables, not replace it!
The GI does not necessarily relate to food factors that are popularly expected to make a difference, such as the dietary fiber content, the amount of sugar, or the degree of cooking or refining. Many diverse common foods have similar GI values; for example, melba toast, bagels, white bread, 100% whole wheat bread, angel food cake, graham crackers, whole wheat crackers, couscous, corn chips, oatmeal muffins, french fries, mashed potatoes, and canned green pea soup and the breakfast cereals Cream of Wheat, Cheerios, and Golden Grahams have almost identical GI values (range 94-106) (37). On the other hand, the GI values of some foods can vary markedly, depending on variety, processing, and preparation. This does not invalidate the GI concept but may make it more difficult to apply in practice. For example, there are many varieties of rice with different types of starch, processed in different ways that result in different GI values (34-36); the GI of 1-inch cubes of boiled potato can be increased by 25% by mashing them (37); subtle differences in banana ripeness can double its GI (38,39). On the other hand, many foods are fairly consistent when tested in different centers, for example, cornflakes, bread, spaghetti, and lentils (40). Thus, while useful generalizations can be made, detailed knowledge is required for detailed application of the GI, and this is probably beyond the needs and abilities of most people. However, motivated individuals may be able to reduce their diet GI without drastically altering the nature of their diets merely by selecting specific brands or modifying food preparation methods. If there were interest, local varieties of foods likely to have variable GI values could be tested so that reliable local data could be obtained.
My final concern about general use of the GI is that we do not know very much about how to use it. Does using low-GI starchy foods have the same effect as using low-GI foods where the carbohydrate is predominantly sugars, such as certain fruits, yogurt, and dairy products? What is the appropriate place for the GI in terms of patient education? There is some evidence that a primary focus on encouraging the use of low-GI starchy foods in subjects with newly diagnosed diabetes results in a better outcome than traditional advice (27). If using low-GI foods, does the amount of carbohydrate in the diet matter (41)?
My impression is that the people with diabetes and the general public want to know about the GI. If they cannot get the information from reliable sources, they will obtain it from unreliable sources and perhaps receive inappropriate advice. A more balanced approach to the GI would be helpful, and it is time for the ADA to take an objective look at the data rather than evading the issue.
References, Table 1, and Figure 1 Omitted
Dr. Wolever is from the Department of Nutritional Sciences, Faculty of Medicine, Division of Endocrinology and Metabolism, St. Michael's Hospital, University of Toronto, Toronto, Ontario, Canada. Address correspondence and reprint requests to Thomas M.S. Wolever MD, PhD, Department of Nutritional Sciences, University of Toronto, Toronto, Ontario M55 3E2, Canada. Received for publication by Diabetes Care on 23 May 1996 and accepted in revised form 17 September 1996. Published in Diabetes Care, Volume 20, Number 3, March 1997, pp. 452-456, and reprinted here with the written permission of Dr. Wolever.
Address correspondence and reprint requests to Thomas M.S. Wolever MD, PhD, Department of Nutritional Sciences, University of Toronto, Toronto, Ontario M55 3E2, Canada.
Received for publication by Diabetes Care on 23 May 1996 and accepted in revised form 17 September 1996. Published in Diabetes Care, Volume 20, Number 3, March 1997, pp. 452-456, and reprinted here with the written permission of Dr. Wolever.
Go back to Home Page
Go back to Diabetes Directory