Glycemic Index and Glycemic Load: Measurement Issues and Their Effect on Diet–Disease Relationships

Home , Glycemic index, Glycemic load, Insulin index

REVIEW Glycemic index and glycemic load: measurement issues and their effect on diet–disease relationships

BJ Venn and TJ Green

Department of Human Nutrition, University of Otago, Dunedin, New Zealand

Glycemic index (GI) describes the blood glucose response after consumption of a carbohydrate containing test food relative to a carbohydrate containing reference food, typically glucose or white bread. GI was originally designed for people with diabetes as a guide to food selection, advice being given to select foods with a low GI. The amount of food consumed is a major determinant of postprandial hyperglycemia, and the concept of glycemic load (GL) takes account of the GI of a food and the amount eaten. More recent recommendations regarding the potential of low GI and GL diets to reduce the risk of chronic diseases and to treat conditions other than diabetes, should be interpreted in the light of the individual variation in blood glucose levels and other methodological issues relating to measurement of GI and GL. Several factors explain the large inter- and intra-individual variation in glycemic response to foods. More reliable measurements of GI and GL of individual foods than are currently available can be obtained by studying, under standard conditions, a larger number of subjects than has typically been the case in the past. Meta-analyses suggest that foods with a low GI or GL may confer benefit in terms of glycemic control in diabetes and lipid management. However, low GI and GL foods can be energy dense and contain substantial amounts of sugars or undesirable fats that contribute to a diminished glycemic response. Therefore, functionality in terms of a low glycemic response alone does not necessarily justify a health claim. Most studies, which have demonstrated health benefits of low GI or GL involved naturally occurring and minimally processed carbohydrate containing cereals, vegetables and fruit. These foods have qualities other than their immediate impact on postprandial glycemia as a basis to recommend their consumption. When the GI or GL concepts are used to guide food choice, this should be done in the context of other nutritional indicators and when values have been reliably measured in a large group of individuals. European Journal of Clinical Nutrition (2007) 61 (Suppl 1), S122–S131; doi:10.1038/sj.ejcn.1602942

Keywords: glycemic index; glycemic load; methodology; diet–disease relationships

Introduction indication of glucose available for energy or storage following a carbohydrate containing meal. Although GI is usually The glycemic index (GI) concept was introduced by Jenkins tested on individual foods, there are methods described et al. (1981) in the early 1980s as a ranking system for whereby the GI and GL of meals and habitual diets can be carbohydrates based on their immediate impact on blood estimated (Wolever and Jenkins, 1986; Salmeron et al., glucose levels. GI was originally designed for people with 1997a). In addition to a role in the treatment of diabetes, diabetes as a guide to food selection, advice being given to low GI and GL diets have more recently been widely select foods with a low GI (Jenkins et al., 1983). Lower GI recommended for the prevention of chronic diseases includ- foods were considered to confer benefit as a result of the ing diabetes, obesity, cancer and heart disease and in the relatively low glycemic response following ingestion com- treatment of cardiovascular risk factors, especially dyslipi- pared with high GI foods. The GI concept has been extended daemia (Jenkins et al., 2002). to also take into account the effect of the total amount of The usefulness of GI and GL has been questioned on carbohydrate consumed. Thus glycemic load (GL), a product several counts: failure to consider the insulin response of GI and quantity of carbohydrate eaten provides an (Coulston et al., 1984), the high intra- and inter-subject variation in glucose response to a food (Pi-Sunyer, 2002), and a loss of discriminating power when foods are combined in a Correspondence: Dr B Venn, PO Box 56, Department of Human Nutrition, University of Otago, Dunedin, New Zealand. mixed meal (Flint et al., 2004). Furthermore, foods with a E-mail: [email protected] high sugar (sucrose) content and those containing both Glycemic index and glycemic load BJ Venn and TJ Green S123 carbohydrate and fat may have a low GI, but may not be responses when consumed in an amount containing 50 g regarded as particularly appropriate choices because of their available carbohydrate because of rapid and near complete energy density and nature of dietary fat (Freeman, 2005). digestion and absorption in the small intestine. From the This review considers the reliability of the measurement and International Tables, the GI for instant potato, determined as the practical application of GI. Its value in relating dietary the mean of six studies, was 85 (Foster-Powell et al., 2002). attributes to chronic diseases is considered in other papers in Sucrose, a disaccharide of glucose and fructose, has a this series. somewhat lower GI of 68, resulting from the fructose component which has an exceptionally low GI of 19. Adding protein or fat to a carbohydrate containing food can also Definition and measurement lower overall GI (Miller et al., 2006). Resistant starch and dietary fibre are largely undigested and not absorbed in the GI is defined as the blood glucose response measured as area small intestine and therefore contribute little to postprandial under the curve (AUC) in response to a test food consumed glycemia. However, a lowering of glycemic response has by an individual under standard conditions expressed as a been found when purified extracts of fibre, particularly of percentage of the AUC following consumption of a reference the type that forms a viscous gel in water such as guar gum, food consumed by the same person on a different day (FAO/ are added to a test food in sufficient quantity (Jenkins et al., WHO, 1998). The test food and reference food (usually 50 g 1976; Doi et al., 1979; Wolever et al., 1991; Tappy et al., glucose) must contain the same amount of available 1996). GI cannot be predicted from the fibre content of a carbohydrate (Figure 1). It is important to standardize GI carbohydrate containing food or from the terms wholemeal testing conditions, and the procedure for the measurement and wholegrain for which there are no universally accepted of GI is described in detail in the 1998 FAO/WHO report on definitions. For example, from the International Tables, the carbohydrates in human nutrition (FAO, 1998). Hundreds of mean GI of wholemeal bread from 13 studies is 71, while foods have been tested for GI with the aim of ranking foods that of white wheat bread (mean of six studies) is 70 (Foster- within and between food categories. A GI classification Powell et al., 2002). Whole grains, when largely intact, have system is in common use in which foods are categorized as been found to lower GI (Jenkins et al., 1986; 1988; Liljeberg having low (o55), medium (55–69) or high GI (470) (Brand- et al., 1992; Granfeldt et al., 1994; 1995), but wholegrain Miller et al., 2003a). products contain a variable proportion of intact grains. Glucose, a monosaccharide, induces a large glycemic GI does not take into account the amount of carbohydrate response and is often used as the reference food and assigned consumed, an important determinant of glycemic response. a GI of 100. Some polysaccharides, such as those present in For example, watermelon has a high GI (Foster-Powell et al., instant potato for example, may also result in large glycemic 2002) and may not be considered a good food selection as

8 7 6 5 4 3 2 1

Blood glucose (mmol/L) 0 0 306090120 Time (min)

10 10

8 8

6 6

4 4

2 2 Blood glucose (mmol/L) 0 Blood glucose (mmol/L) 0 0 30 60 90 120 0 306090120 Time (min) Time (min) Figure 1 Example of an individual’s data used to estimate glycemic index (GI). Area under the curve (AUC) refers to the area included between the baseline and incremental blood glucose points when connected by straight lines. The area under each incremental glucose curve is calculated using the trapezoid rule (note: only areas above the baseline are used). GI ¼ AUCFood/mean (AUCReference) Â 100.

European Journal of Clinical Nutrition Glycemic index and glycemic load BJ Venn and TJ Green S124 350 175

300 150 min) ·

250 125 iAUC food 200 100

150 75 100 50 GL direct measure 50 25 Blood glucose iAUC (mmol/L·min) Blood glucose iAUC 0 (mmol/LBlood glucose iAUC 0 12.5 25 37.5 50 62.5 75 0 Available carbohydrate (g) 0 12.5 25 37.5 50 62.5 75 Glycaemic load (g) Figure 2 Blood glucose area under the curve (AUC) responses to increasing amounts of glucose and granola bar tested in 20 people. Figure 3 Example of an individual’s standard glucose curve generated using glucose doses of 12.5, 25, 50 and 75 g. A test food is consumed and the resulting area under the curve (AUC) used to impute the glycemic load. AUC refers to the area included between part of a low GI diet. However, watermelon only contains 5 g the baseline and incremental blood glucose points when connected of carbohydrate per 100 g, thus it would have a minimal by straight lines. The area under each incremental glucose curve is glycemic effect. GL takes into account how much carbohy- calculated using the trapezoid rule (Note: only areas above the baseline are used.) drate a serving of a food contains and may be determined by indirect and direct methods. The indirect method involves multiplying the GI of a food Table 1 Examples of GL arranged by classification taken from the by the amount of available carbohydrate in the portion of international tables (Foster-Powell et al., 2002) food consumed. This method implies that GL is directly Food GI Serving Available GL proportional to the amount of the particular food eaten. This size (g) carbohydrate (g) is perhaps counterintuitive, because the blood glucose AUC does not increase in direct proportion to the amount Watermelon 72 120 6 4 Ice cream (high fat) 37 50 9 4 consumed. For example, eating six times the amount of bread results in an approximately threefold increase in AUC Mashed potato 74 150 20 15 (Brand-Miller et al., 2003c). In other words, as the amount of Macaroni 47 180 48 23 food increased, the rate of increase in AUC declines, an effect shown in Figure 2 (Venn et al., 2006). Therefore, it is implicit Parboiled rice 64 150 36 23 Chocolate bar 65 60 40 26 in the calculation of GL that the AUC for both the test and the reference foods are attenuated to the same degree with Porridge 58 250 22 13 increasing amounts consumed. Corn flakes 81 30 26 21 Glycemic equivalence is a method of directly determining Abbreviations: GI, glycaemic index; GL, glycaemic load. GL. For each subject an AUC for glucose is calculated for a range of doses of the reference food measured on different days. A standard curve is constructed for each subject with increasing amounts of the reference on the x axis with its GI Â available carbohydrate agrees well with GL measured corresponding AUC for blood glucose on the y axis (Venn directly, at least when food is consumed over a range of usual et al., 2006). The AUC in response to a food consumed at any intakes (Venn et al., 2006). A GL classification system is used portion size, typically a usual serving, is compared to that in which foods are categorized as having low (p10), medium individual’s glucose standard curve as depicted in Figure 3 (410–o20) or high GL (X20). (Venn et al., 2006). Using this technique, glycemic equiva- The relationship between GI and GL is not straightfor- lence is the amount of glucose that would theoretically ward; for example, a high GI food can have a low GL if eaten produce the same blood glucose AUC as that particular in small quantities. Conversely, a low GI food can have a portion size of food consumed. Major drawbacks of the direct high GL dependent upon the portion size eaten. This effect is method are increased time and cost required to determine demonstrated in Table 1, in which various foods from the the GL of a food. The reference must be tested at several International Tables have been selected (Foster-Powell et al., doses in each subject and the GL of a food cannot be 2002). A ‘serving size’ of watermelon, a high GI food, has the estimated from currently available GI values. Data from our same GL as a serving size of high fat ice cream, a low GI food. laboratory support the premise that GL is linearly related to Mashed potato and macaroni may be contrasted with the the amount of food consumed that is, GL calculated using lower GI food (macaroni) having a higher GL per serving.

European Journal of Clinical Nutrition Glycemic index and glycemic load BJ Venn and TJ Green S125 Foods having very different nutrient profiles can have The use of a test food referenced to a standard could be similar GIs and GLs per serving, such as parboiled rice and used as an argument that GI is a property of food, rather than a chocolate bar. GI and GL can also be positively related to a characteristic of the individual consuming the food. each other, for example comparing porridge and corn flakes, However, postprandial glycemia may be influenced by the in which the higher GI food (corn flakes) predicts a higher extent to which individuals chew food prior to swallowing GL per serving. Although a food is assigned a fixed GI value, (Read et al., 1986; Suzuki et al., 2005) as well as the expected any food could have a low, medium or high GL because GL is biological variation in rates and extent of digestion and dependent upon the amount eaten. absorption. These variables may not apply equally to test and The glycemic load of a diet can be calculated by summing reference foods; a reference food commonly used is a glucose the glycemic loads for all foods consumed in the diet. A low beverage. The observed intra- and inter-individual differ- GL diet could be achieved by choosing small servings of ences in GI and GL which are apparent even when measured foods relatively high in carbohydrate having a low GI. under standardized conditions may be further exaggerated Alternatively, a low GL diet could comprise foods having a by differing physical and chemical nature of apparently high fat, high protein, low carbohydrate content. The similar food products (Wolever, 1990). For example, gelati- heterogeneity of foods that could be used to construct a nization, the process of rendering starches water soluble; low GL diet indicates that food selection should not be made retrogradation, a realignment of starch molecules during on GL alone. Knowledge of other qualities of the food, for cooling and storage; starch type; and dietary fibre, are all example fat content, type of fat, energy density, fibre factors with potential glycemic-modifying effects. Some of content and appropriate serving size should be taken into these factors are affected by cooking times and methods, and consideration. the temperature of the food consumed, potentially providing a source of variability both in GI measurement and in day-to-day variability of glycemic responses to the same GI and GL labelling food. Voluntary GI labelling of foods by food manufacturers occurs in several countries. Products may need to meet nutritional Reliability of GI values for individual foods compositional requirements to be eligible for GI testing and labelling, such as a limit on the type or amount of fat The 1998 Joint FAO/WHO Expert Consultation on carbo- contained in the food. However, compositional require- hydrates suggested that for the determination of the GI of a ments are not standardized either within a country, where food, six subjects would be required; although the basis for more than one laboratory may provide a GI-testing service, this number was not given (FAO, 1998). More recently, it has or among countries around the world. Standardized elig- been recommended that a sample of 10 should be used, on ibility criteria would give consumers, health professionals the grounds that it allows for a ‘reasonable degree of power and regulators more confidence in the suitability of a food to and precision for most purposes’, although it was acknowl- display its GI. It could be argued that GL should be labeled edged that more people would be necessary if greater because GL more closely reflects the glycaemic impact precision was required (Brouns et al., 2005). However, there associated with consuming an amount of the food. are strong indications that using 10 people is insufficient to obtain reliable estimates of GI, particularly if GI levels are Factors affecting the measurement high, because variance increases with the mean. Large variation in glycemic response between- and within-people Postprandial glucose concentrations are dependent upon makes it difficult to show differences among foods. For several factors. In people with impaired glucose tolerance example, Henry et al. (2005) tested eight varieties of potato and diabetes the glycemic response measured as blood in groups of 10 people and reported mean7s.e.m. GIs glucose AUC is increased compared with healthy individuals. ranging from 5673to94716. Despite a wide range of GIs, it However, GI is the AUC in response to a test food relative to was not possible to demonstrate statistically significant that of a reference food and given that each person acts as differences among the potato varieties. Because of the large his/her own control the GI of a food should not differ in variation there is the potential to miss-classify foods into those with and without abnormalities of glucose metabo- categories of low, medium, or high GI. lism. GI testing has been carried out, and values published in In an inter-laboratory study, seven laboratories tested the international tables, using normoglycaemic individuals as GIs of centrally provided foods, each using 8–12 participants well as those with impaired glucose tolerance (Foster-Powell (Wolever et al., 2003). A range in mean GI values among et al., 2002). Despite broadly comparable results Brouns et al. laboratories was obtained for each of the test foods; potato (2005) have recommended using people with normal glucose 65.2744.6–98.5720.6; bread (locally sourced) 64.2715.4– tolerance for the determination of GI because variability in 78.9726.1; rice 54.8724.1–85.0728.6; spaghetti 36.4735.8 glycemic response is greater in people with impaired glucose –69.9718.8; and barley 23.2724.6–47.1749.7. Rice would tolerance or diabetes. have been classified as low GI (54.8724.1) by one laboratory,

European Journal of Clinical Nutrition Glycemic index and glycemic load BJ Venn and TJ Green S126 medium GI (62.6725.0, 63.378.1, 68.4748.0) by three individual GIs of foods in a meal can be used to reliably laboratories, and high GI (76.9712.9, 85.0728.6, calculate the GI of the meal. Flint et al. (2004) used GI values 87.0775.9) by the other three laboratories. It appeared that taken from the International Table of GI (Foster-Powell et al., results were more consistent when GI was calculated using 2002) to predict the GI of 13 simple breakfast meals, each capillary blood obtained by finger prick rather than venous providing 50 g available carbohydrate. There was no associa- blood. A much better estimate was obtained when data from tion between the GI calculated from the Tables and the the laboratories using capillary blood were combined. Using measured GI. It was pointed out that the energy content of a pooled sample of 47 participants, mean GI for rice was 69 the meals had not been standardized (Brand-Miller and with narrower confidence intervals (95% CI: 63, 76). Thus, in Wolever, 2005); however, it might be argued that GI testing addition to indicating the most appropriate method for is standardized to an available amount of carbohydrate, not blood sampling, the data demonstrate the enhanced relia- to energy content. In another study, a closer relationship was bility of a measurement when studying large numbers of reported between calculated GI and glycemic responses to individuals. Most of the variation in GI differences between various breakfast meals, but the agreement was not entirely laboratories was attributed to random within-person varia- consistent (Wolever et al., 2006). Two of the meals, a bagel bility (Wolever et al., 2003). There is no ready explanation for with cream cheese and orange juice; and a meal of rye bread, this random day-to-day variability in glycemic response that margarine, cereal, milk, sugar and orange juice each occurs even to repeat challenges of the same food under contained approximately 69 g available carbohydrate. The standardized conditions. mean7s.e. glycemic responses to the meals, measured as Within-person variability can be reduced to some extent AUC, were similar (148714 and 143713 mmol/l min, re- by increasing the number of replicates for each subject. The spectively). However, using published values the calculated current recommendation is that the reference food should be GIs of the meals were predicted to be 67 and 51, respectively. tested two or three times in each subject (Brouns et al., 2005). Sugiyama et al. (2003) found that the ingestion of milk with However, within-person variability is also present for the test rice resulted in a significantly lower GI than when rice was food. Using data obtained from our laboratory in which a eaten alone. When cheese was added to potato, a dramatic test food and a reference food were tested three times and lowering of the potato’s mean7s.e.m. GI from 9378to four times, respectively, in 20 people, we have calculated 3975 was found (Henry et al., 2006). Thus while it is clear sample sizes necessary to be confident of a difference of 10 that combining foods does influence GI and that the GI units between foods. The sample size is dependent on the addition of protein and fat to a carbohydrate containing level of GI. For a difference of 10 units, between 30 and 40 meal can appreciably reduce the glycemic response (Collier for instance, it was estimated that 25 people would be and O’Dea, 1983; Nuttall et al., 1984) there is insufficient required if the food was tested once and the reference food information to accurately predict the effect of different three times, or 19 people if the test and reference foods were combinations of foods. Aggregating the GIs of individual both tested twice. These estimates are comparable with the components of a meal does not reliably predict the observed sample size estimates shown by Brouns et al. (2005). The GI of the meal as a whole. same likelihood of detecting differences of 10 GI units between foods having GIs toward the upper end of the scale (70 and 80), would require a sample size of 114 people Published glycemic index and glycemic load values testing the food once and the reference food three times, or 86 people if the test and reference foods are both tested Care must be taken when using published GI and GL values. twice. Increasing the sample and/or repeating the test food Variability of GI and GL among apparently similar foods has would appreciably increase the cost of testing, perhaps a led to recommendations that some foods, for example rice, necessary expense, if more precision in GI measurement is to should be tested in the geographical region in which they are be achieved. consumed. This may be necessary to account for differences in variety and cooking conditions (Foster-Powell et al., 2002). Testing in specific populations may also be important if GI is Mixed meals not solely a property of food. Mettler et al. (2007) found that the training state of athletes affected GI of the same food. It The ranking of meals by GI has been found to reflect the was suggested that Flint et al. (2004) who reported that ranking of the major carbohydrate component in the meal. combining the GI of single foods did not predict the GI of For example, baked potato was found to have a higher GI mixed meals, chose incorrect GI values from published tables than rice (Jenkins et al., 1984). When these foods were (Wolever et al., 2006). This problem is likely to occur when incorporated into meals, there was a tendency for the multiple entries for the same food are presented. For postprandial glycemic ranking of the meals to be maintained example, in the International Tables baked Russet Burbank according to the GI ranking of the food that provided the potatoes eaten without added fat (that is butter) are listed as major carbohydrate source (Wolever and Jenkins, 1986). having GI values of 56, 78, 94 and 111 (Foster-Powell et al., However, there is debate as to whether summing the 2002). Multiple entries for the same foods also complicate

European Journal of Clinical Nutrition Glycemic index and glycemic load BJ Venn and TJ Green S127 research activities and food selection for individuals trying to preparation and consumption may all contribute to the follow a low GI diet. For example, boiled carrots have mean marked inter- and intra-individual variation observation in GI values listed in the International Tables of 92, 49 and 32 the glycemic response to foods. GI and GL testing on larger (Foster-Powell et al., 2002). If GI were a major criterion in numbers of individuals than previously undertaken or food selection, a value of 92 might have discouraged people increasing the number of replicates carried out on an from eating boiled carrots. On the other hand, low GI values individual will improve the reliability and precision of GI of 32 and 49 would have suggested boiled carrots as a highly estimates. Specifying origin and other details of product (for suitable choice. It has been suggested that reliability of the example, variety of fruit or vegetable) will further enhance estimates might have contributed to the differences in GI confidence in the measurement. However, the usefulness of (Foster-Powell et al., 2002), an argument in favour of the index will always be limited to some extent by the studying large numbers of individuals under the standar- variation between and within individuals. It is common dized conditions described above. practice to place foods into broad categories of GI. Classifica- tion works well when there is a large separation in GI values between foods, but there are still uncertainties into which Dietary instruments category of GI many foods belong because of the variability around group mean GI values. More certainty in the relative Questions have been raised regarding the appropriateness of ranking of foods by GI would be attained if larger group sizes the dietary instruments used when examining the relation- were used to estimate GI. The degree to which limitations of ship between GI and GL and various diseases in observa- currently available data influence present use of the concepts tional studies (Pi-Sunyer, 2002). Food frequency of GI and GL in understanding cause or influencing questionnaires used in several studies were not designed management of disease is considered further in the section specifically to obtain information on GI and GL. Individual on recommendations. foods were not assigned GI or GL values, rather they were collapsed into categories. Most of the published studies have not described how foods were grouped, but an Australian Glycemic index and glycemic load and human study gave some insight into the process (Hodge et al., 2004). health The food frequency questionnaire used in the Australian study had a category of ‘cereal foods, cakes and biscuits’. The GI/GL concept has been widely advocated as a means of Within that category were 17 items, two of which were identifying foods that might protect against chronic diseases ‘muesli’ and ‘other breakfast cereals’. A GI value of 46% was or be useful in disease management. Potential protection assigned to muesli and 62% to ‘other breakfast cereals.’ The against diabetes and cardiovascular disease is considered in use of a single value to describe the GI of breakfast cereals the paper by Mann (2007). Several other health issues are seemed inappropriate given that the GIs of Australian cereals discussed here. range from 30 (All-Bran) to 85 (Rice Bubbles) (Foster-Powell et al., 2002). Pi-Sunyer has drawn attention to the even broader groupings that were used in the Health Profes- Long-term glycemic control in diabetes sionals’ Follow-up Study and the Nurses’ Health Study. Categories used were—all whole grains; all refined grains; The GI concept was used in the management of diabetes all cold breakfast cereals; all fruit; and all fruit juices (Pi- before being used in other clinical situations. Glycated

Sunyer, 2002). Such observations do raise some concerns haemoglobin (HbA1C) is measured in people with diabetes about the degree of confidence that can be placed on the to assess overall glycemic control over a period of approxi- findings of these cohort studies with respect to dietary GI or mately 2–3 months prior to the measurement being made GL and disease outcome. On the other hand one might argue (Goldstein et al., 2004). Fructosamine, another glycated that misclassification leads to an underestimate of the true protein, is also occasionally used as a measure of glucose association between GI and GL and disease (see paper by control over the preceding 2–3 weeks. Several studies have Mann in this series). The use of dietary questionnaires been carried out in people with diabetes to examine the specifically designed to gather information on GI and GL, effect on HbA1C or fructosamine of diets differing principally such as those currently being developed, may be expected to with respect to GI. Data from these studies form the basis of generate data which can be interpreted with greater two meta-analyses. There was a modest reduction in HbA1C confidence (Flood et al., 2006; Neuhouser et al., 2006). in people consuming low GI diets, estimated to be 0.33% units (95% CI: 0.07, 0.59) in one meta-analysis (Brand-Miller et al., 2003b), and 0.27% units (95% CI: 0.03, 0.5) in the Overall comment on reliability other (Opperman et al., 2004). Fructosamine concentrations were also lower in favour of low GI dietary interventions. In Biological variation, differing chemical and physical struc- one meta-analysis, the estimated difference between low and ture of apparently similar foods and method of food high GI dietary periods was 0.19 mmol/l (95% CI: 0.06, 0.32)

European Journal of Clinical Nutrition Glycemic index and glycemic load BJ Venn and TJ Green S128 (Brand-Miller et al., 2003b), and in the other 0.1 mmol/l altering the GI of test diets. There was a difference in total (95% CI: 0.00, 0.20) (Opperman et al., 2004). Although these and LDL-cholesterol concentrations of 0.33 (95% CI: 0.18,

reductions in HbA1C or fructosamine are small it is important 0.47) mmol/l and 0.15 (95% CI: 0.00, 0.31) mmol/l, favoring to note that these effects are in addition to other dietary the low GI diets, but no difference in HDL cholesterol changes or pharmacological treatments used in diabetes concentrations between people consuming low and high GI management. Whether changes in glycated proteins of this diets. magnitude affect long-term health outcomes is untested. Thus, the reasonably consistent finding in observational Trials using drugs such as acarbose, which lower postprandial studies of an inverse association between dietary GI and HDL hyperglycaemia, suggest that acarbose may be effective in cholesterol concentrations is not confirmed by meta-ana- reducing cardiovascular complications in people with type 2 lyses of randomized controlled trials in which people diabetes mellitus (Hanefeld et al., 2004). However, reductions consumed diets that had been designed specifically to

in HbA1C of 0.6–0.8% were achieved in these trials (Hanefeld achieve differences in GI (Kelly et al., 2004; Opperman et al., 2004; van de Laar et al., 2005). Nevertheless, any et al., 2004). On the other hand, the meta-analyses show dietary strategy that resulted in improved glycaemic control differences in total and LDL cholesterol not found in the would be welcome and given the difference in the acute observational data. There is no obvious explanation for this effect that low and high GI foods have on postprandial inconsistency. hyperglycaemia, the proposition that changing foods in the diet from high to low GI might improve markers of glycemic control is entirely plausible. However, some caveats may be Glycemic index and glycemic load and insulin appropriate. In many of the studies included in the meta- response analyses described above, the low GI foods tended to have low energy density and a high fibre content, such as whole The GI of a food is affected not only by the rate of absorption fruit, oats, whole grain, pulses and pasta (Frost et al., 1998; of carbohydrate, but also by the rate of glucose removal from Heilbronn et al., 2002). Thus, modest changes in glycaemic the plasma. When comparing two breakfast cereals with control were achieved under study conditions that required different GI values (131733 and 54.577.2), the rate of people to be compliant with relatively major changes in glucose removal was a major determinant of postprandial dietary habits. The findings may not necessarily apply to the hyperglycaemia (Schenk et al., 2003). It was found that the many low GI functional and convenience foods currently lower GI breakfast cereal had induced hyperinsulinaemia available, which may be relatively high in sugars and energy earlier than the higher GI cereal, resulting in an earlier dense. increase in more rapid removal of glucose from circulation. It has been known for some time that insulin response cannot be predicted based solely on the glycemic response to a food. Glycemic index and glycemic load and blood lipids Collier and O’Dea (1983) found marked differences in the glycemic response to potato with or without added butter, Relationships between dietary GI and blood lipid fractions but a very similar insulin response. The effect of GI on have been assessed in several prospective observational insulin response may also depend upon insulin sensitivity. studies. A reasonably consistent finding has been an inverse Dietary GI has not been shown to have a marked effect on association between fasting HDL cholesterol concentrations insulin sensitivity whereas dietary fibre has (McAuley and and dietary GI (Liu et al., 2001; Amano et al., 2004; Slyper Mann, 2006). et al., 2005), although one study found no association (Murakami et al., 2006). Ma et al. (2006) found inverse associations between dietary GI and GL in a cross-sectional Glycemic index and satiety analysis, but the associations were lost during follow-up. An inverse association between GI and HDL-cholesterol con- An important justification for the claim of an overall health centration has also been found in a nationally representative benefit of low GI foods is that low GI foods may aid weight sample of US adults (Ford and Liu, 2001). control because they promote satiety (Brand-Miller et al., Findings from intervention trials differed from those of 2002). Ideally, weight loss studies comparing low and high observational studies. Kelly et al. (2004) conducted a meta- GI diets would need to assess differences between diets based analysis of intervention trials that had examined the effect of on ad libitum intake to show that the apparently greater low GI diets on coronary heart disease risk factors. Results satiating effect of low GI foods led to a reduced energy from that analysis showed limited and weak evidence of an intake. Holt and colleagues have carried out the most inverse relationship between GI and total cholesterol, with comprehensive study investigating the relationship between no effect of dietary GI on LDL and HDL cholesterol, GI and satiety, reporting the same work in several articles triglycerides, fasting glucose and fasting insulin. Opperman (Holt et al., 1995; Holt et al., 1996; Holt et al., 1997). Iso- et al. (2004) conducted a meta-analysis of 14 randomized energetic (1000 kJ) servings of 38 foods were tested for satiety controlled trials relating to the effects on blood lipids of rating and glucose and insulin response. The food with the

European Journal of Clinical Nutrition Glycemic index and glycemic load BJ Venn and TJ Green S129 highest satiety score was boiled potato. When comparing a that report some of the limitations of the GI and GL concepts high GI food (potato) with a low GI food (white pasta) on an have become increasingly apparent. With regard to measure- iso-energetic, equi-carbohydrate (49 g) basis, the high GI ment there is clearly a need to study a larger number of food had the highest satiety rating. The opposite was true subjects under standard conditions to obtain more precise when comparing oranges (lower GI) and white bread (higher estimates of the GI and GL of individual foods. The GI), where the lower GI food had the higher satiety rating. introduction of instruments for assessing dietary intake in Porridge and natural muesli had similar glycemic and epidemiological studies that have been designed to include insulinemic scores, but porridge had a greater satiety index more direct measures of GI and GL will enhance the than muesli (Po0.001). These results suggest that there is confidence in findings from such studies. Despite these little or no relationship between GI and satiety, at least when reservations it does appear that distinguishing between foods comparing food portions of equal energy content. Rather, with appreciable differences in the indices may produce energy density appeared to be inversely related to satiety, some benefit in terms of glycemic control in diabetes and presumably because of the high bulk required to obtain a lipid management. However, caution should be exercised in serving containing 1000 kJ when low energy-dense foods food choice based solely on GI or GL because low GI and GL were tested. When iso-energetic, iso-volumetric carbohy- foods may be energy dense and contain substantial amounts drate-containing beverages were tested, high GI beverages of sugars or undesirable fatty acids that contribute to the resulted in lower energy intakes during a subsequent meal, diminished glycemic response but not necessarily to good while low GI beverages were found not to suppress appetite health outcomes. This may apply especially to some of the and food intake in the short-term (Anderson et al., 2002). A manufactured products that have been GI and GL tested and review of the effect of glycemic carbohydrates on short-term are available in many countries. Given that most of the satiety has been published (Anderson and Woodend, 2003). studies which have demonstrated a health benefit of low GI One conclusion was that high GI carbohydrates suppress and GL involved the use of naturally occurring and short-term (1 h) food intake more effectively than low GI minimally processed foods it would seem to be appropriate carbohydrates, whereas low GI carbohydrates appeared to be for such products to be further tested for their health benefits more effective over longer periods (6 h). directly, rather than on the basis of their functionality (that How dietary GI and GL affects satiety and food intake over is, a low glycemic response). Although some data suggest a number of years is not entirely clear. The effectiveness of that the low GI effect is not explained by the dietary fibre dietary GI and GL on weight loss or maintenance is covered content of the foods it remains conceivable that food by van Damm in this series. The results of several observa- structure or composition explain some of the health tional studies have shown little difference in body mass benefits. GI may be a useful indicator to guide food choice index (BMI) across categories of GI and GL (Salmeron et al., if for example bread with a high GI is replaced on a slice-for- 1997a, b; Hodge et al., 2004; Schulze et al., 2004). Murakami slice basis with a lower GI bread, thereby achieving a lower et al. (2006) found a positive association between GI and GL. However, the complexity of the relationship between GI body mass index in Japanese female farmers, but no and GL is probably not well understood whereby GI and the association between GL and body mass index. On the other amount of a food eaten are both important determinants of hand, Ford and Liu reported inverse associations between GI the postprandial glycemic response. For the present it would and GL and body mass index in a nationally representative seem appropriate that when GI or GL are used to guide food sample of US adults (Ford and Liu, 2001). These contra- choice, it should only be done in the context of other dictory findings might suggest that dietary GI and GL is not nutritional indicators and when values have been measured a major determinant of dietary energy intake over the long- in a large group of individuals. term. A plausible reason is that GI appears not to be related to energy density. Potatoes and lentils for example represent foods with widely differing GIs but comparable energy Acknowledgements densities of around 3–4 kJ/g. On the other hand, cakes, cookies and fresh oranges have similar GIs in the low to We wish to thank Dr Jennie Brand-Miller, Professor Gary medium range, but energy densities some 10-fold different Frost, Professor Philip James, Professor Simin Liu, Professor (Holt et al., 1996). Jim Mann, Dr Gabriele Riccardi, Dr M Robertson and Professor HH Vorster for their valuable comments.

Recommendations Conflict of interest During the preparation and peer-review of this paper in The FAO/WHO Report on Carbohydrates in Human Nutri- 2006, the authors and peer-reviewers declared the following tion suggests that the concept of GI provides a useful means interests. of selecting the most appropriate carbohydrate containing foods for the maintenance of health and the treatment of Authors several disease states (FAO, 1998). Since the publication of Dr Bernard J Venn: None declared.

European Journal of Clinical Nutrition Glycemic index and glycemic load BJ Venn and TJ Green S130 Dr Tim Green: Affiliated with GI Otago, a commercial National Cancer Institute Diet History Questionnaire database. glycemic index testing service. J Am Diet Assoc 106, 393–402. Ford ES, Liu S (2001). Glycemic index and serum high-density lipoprotein cholesterol concentration among US adults. Arch Peer-reviewers Intern Med 161, 572–576. Dr Jennie Brand-Miller: Publishing books in the popular Foster-Powell K, Holt SH, Brand-Miller JC (2002). International table press: ‘The New Glucose Revolution Series’; Director of a of glycemic index and glycemic load values: 2002. Am J Clin Nutr University-based service for GI testing; Director of a not-for- 76, 5–56. Freeman J (2005). The glycemic index debate: does the type of profit food-labelling programme based on the GI. carbohydrate really matter? Diabet Forecast 58, 11. Professor Gary Frost: None declared. Frost G, Leeds A, Trew G, Margara R, Dornhorst A (1998). Insulin Professor Philip James: None declared. sensitivity in women at risk of coronary heart disease and the Professor Simin Liu: None declared. effect of a low glycemic diet. Metabolism 47, 1245–1251. Professor Jim Mann: None declared. Goldstein DE, Little RR, Lorenz RA, Malone JI, Nathan DM, Peterson CM (2004). Tests of glycemia in diabetes. Diabetes care 27 Dr Gabriele Riccardi: None declared. (Suppl 1), S91–S93. Dr M Robertson: Research Grant from National Chemical Granfeldt Y, Liljeberg H, Drews A, Newman R, Bjorck I (1994). and Starch. Glucose and insulin responses to barley products: influence of Professor HH Vorster: Member and Director of the Africa food structure and amylose-amylopectin ratio. Am J Clin Nutr 59, 1075–1082. Unit for Transdisciplinary health Research (AUTHeR), Re- Granfeldt Y, Hagander B, Bjorck I (1995). Metabolic responses to search grant from the South African Sugar Association. starch in oat and wheat products. On the importance of food structure, incomplete gelatinization or presence of viscous dietary fibre. Eur J Clin Nutr 49, 189–199. Hanefeld M, Cagatay M, Petrowitsch T, Neuser D, Petzinna D, Rupp M (2004). Acarbose reduces the risk for myocardial infarction in References type 2 diabetic patients: meta-analysis of seven long-term studies. Eur Heart J 25, 10–16. Amano Y, Kawakubo K, Lee JS, Tang AC, Sugiyama M, Mori K (2004). Heilbronn LK, Noakes M, Clifton PM (2002). The effect of high- and Correlation between dietary glycemic index and cardiovascular low-glycemic index energy restricted diets on plasma lipid and disease risk factors among Japanese women. Eur J Clin Nutr 58, glucose profiles in type 2 diabetic subjects with varying glycemic 1472–1478. control. J Am Coll Nutr 21, 120–127. Anderson GH, Catherine NL, Woodend DM, Wolever TM (2002). Henry CJ, Lightowler HJ, Kendall FL, Storey M (2006). The impact of Inverse association between the effect of carbohydrates on blood the addition of toppings/fillings on the glycaemic response glucose and subsequent short-term food intake in young men. Am to commonly consumed carbohydrate foods. Eur J Clin Nutr 60, J Clin Nutr 76, 1023–1030. 763–769. Anderson GH, Woodend D (2003). Effect of glycemic carbohydrates Henry CJ, Lightowler HJ, Strik CM, Storey M (2005). Glycaemic index on short-term satiety and food intake. Nutr Rev 61, S17–S26. values for commercially available potatoes in Great Britain. Br J Brand-Miller J, Hayne S, Petocz P, Colagiuri S (2003b). Low-glycemic Nutr 94, 917–921. index diets in the management of diabetes: a meta-analysis of Hodge AM, English DR, O’Dea K, Giles GG (2004). Glycemic index randomized controlled trials. Diabetes Care 26, 2261–2267. and dietary fiber and the risk of type 2 diabetes. Diabetes Care 27, Brand-Miller JC, Holt SH, Pawlak DB, McMillan J (2002). Glycemic 2701–2706. index and obesity. Am J Clin Nutr 76, 281S–285S. Holt SH, Miller JC, Petocz P (1997). An insulin index of foods: the Brand-Miller JC, Thomas M, Swan V, Ahmad ZI, Petocz P, Colagiuri S insulin demand generated by 1000-kJ portions of common foods. (2003c). Physiological validation of the concept of glycemic load Am J Clin Nutr 66, 1264–1276. in lean young adults. J Nutr 133, 2728–2732. Holt SH, Miller JC, Petocz P, Farmakalidis E (1995). A satiety index of Brand-Miller JC, Wolever TM (2005). The use of glycaemic index common foods. Eur J Clin Nutr 49, 675–690. tables to predict glycaemic index of breakfast meals. Br J Nutr 94, Holt SH, Brand Miller JC, Petocz P (1996). Inter-relationships among 133–134. postprandial satiety, glucose and insulin responses and changes in Brand-Miller J, Wolever TM, Foster-Powell K, Colagiuri S (2003a). The New Glucose Revolution. Marlowe & Company: New York. subsequent food intake. Eur J Clin Nutr 50, 788–797. Brouns F, Bjorck I, Frayn KN, Gibbs AL, Lang V, Slama G et al. (2005). Jenkins DJ, Goff DV, Leeds AR, Alberti KG, Wolever TM, Gassull MA Glycaemic index methodology. Nutr Res Rev 18, 145–171. et al. (1976). Unabsorbable carbohydrates and diabetes: decreased Collier G, O’Dea K (1983). The effect of coingestion of fat on the post-prandial hyperglycaemia. Lancet 2, 172–174. glucose, insulin, and gastric inhibitory polypeptide responses to Jenkins DJ, Kendall CW, Augustin LS, Franceschi S, Hamidi M, carbohydrate and protein. Am J Clin Nutr 37, 941–944. Marchie A et al. (2002). Glycemic index: overview of implications Coulston AM, Hollenbeck CB, Reaven GM (1984). Utility of studies in health and disease. Am J Clin Nutr 76, 266S–273S. measuring glucose and insulin responses to various carbohydrate- Jenkins DJ, Wolever TM, Buckley G, Lam KY, Giudici S, Kalmusky J containing foods. Am J Clin Nutr 39, 163–167. et al. (1988). Low-glycemic index starchy foods in the diabetic diet. Doi K, Matsuura M, Kawara A, Baba S (1979). Treatment of diabetes Am J Clin Nutr 48, 248–254. with glucomannan (konjac mannan). Lancet 1, 987–988. Jenkins DJ, Wolever TM, Jenkins AL, Giordano C, Giudici S, FAO (1998). Carbohydrates in human nutrition. Report of a Joint Thompson LU et al. (1986). Low-glycemic response to traditionally FAO/WHO Expert Consultation. FAO Food Nutr Pap 66, processed wheat and rye products: bulgur and pumpernickel 1–140. bread. Am J Clin Nutr 43, 516–520. Flint A, Moller BK, Raben A, Pedersen D, Tetens I, Holst JJ et al. Jenkins DJ, Wolever TM, Jenkins AL, Josse RG, Wong GS (1984). (2004). The use of glycaemic index tables to predict glycaemic The glycaemic response to carbohydrate foods. Lancet 2, index of composite breakfast meals. Br J Nutr 91, 979–989. 388–391. Flood A, Subar AF, Hull SG, Zimmerman TP, Jenkins DJ, Schatzkin A Jenkins DJ, Wolever TM, Jenkins AL, Thorne MJ, Lee R, Kalmusky J (2006). Methodology for adding glycemic load values to the et al. (1983). The glycaemic index of foods tested in diabetic

European Journal of Clinical Nutrition Glycemic index and glycemic load BJ Venn and TJ Green S131 patients: a new basis for carbohydrate exchange favouring the use Salmeron J, Manson JE, Stampfer MJ, Colditz GA, Wing AL, Willett of legumes. Diabetologia 24, 257–264. WC (1997b). Dietary fiber, glycemic load, and risk of non-insulin- Jenkins DJ, Wolever TM, Taylor RH, Barker H, Fielden H, Baldwin JM dependent diabetes mellitus in women. JAMA 277, 472–477. et al. (1981). Glycemic index of foods: a physiological basis for Schenk S, Davidson CJ, Zderic TW, Byerley LO, Coyle EF (2003). carbohydrate exchange. Am J Clin Nutr 34, 362–366. Different glycemic indexes of breakfast cereals are not due to Kelly S, Frost G, Whittaker V, Summerbell C (2004). Low-glycaemic glucose entry into blood but to glucose removal by tissue. Am index diets for coronary heart disease. Cochrane Database Syst Rev J Clin Nutr 78, 742–748. CD004467. Schulze MB, Liu S, Rimm EB, Manson JE, Willett WC, Hu FB (2004). Liljeberg H, Granfeldt Y, Bjorck I (1992). Metabolic responses to Glycemic index, glycemic load, and dietary fiber intake and starch in bread containing intact kernels versus milled flour. Eur incidence of type 2 diabetes in younger and middle-aged women. J Clin Nutr 46, 561–575. Am J Clin Nutr 80, 348–356. Liu S, Manson JE, Stampfer MJ, Holmes MD, Hu FB, Hankinson SE Slyper A, Jurva J, Pleuss J, Hoffmann R, Gutterman D (2005). et al. (2001). Dietary glycemic load assessed by food-frequency Influence of glycemic load on HDL cholesterol in youth. Am J Clin questionnaire in relation to plasma high-density lipoprotein Nutr 81, 376–379. cholesterol and fasting plasma triacylglycerols in postmenopausal Sugiyama M, Tang AC, Wakaki Y, Koyama W (2003). Glycemic women. Am J Clin Nutr 73, 560–566. Ma Y, Li Y, Chiriboga DE, Olendzki BC, Hebert JR, Li W et al. (2006). index of single and mixed meal foods among common Japanese Association between carbohydrate intake and serum lipids. JAm foods with white rice as a reference food. Eur J Clin Nutr 57, Coll Nutr 25, 155–163. 743–752. Mann J (2007). Dietary carbohydrate: relationship to cardiovascular Suzuki H, Fukushima M, Okamoto S, Takahashi O, Shimbo T, Kurose disease and disorders of carbohydrate metabolism. Eur J Clin Nutr T et al. (2005). Effects of thorough mastication on postprandial 61(Suppl 1), S100–S111. plasma glucose concentrations in nonobese Japanese subjects. McAuley K, Mann J (2006). Thematic review series: patient-oriented Metabolism 54, 1593–1599. research. Nutritional determinants of insulin resistance. J Lipid Res Tappy L, Gugolz E, Wursch P (1996). Effects of breakfast cereals 47, 1668–1676. containing various amounts of beta-glucan fibers on plasma Mettler S, Lamprecht-Rusca F, Stoffel-Kurt N, Wenk C, Colombani PC glucose and insulin responses in NIDDM subjects. Diabetes Care (2007). The influence of the subjects’ training state on the 19, 831–834. glycemic index. Eur J Clin Nutr 61, 19–24. van de Laar FA, Lucassen PL, Akkermans RP, van de Lisdonk EH, Miller CK, Gabbay RA, Dillon J, Apgar J, Miller D (2006). The effect of Rutten GE, van Weel C (2005). Alpha-glucosidase inhibitors for three snack bars on glycemic response in healthy adults. J Am Diet patients with type 2 diabetes: results from a Cochrane systematic Assoc 106, 745–748. review and meta-analysis. Diabetes Care 28, 154–163. Murakami K, Sasaki S, Takahashi Y, Okubo H, Hosoi Y, Horiguchi H Venn BJ, Wallace AJ, Monro JA, Perry T, Brown R, Frampton C et al. et al. (2006). Dietary glycemic index and load in relation to (2006). The glycemic load estimated from the glycemic index does metabolic risk factors in Japanese female farmers with traditional not differ greatly from that measured using a standard curve in dietary habits. Am J Clin Nutr 83, 1161–1169. healthy volunteers. J Nutr 136, 1377–1381. Neuhouser ML, Tinker LF, Thomson C, Caan B, Horn LV, Snetselaar L Wolever TM, Jenkins DJ (1986). The use of the glycemic index in et al . (2006). Development of a glycemic index database for food predicting the blood glucose response to mixed meals. Am J Clin frequency questionnaires used in epidemiologic studies. J Nutr Nutr 43, 167–172. 136, 1604–1609. Wolever TM, Vorster HH, Bjorck I, Brand-Miller J, Brighenti F, Mann Nuttall FQ, Mooradian AD, Gannon MC, Billington C, Krezowski P JI et al. (2003). Determination of the glycaemic index of foods: (1984). Effect of protein ingestion on the glucose and insulin interlaboratory study. Eur J Clin Nutr 57, 475–482. response to a standardized oral glucose load. Diabetes Care 7, 465–470. Wolever TM (1990). The glycemic index. World Rev Nutr Diet 62, Opperman AM, Venter CS, Oosthuizen W, Thompson RL, Vorster HH (2004). Meta-analysis of the health effects of using the glycaemic 120–185. index in meal-planning. Br J Nutr 92, 367–381. Wolever TM, Vuksan V, Eshuis H, Spadafora P, Peterson RD, Chao ES Pi-Sunyer FX (2002). Glycemic index and disease. Am J Clin Nutr 76, et al. (1991). Effect of method of administration of psyllium on 290S–298S. glycemic response and carbohydrate digestibility. J Am Coll Nutr Read NW, Welch IM, Austen CJ, Barnish C, Bartlett CE, Baxter AJ 10, 364–371. et al. (1986). Swallowing food without chewing; a simple way to Wolever TM, Yang M, Zeng XY, Atkinson F, Brand-Miller JC reduce postprandial glycaemia. Br J Nutr 55, 43–47. (2006). Food glycemic index, as given in glycemic index Salmeron J, Ascherio A, Rimm EB, Colditz GA, Spiegelman D, Jenkins tables, is a significant determinant of glycemic responses DJ et al. (1997a). Dietary fiber, glycemic load, and risk of NIDDM elicited by composite breakfast meals. Am J Clin Nutr 83, in men. Diabetes Care 20, 545–550. 1306–1312.

European Journal of Clinical Nutrition