Different GI values for different groups were found when a glucose beverage was used as the reference when testing starchy foods. These findings are contrary to the basic concept that the index represents a property of the food independent of the consumer . Indeed, it has been reported that GI was not different among groups comprising people with impaired glucose tolerance; diabetes mellitus; lean with normal glucose tolerance; and obese with normal glucose tolerance . Thus, although the BMI of our Chinese and European groups differed, and the BMI of our younger and older participants differed, this should not have affected the GI. We confirmed this using the Chinese and European data in a regression analysis whilst adjusting for BMI in which the statistical significance and the inference were unchanged (data not shown). Additionally, if BMI was postulated to affect GI, a higher BMI was associated with a lower GI in the Europeans whilst in the elderly, the opposite was the case; a higher BMI was associated with a higher GI. Thus, we consider it unlikely that BMI provides an explanation for the observed differences in group GI.
In the series of tests presented here, GI differences for the same food differed by as much as 10 – 17 GI units when considering different ethnic groups or age categories. Such a difference is substantial given that there is only 15 GI unit difference between the nominal ‘high’ (≥70) and ‘low’ (≤55) GI classification. The extent of the group differences even affected the classification. For example, cornflakes was medium GI in the younger group and high GI in the older group. For the European group, the varieties of rice we tested would generally be classified as medium GI with the confidence intervals in the low to medium GI range whereas in the Chinese, rice would be classified as a medium to high GI food. Wolever and colleagues, using a glucose reference, reported a similar finding in which white bread was medium GI in a group of Caucasians (GI = 66) and high GI in non-Caucasians (GI = 78) . Identifying a problem with the reference food suggest that changes should be made to the testing methodology of starchy foods to remove this source of variability, thereby enabling the consistent categorization of GI, necessary both for dietary advice and for generalizability on food labels.
Different GIs for the same food are also found in the International Tables of GI . For example, under ‘boiled white Basmati rice’ the GI values range from 43 to 69. Some of this variability may be due to the growing, processing and cooking methods as alluded to in the preface of the International Tables. However, despite standardization by supplying a single source of long grain white rice with cooking instructions to seven GI testing centres around the world, the GI of the rice was still variable ranging from 55 to 87 . It is of note that age and ethnic composition of the groups differed among the centres although it was stated that these had no significant effect on GI. Some of the variability could have been due to laboratory differences in GI testing protocol. In our series of experiments in which rice was tested in Chinese and European groups, the source of rice, cooking and laboratory methods were all standardised. Similarly, in the young and older participants, the breakfast cereals and testing procedures were the same for both groups. Hence, the observed differences in GI when using a glucose beverage as a reference food are attributable to differences in group characteristics.
An effect of group characteristics on GI was apparent when people were dichotomized as having high or low salivary alpha-amylase activity . A significant between-group difference in postprandial glucose response was noted after participants had consumed a corn starch solution but there was no between-group difference in response to the glucose reference beverage. From an earlier observation, Wolever and colleagues had proposed that the use of a glucose beverage may lead to false conclusions regarding the relevance of observed ethnic differences in GI if starch digestion differed by ethnicity . We measured salivary alpha-amylase concentrations in our Chinese and European groups and found these did not explain the different GIs when glucose was used as a reference . However, the majority of starch digestion occurs in the small intestine via the action of pancreatic alpha-amylase  with heterogeneity noted in the entire amylase gene family . Although variability in alpha-amylase could plausibly account for differing rates of starch digestion between Caucasians and non-Caucasians, and between groups having different salivary amylase characteristics, we have found no reports suggesting that alpha-amylase is raised in older people. One possibility is that the elderly are more prone to hypochlorhydria and achlorhydria . The activity of salivary amylase is pH dependent  such that a deficiency or lack of stomach acid may enable salivary alpha-amylase to continue working whilst the food is in the stomach.
The potential for different rates of starch digestion are indicative that a solid starchy food may be a physiologically more relevant reference than a glucose beverage when testing the GI of starchy foods. Rice has been used as a reference in a Japanese study due to availability and palatability  and although white bread used to be preferred , glucose was subsequently recommended as the reference . Consequently, in the International Tables of GI, the GI values of many starchy foods have been determined with reference to glucose . How generalizable those values are is unclear, with our work and that of others presented here being suggestive that for GI to have a universal value, regardless of ethnicity, age or other participant characteristics, that the choice of reference is important. A simple conversion from one reference scale to the other using a multiplicative constant does not resolve the problem. The factor we obtained for converting from a Jasmine rice scale to a glucose scale was 0.8 for the Chinese participants. This is the same value as that found by Sugiyama and colleagues converting from a rice scale to a glucose scale in a Japanese sample . However, the conversion factor for Jasmine rice to glucose in our European sample was 0.7; and this was significantly different to the conversion factor for the Chinese.
For sugary foods, the use of glucose as a reference may be appropriate. In a study by Wolever and colleagues, there were no ethnic differences in GI for chocolate chip cookies and fruit bars, notably with the majority of carbohydrate in those foods coming from sugars rather than starch . Similarly, for the sugary food we tested (LoGiCane™), there was no ethnic difference in GI, supporting the proposition that glucose would appear to be a suitable reference for sugary foods. In essence, glucose may be an appropriate reference for sugary foods but not for starchy foods. The opposite may also be true, that a starchy food would be an inappropriate reference for testing a sugary food. The GI of sucrose tended to be higher in the European group (88) than in the Chinese group (78) when rice was used as the reference (p = 0.19).
Interestingly, the rank order of the rice varieties we tested were largely maintained regardless of the reference food (Table 2). If GI were simply used to rank foods then the choice of reference may be of little concern. However, GI has moved on such that classifications (low, medium and high) and absolute numbers of GI now appear on food labels and in the public domain . For starchy foods tested with glucose as the reference, the applicability of those classifications and values to individuals is questionable and it would be unrealistic, and contrary to the GI concept, to test and label the GI of a food specific to particular subgroups within a population. It would appear that the use of a starchy reference when testing starchy food would avoid this problem, although at present Standards Australia specify glucose as the reference . The International Organization for Standards is more flexible and endorses the use of white bread or other reference foods .
The choice of reference, if any, may relate to the purpose of conducting the tests. For example, if we were simply interested in recommending one type of rice over another based on glycemic response then conclusions could be reached by inspection of the AUCs directly (Figure 1) without the need of a ‘reference’. This is really only valid if the same people are testing all of the foods. On the other hand, if the purpose was to make numerical comparisons among studies in which different groups of participants had been involved, then for comparability it would be desirable to settle on a standard reference for the testing of starchy foods. White bread has a history of use and although the composition may differ somewhat from location to location, in an international comparison the GI values obtained by different testing centers for white bread obtained locally were no more variable than for other starchy foods that had been provided from a single source . A limitation of our conclusion that starchy foods require a starchy reference is the small number of foods tested by ourselves (five rice varieties and two breakfast cereals) and others (white bread)  and corn starch . There is also a limited number of comparison groups (Chinese vs European; younger vs older; Caucasian vs non-Caucasian; higher vs lower salivary alpha-amylase excreters). A dependence of GI on a particular reference food may be variable among groups, for example between other ethnic groups or groups with a different age separation, thus limiting the generalizability of our findings. It would be of interest to test whether converting from one starchy reference (eg: rice) to another starchy reference (eg: white bread) would be independent of participant characteristics.