We explored multiple and novel measures of dietary glycemic carbohydrate in a group of older Australian women to provide insight into which variables were most predictive of an association with insulin resistance. Mean intakes for all glycemic measures were consistently higher in women who were insulin resistant compared to those who were not insulin resistant. Contrary to our hypothesis however, the largest contributors to prediction of insulin resistance, independently of other risk factors, were the daily glycemic measures rather than meal based measures (Table 2). Our data suggest GL/Mcal was the glycemic variable that had the highest contribution to the prediction of insulin resistance. Adjusting for overall energy intake, in the form of GL/Mcal, increased the predictive ability with insulin resistant status compared to the GL unadjusted. This could be especially important when examining dietary glycemic intake in relation to disease risk in more diverse population groups with wider variations in energy requirements due to gender, age or physical activity. The GL/Mcal became more clinically important in subjects who were older, sedentary or had a larger waist circumference, as these factors potentiated the effect (Figure 1).
The difference in predictive ability of the four highest contributing glycemic variables (GL/Mcal, GL, carbohydrate (g) and carbohydrate percentage) was minor, with a separation of only 0.006 in Nagelkerke's R2 coefficients between the first and fourth placed variables. These four glycemic variables were also the highest contributors per standard deviation to prediction of HOMA as a continuous variable. As carbohydrate is used to calculate GL, this was partly due to colinearity between the measures. The advantage of using the GL over carbohydrate in prediction of insulin resistance in this population is slight. Therefore in practice, the prescription of carbohydrate reduction may be simpler to follow than reducing GL, as long as intakes of dietary fibre, vitamins, and minerals remain nutritionally adequate. The type of fat used to replace carbohydrate should also be considered, as an increase in saturated fat has the potential to raise low-density lipoprotein cholesterol .
The lack of significant associations when using HOMA as a continuous variable suggests that relationships between insulin resistance and glycemic intake variables may be more noticeable when subjects with higher HOMA values are identified as a separate group. Causality cannot be inferred due to the cross-sectional design, however our results support further investigation of the effect of dietary glycemic intake on development of insulin resistance.
A unique aspect of this study was investigation of the association between insulin resistance and dietary glycemic carbohydrate on a meal basis. Post-prandial glucose response can vary with the time of day. In previous studies, plasma glucose concentrations were higher after breakfast than lunch, despite a lower GL for the breakfast meal  or when both breakfast and lunch contained the same type and amount of cereal . In our study, lunch GL but not breakfast or dinner GL, was a significant predictor of insulin resistance when adjusted for confounding factors. The relevance of lunch per se is not clear; however evidence for the possible importance of lunch in relation to risk of chronic disease was also seen in a five-year prospective study, in which blood glucose concentrations after lunch predicted cardiovascular events in women with diabetes .
We also used the analysis of GL on a meal basis to conceptualise the GL peak score, a new measure designed to represent the magnitude of glycemic peaks at meals during the day. The magnitude of postprandial hyperglycemia has been shown to influence the degree of oxidation of low-density lipoprotein cholesterol in subjects with diabetes  and oxidative stress in healthy subjects . For comparison with a disease outcome such as insulin resistance, it was hypothesised that a measure of hyperglycemic spikes could be a more relevant predictor of the glycemic potential than the daily measures used in previous studies. We designed the GL peak score to express the magnitude of GL peaks during the day, relative to each subject's mean dietary GL, and to quantify high peaks across all meals, not just the single meal with the highest peak. While our data indicated that a dietary pattern consisting of relatively high peaks was a significant independent predictor of insulin resistance, the contribution was less than that of the dietary GL. Depending on the degree of fluctuation of meal GL values, a subject with a high GL peak score can have a low or a high dietary GL, as evident in the relatively low correlation we observed between these two measures (r = 0.42). However, the potential impact of these alternatives on insulin resistance is not clear from our data. The use of mean meal GL to calculate the GL peak score may also have misrepresented subjects who alternated between missing meals and consuming large meals (for example, at breakfast). While the technique we developed to calculate GL peak score was carefully considered, further research may suggest another technique that provides a more optimal measure of hyperglycemic peaks for comparison with chronic disease status.
A strength of our study was the comprehensive assessment of usual dietary glycemic intake by diet history interviews. This allowed us to quantify individual carbohydrate serve sizes to optimise the precision in our estimates of dietary glycemic carbohydrate, and provided data on meal patterns. A possible limitation to application of this research to the wider population is that the BMI and waist circumference of subjects in this analysis were lower than other subjects in the larger LAW study (Table 1), in part due to exclusion of women with diabetes for the current analyses.