Methodological challenges in the application of the glycemic index in epidemiological studies using data from the European Prospective Investigation into Cancer and Nutrition.
van Bakel MM., Slimani N., Feskens EJ., Du H., Beulens JW., van der Schouw YT., Brighenti F., Halkjaer J., Cust AE., Ferrari P., Brand-Miller J., Bueno-de-Mesquita HB., Peeters P., Ardanaz E., Dorronsoro M., Crowe FL., Bingham S., Rohrmann S., Boeing H., Johansson I., Manjer J., Tjonneland A., Overvad K., Lund E., Skeie G., Mattiello A., Salvini S., Clavel-Chapelon F., Kaaks R.
Associations between the glycemic index (GI) or glycemic load (GL) and diseases are heterogeneous in epidemiological studies. Differences in assigning GI values to food items may contribute to this inconsistency. Our objective was to address methodological issues related to the use of current GI and GL values in epidemiological studies. We performed ecological comparison and correlation studies by calculating dietary GI and GL from country-specific dietary questionnaires (DQ) from 422,837 participants from 9 countries participating in the European Prospective Investigation into Cancer and Nutrition study and single standardized 24-h dietary recalls (24-HDR) obtained from a representative sample (n = 33,404) using mainly Foster Powell's international table as a reference source. Further, 2 inter-rater and 1 inter-method comparison were conducted, comparing DQ GI values assigned by independent groups with values linked by us. The ecological correlation between DQ and 24-HDR was good for GL (overall r = 0.76; P < 0.005) and moderate for GI (r = 0.57; P < 0.05). Mean GI/GL differences between DQ and 24-HDR were significant for most centers. GL but not GI from DQ was highly correlated with total carbohydrate (r = 0.98 and 0.15, respectively; P < 0.0001) and this was higher for starch (r = 0.72; P < 0.0001) than for sugars (r = 0.36; P < 0.0001). The inter-rater and inter-method variations were considerable for GI (weighted kappa coefficients of 0.49 and 0.65 for inter-rater and 0.25 for inter-method variation, respectively) but only mild for GL (weighted kappa coefficients > 0.80). A more consistent methodology to attribute GI values to foods and validated DQ is needed to derive meaningful GI/GL estimates for nutritional epidemiology.