A total of 641 participants with 2,795 observations were available for analysis. Because our goal was to examine the longitudinal effect of dietary GI or GL on hs-CRP, participants with data available for less than 2 times point were excluded (n=49 observations). An additional 617 observations were excluded because both dietary measurements and hs-CRP were not available at the same quarter. We excluded 65 observations where hs-CRP was greater than 10 mg/L because such elevated levels are likely to be caused by an acute infection or underlying medical problem not related to diet [24
]. We also excluded one observation with an extreme outlier; GL= 1085. Therefore a total of 582 participants and 2063 observations remained for analysis. Hs-CRP values were highly skewed, therefore, the data were analyzed using log transformed values for hs-CRP.
presents participant characteristics. Participants were predominantly white (86%), with an average age of 48 years and approximately equal distributions of males and females. Average BMI was 27.4 kg/m2, and 64% of the participants were either overweight or obese. Average and median hs-CRP values were, 1.84 mg/L and 1.21 mg/L, respectively (). For this paper GI is reported using white bread as the reference, where white bread equals a GI score of 100. The average daily dietary GI score was 85, considered to be in the intermediate range for GI, with values ranging from 49 to 101 (). The average GL value was 198, considered to be in the high range for GL, with a minimum of 45 and a maximum value of 489 (). All-purpose flour, white sugar, white bread, white rice and cola beverage, were the top contributors to GL these five foods alone account for a cumulative GL of 52. The graphs in figures one and two suggested a slightly inverse relationship between average GI or GL and hs-CRP levels (, ).
Selected characteristics of the study population.
Mean median and range of average high sensitivity CRP, average dietary glycemic index and average dietary glycemic load.
Scatterplot between Average CRP and Average GI
Scatterplot between Average CRP and Average GL
We used a multivariable linear mixed model to examine both the cross-sectional and longitudinal relationship between log hs-CRP and GI or GL (). We found no association between GI and log hs-CRP in either the cross-sectional (regression coefficient (β)=0.009, p=0.24) or the longitudinal analysis (β=-0.002, p= 0.39). We did however observe the suggestion of an inverse association between GL and log hs-CRP in the cross-sectional analyses, but no association in the longitudinal analyses. Specifically, the coefficient for the cross-sectional effect of GL (β= -0.00194) was suggestive of an inverse relationship with log hs-CRP (p= 0.002).
Regression coefficients predicting log high sensitivity CRP from linear mixed models.
After statistically adjusting for BMI, smoking status, age and infection status, the cross-sectional findings for GL and hs-CRP are attenuated and no longer statistically significant (p=0.07). The longitudinal association was attenuated to -0.00012 (p= 0.72) (). None of the variables produced notable changes in the strength or direction of the estimate.
Regression coefficients predicting log high sensitivity CRP from multivariable linear mixed models.
We then stratified the analysis by BMI category and gender. Stratification by gender did not yield results of statistical significance (for cross-sectional results for GL; men β=-0.0055 p=0.43, women β= -0.00141 p=0.18). When stratified by BMI, we found that mean GL was a significant predictor of hs-CRP in the cross-sectional analysis only among obese individuals (β= -0.00185 p=0.04) ().
Regression coefficients predicting log high sensitivity CRP from multivariable linear mixed models for glycemic load stratified by BMI categories.
To further explore how BMI modified the relationship between dietary GL and hs-CRP, we fitted a linear model to clarify the interaction, without including any of the covariates. We then converted the results to the natural scale and plotted the predicted hs-CRP versus average GL (). An inverse relationship was seen across all strata, with individuals in the highest BMI category having the greatest reduction of hs-CRP. However as GL increased, there was no statistical difference between the slope of obese participants and the slope of the other BMI categories (p values>0.05).
Predicted HS-CRP vs. Average Glycemic Load by BMI Category