A cross-sectional study of food group intake and C-reactive protein among children

The NHANES are cross-sectional population-based surveys conducted by National Center for Health Statistics for the purpose of assessing the health and nutritional status of U.S. adults and children[14, 15]. The survey samples are selected using a stratified multistage probability design with random sampling of the civilian non-institutionalized population, with over-sampling of certain subgroups. Our analyses were conducted with the approval of our Institutional Review Board.

Of the 21,004 individuals who participated in 1999-2002 surveys, we included 5,857 children and adolescents, ages 5 to 16 years. Adolescents' ages 17 years and older were not included in the study since dietary habits during this early post-high school period are often unstable due to significant lifestyle changes [16, 17]. We excluded 539 subjects with missing or unreliable dietary data as well as 204 individuals with extreme energy intake (upper and lower 2% of the distribution). We further excluded subjects meeting any of the following criteria: missing CRP data (n = 590), pregnant or breastfeeding (n = 15), on dialysis or taking oral medications or insulin for diabetes (n = 11), missing data on covariates (n = 204), and those not able to be classified into one of primary racial groups (n = 107). Finally, to eliminate illness as a possible cause of elevated CRP, we excluded 77 subjects (mean CRP = 7.02 mg/L) whose white blood cell count was above the age- and sex-specific upper limit of normal as specified in the NHANES laboratory manuals[18]. The final study population included 4,110 children and adolescents.

From 1999 to 2001 dietary intake was assessed in NHANES with a single, multipass 24-hour recall, carried out using a computer-assisted dietary interview system that has been previously described[19]. A revised five step automated multiple-pass instrument was used starting in 2002[20]. Participants aged 12 years and older completed the interview independently. A proxy respondent responded for five-year olds while children ages 6 to11 years completed the interview together with a proxy respondent.

The dietary recalls were analyzed by the USDA using nutrient files that were appropriate to the time period of data collection[21, 22]. Food Pyramid servings were estimated for each subject as defined by the USDA's Dietary Guidelines[23] (see additional file 1).

Blood samples were collected from all subjects, stored at -20 degrees centigrade and then shipped to the University of Washington Medical Center for analysis. The CRP concentration in serum was measured by a high-sensitivity assay, latex enhanced nephelometry, on a Dade Behring Nephelometer II Analyzer System. The analytical performance of this test has been validated[24] and shown to be comparable to that of ELISA[25].

We considered the following potential confounding variables in our analysis: exact age, race/ethnicity, height (in meters), height-for-age Z-Scores (HAZ), socioeconomic status (SES), body mass index (BMI) and BMI-Z Scores, waist circumference, sedentary behavior (hours of TV watching and video and computer time per day), energy intake (in kilo calories), percentage (%) of calories from carbohydrates, protein and fat and fiber intake per 1000 kilocalories of energy.

Racial/ethnic origin was classified into three categories: non-Hispanic whites, non-Hispanic blacks and Hispanics (i.e., Mexican-Americans). SES was estimated using combined data on income and food insecurity. Weight was measured on a Toledo Digital Scale and standing height (to the nearest 0.1 centimeter) using a fixed stadiometer. Waist circumference was measured with a steel tape (to the nearest 0.1 centimeter) at the high point of iliac crest at the end of normal expiration[26]. BMI was calculated as weight in kilograms divided by the square of height, in meters. CDC growth charts were used to calculate age and sex-specific BMI Z-scores and height-for-age Z-scores using the Lambda, Mu, Sigma (LMS) method[27].

The goal of the analysis was to evaluate the association between serum CRP levels at 5 to16 years of age and dietary intake in the USDA food pyramid groups. Since NHANES provides only a single 24-hour dietary recall and single recalls may provide reliable estimates of population group means rather than individual means, we chose to estimate the adjusted food group means according to the child's CRP group assignment[28]. CRP was classified into one of three categories that have been previously shown to reflect low, average and high risk of future CVD[29]: <1.0 mg/L, 1.0-3.0 mg/L and >3.0 mg/L, respectively. Adjusted mean intakes of dairy, grains, fruit, vegetables and meat/other proteins were estimated in each of these categories using analysis of covariance modeling with Proc GLM.

All analyses were carried out using SAS, Version 9.1 (SAS Institute, Cary, NC). We used Proc Surveyreg to calculate adjusted mean differences in food group intake by CRP levels accounting for stratification and clustering within primary sampling units. Since we adjusted for the covariates representing the over-sampled subgroups, we did not adjust for sampling weights in our analysis; this adjustment is regarded as a good compromise between efficiency and bias[30]. The results of the Surveyreg Procedure were almost identical to the simpler GLM modeling, leading us to retain the GLM modeling in the results of this manuscript.

The weak but beneficial association between CRP level and dairy consumption in the baseline analysis was attenuated by controlling for either BMI Z-score (p for linear trend changed from p = 0.047 to p = 0.304) or waist circumference (changed from p = 0.047 to p = 0.770). There was only a small amount of attenuation of the associations of grains and vegetables with CRP level. In the final analysis in this table, all other food groups were added to the baseline multivariable models. In this analysis, only an independent association remained for grains and vegetables with CRP. In contrast, the inclusion of energy-adjusted macronutrients and fiber to the models did not fully attenuate the observed associations between any of the food groups and CRP in these children.

The associations of food subgroups and CRP levels are shown in Table 3. A number of the food subgroups here are consumed infrequently by these children and adolescents (e.g., yogurt, whole grains, dark green/leafy vegetables, deep yellow/orange vegetables), giving us little statistical power to examine their relationship with CRP. The milk subgroup was inversely associated with CRP (p = 0.040) while cheese consumption was not. Refined grain intake appeared to be beneficially associated with CRP (p < 0.0001). Other food subgroups found to be beneficially related to CRP level included citrus, melons, and berries (p = 0.053), tomatoes (p = 0.001), total non-starchy vegetables (p = 0.004), and total starchy vegetables (p = 0.012). Neither total meat nor any of the meat/other protein subgroups were associated with CRP levels.