The weighted response rate for the NSHAP study was 75.5% (74.8% unweighted). Distributions for the variables considered in our analyses are presented in . The demographic characteristics of the subset of respondents with complete sociodemographic and CRP data closely match those of respondents to the 2002 Current Population Survey (U.S. Census Bureau, 2003
) as is true for the NSHAP sample overall (Lindau et al., 2007
). The median blood spot CRP concentration was 1.5 mg/L (25th percentile: 0.6 and 75th percentile: 3.5). Individuals with CRP >8.6 mg/L excluded from subsequent analyses did not differ significantly from the rest of the sample with respect to age. However, Blacks compared with Whites (odds ratio [OR] = 2.95, 95% confidence interval [CI] = 1.84–4.72, p
< .01) as well as females (OR = 1.96, 95% CI = 1.31–2.95, p
< .01) and individuals with lower levels of educational attainment (OR = 1.42, 95% CI = 1.21–1.68, p
< .01) and higher waist circumference (OR = 1.03; 95% CI = 1.02–1.04, p
< .01) were more likely to be excluded due to acute elevations in CRP.
Table 1. Descriptive Statistics for Entire Sample (N = 1,701). Population Mean (SE) Values are presented for continuous variables; population proportions (SE) are presented for categorical variables. Median (25%–75%) values are presented for CRP. All estimates (more ...)
CRP concentration was significantly higher in non-Hispanic Black participants compared with Whites and among women compared with men (, Model 1). However, these differences were substantially attenuated when education and assets were considered (, Model 2). Higher educational attainment was associated with reduced CRP. Higher levels of household assets were also associated with lower CRP, but this association was significant only for individuals with the highest levels of assets (>$500,000). Compared with non-Hispanic Whites, non-Hispanic Blacks had marginally higher concentrations of CRP, and Hispanics had significantly lower concentrations. There was no significant gender difference in CRP after controlling for other sociodemographic factors.
Multiple Linear Regression Models Predicting Log-Transformed C-Reactive Protein Concentration.
We found no evidence for significant interactions between age and educational status (p = .57) or household assets (p = .18). However, associations between CRP and race/ethnicity were significantly moderated by age (p < .01). In the youngest age group, a similar pattern in CRP across race/ethnicity was seen as reported above, with higher CRP among non-Hispanic blacks (B = 0.105, SE = .077) and lower CRP among Hispanics (B = −0.197, SE = 0.069) compared with Whites. In the middle age group, the association between race/ethnicity and CRP was weaker (p = .07). In the oldest age group, differences in CRP concentrations across the racial/ethnic groups were further attenuated (p = .58).
We next investigated the extent to which associations between CRP and social status were independent of health behaviors, psychosocial factors, and medical factors that bivariate analyses revealed as significant predictors of CRP in our sample (, full model). Waist circumference and smoking (both currently and in the past) were positively associated with CRP. The interviewer’s rating of the cleanliness of the respondent’s household was also associated with CRP, with lower CRP concentrations for respondents living in a household evaluated as cleaner. The use of sex hormones and immunosuppressive medications were associated with elevated CRP, whereas statin use predicted lower CRP. Physical activity levels, sleep quality, alcohol use, history of heart failure or arthritis, depressive symptoms, and being married or cohabitating were significantly associated with CRP in bivariate analyses but not in the fully adjusted model. Remaining behavioral, psychosocial, and medical factors were not significantly related to CRP in bivariate analyses.
Educational status remained as a significant predictor of CRP in the fully adjusted model, although the strength of this association was substantially reduced. The interaction between race/ethnicity and age was also statistically significant in the full model (p < .01). Level of assets was not a significant predictor of CRP after controlling for more proximate behavioral, psychosocial, and medical factors.