Six hundred one participants in the study sample had complete information for age, race, prevalent chronic disease information, diabetes, smoking status, and GPx activity. Baseline characteristics are summarized in . Age ranged from 65 to 100 years, with mean age of 77.6 (±7.8). The mean number of diseases per participant was 2.5 (±1.3). More than half of the participants had difficulty with at least one ADL task, at least one IADL task, and walking one-quarter mile. Almost one third of the participants (n = 191) reported that they were not able to provide physical activity information due to a health reason. summarizes the results of univariate analysis of the association of GPx activity with the covariates. Age was strongly and significantly inversely associated with GPx activity with a regression coefficient of −2.89 (p < .001), indicating that for each year of increasing age, GPx activity was 2.89 μmol/min/L lower. Results of regression analyses using quadratic and spline terms to test the presence of a nonlinear association between GPx activity and age were not statistically signifi+cant. After examining scatterplots of GPx activity versus age, a potential outlier at age 100 years was identified. The direction and significance of the relationship between GPx activity and age did not change after removing this data point from the analyses; therefore, this value was included in all analyses. Significant positive associations were also observed, in separate models, between GPx activity and BMI, total cholesterol, Hgb, and weight loss (p < .05), but not with race, smoking status, coronary disease, diabetes, Se, albumin, IL-6, HDL, systolic blood pressure, or physical activity.
Baseline Characteristics for the Study Sample (N = 601)
Association of GPx with Age and Covariates Separately (Simple Linear Regression Results Between GPx and Individual Covariate)
To further investigate the finding that GPx activity decreases with age and whether this is an independent association and if it is potentially explained by the presence of disease, we analyzed the association between GPx activity and age by using sequential multiple linear regression models, adjusting for covariates (). Because we found Se to be inversely associated with age (Pearson correlation coefficient = −0.2, p < .001), Se was also added to the model. The results demonstrate that the association between GPx activity and age is significant even after adjusting for Hgb, coronary disease, diabetes, and Se level, with a β coefficient of −2.38, p = .01. The addition of coronary disease into the model adjusted for Hgb concentration minimally changed the regression coefficient, implying that the relationship between GPx activity and age is not confounded by coronary disease. However, although not statistically significant, the effect of coronary disease was of appreciable size, with a regression coefficient ranging from −17 to −21 in these models. When diabetes was added to this model, the regression coefficient was found to decrease modestly (from 2.5 to 2.2). However, an interaction term (Diabetes × Age) added to the model was not significant, indicating that diabetes is not a statistically significant effect modifier in the association between GPx activity and age.
Table 3 Association of GPx With Age, Adjusted for Covariates (Multivariate Linear Regression Models of GPx Versus Age, Sequentially Adjusting for Hemoglobin, Coronary Disease, Diabetes, and Selenium Using a Common Sample of 538 Women With Complete Data on These (more ...)
To more closely examine the relationship between GPx activity and BMI, a scatterplot of GPx activity versus BMI was performed, which suggested that below a BMI of 30 kg/m2, the association between GPx activity and BMI appears to be positively correlated; such an association becomes less apparent above this point. We added a spline term at BMI = 30 kg/m2 to determine if it represents a threshold in the association between GPx activity and BMI (results shown in ). This result shows that, for participants with BMI < 30 kg/m2, the association between GPx activity and BMI is significantly positively correlated (p = .012). However, no significant association was found between GPx activity and BMI for participants with BMI > 30 kg/m2 (p = .64). In a final multiple linear regression model with BMI and weight loss as covariates, the association between GPx activity and age was no longer significant (p = .061) (). In this model, BMI was considered with the previously modeled spline term for a nonlinear association. Because there were 38 missing values for BMI, and because all but one of these participants with missing BMI had a missing value for weight, we performed subsequent analyses to characterize these individuals. A higher proportion of those participants missing BMI information were older and had difficulty with high functioning and self-care tasks (p < .05). GPx activity for individuals with missing BMI information was 470.0 ± 157.5 μmol/min/L and was no different than for those with BMI information (p = .30).
Association of GPx With Age, Adjusted for Covariates and BMI Below/Above 30 kg/m2 (Linear Regression Model of GPx Versus Age, Including Spline Term for BMI of 30 kg/m2 Using a Common Sample of 501 Women With Complete Data on These Covariates and GPx)