“Linearity assessment methods for sex steroid hormones and carrier proteins among men in the National Health and Nutrition Examination Survey (NHANES III)”

• Correlation shows sex hormone and carrier protein relationships are not linear.
• A categorical carrier protein variable is advised when linearity is questionable.
• Differing sex hormone adjustments may explain variation in health disparity studies.

IntroductionIt has been hypothesized that racial disparities among several diseases are explained by differences in testosterone (T), 17-β estradiol (E), sex hormone binding globulin (SHBG) and albumin (A) levels, yet epidemiologic results have been mixed. Statistical advice regarding appropriate adjustment methods for carrier proteins of sex steroid hormones in the literature is scant. Therefore, we investigated different adjustment methods for carrier proteins.MethodsData for 1496 men, >17 years from the Third National Health and Nutrition Examination Survey (NHANES III) 1988–91 were used to analyze linearity between sex hormones and carrier proteins by examining correlation, plots, and regression models. The statistical importance of age, body mass index (BMI), and race/ethnicity were examined for changes in results by the adjustment method.ResultsT was weakly correlated with SHBG and A (r-squared, 0.25, 0.13, respectively) and E was weakly negatively correlated with A (p < 0.0001), but not SHBG (p < 0.1799). Based on the model residual plots and r-squared, the categorical model performed better than linear models. Regression coefficients for age, BMI, and race/ethnicity groups using quotient (e.g., T/A and E/A) models differed from continuous and categorical models.ConclusionChoosing an appropriate adjustment for carrier proteins is important to prevent bias in analyses and inconsistency in findings across studies. Linearity between variables should not be assumed when adjusting models, and should be conducted and reported. An independent categorical carrier protein variable is recommended in analysis exploring factors predicting sex hormone levels, although statistical testing should always be employed.

Figure optionsDownload as PowerPoint slide