Correcting for nonlinear measurement errors in the dependent variable in the general linear model

Linear models are considered in which measurement error is present in the dependent variable. Observed values are related to true values via nonlinear regression models with the parameters in the measurement error models being estimated with the use of independent, external data, collected using standards. Pseudo-maximum likelihood estimators and their asymptotic properties are developed under normality assumptions and the common approach of simply analyzing imputed values obtained from the nestimated calibration curves is assessed. A small simulation evaluates the procedures. An example is presented in which urinary neopterin (measured via radioimmunoassay) is nbeing compared between two groups of individuals.     

Serum testosterone measurement in men: Evaluation of modern immunoassay technologies

Accurate measurement of serum testosterone (T) is essential for proper diagnosis of androgen deficiency. There are now several modern assay technologies, including automated ones, for measurement of T. In this study, we compared analytical performance of five modern immunoassay technologies commonly used for measurement of total T: Vitros ECi (Ortho-Clinical Diagnostics; normal range (n.r.) 4.6–34 nmol/L); Architect (Abbott Laboratories; n.r. 9.7–34 nmol/L); Access (Beckman Coulter; n.r. 5.3–23 nmol/L); Delfia (Perkin-Elmer; n.r. 9.3–34 nmol/L); and manual EIA DRG kits (n.r. 8.3–42 nmol/L), with the classical RIA (³H–T), after extraction (n.r. 11–33 nmol/L), as a reference method. Total T was measured using all above-mentioned methods in serum samples from 100 male patients, aged 16–65 years. Mean T concentrations in these 100 serum samples assayed by all non-isotopic methods were statistically significantly higher than those obtained by RIA. Delfia showed the highest T levels (19.3 nmol/L versus 12.1 nmol/L by RIA) with a positive bias 60–100%. Almost similar results were obtained using Architect, with a positive bias 40–70%. The closest correlation in results was found between Vitros ECi and RIA (12.7 nmol/L versus 12.1 nmol/L). In the studied samples, the median of differences ranged from minimal (?0.4 nmol/L for Vitros ECi) to maximal (?7.25 nmol/L for Delfia). For all non-isotopic methods, with the exception of Vitros ECi, differences in subjects with low T level (<10 nmol/L) were statistically significantly larger than in the subjects with high T (T > 10 nmol/L). All other methods showed different degrees of dissimilarities with the RIA, especially in the range of low testosterone concentrations, which is of importance in the clinical assessment of women and pubertal boys.     

Free testosterone by direct and calculated measurement versus equilibrium dialysis in a clinical population

Abstract

Introduction: The value of clinically available free testosterone (FT) assays remains controversial. Here, we evaluate the agreement between the radioimmunoassay (RIA) and calculated FT (cFT) versus equilibrium dialysis (EqD), considered the gold standard.

Methods: Fifty-six consecutive men (aged 26–77) had blood samples assessed for FT, including men with treated and untreated testosterone deficiency (TD) and men without TD. Samples were split and tested by the two methodologies at a Quest Diagnostics national reference laboratory. cFT was calculated by the Vermeulen method.

Results: A robust correlation was noted for RIA and EqD (r?=?0.966) and for cFT and EqD (r?=?0.986). Strong correlations were observed for men receiving testosterone therapy and for men in the lowest and highest quartiles for total and FT. The correlation of total testosterone with FT was similar for cFT (r?=?0.843), RIA (r?=?0.806), and EqD (r?=?0.809). Sex-hormone binding globulin (SHBG) was not correlated with any measure of FT. Bland–Altman analysis demonstrated similar bias for both cFT and RIA, although cFT consistently overestimated FT. Numerical values for RIA were approximately one seventh of EqD values.

Conclusions: These results support the clinical use of both RIA and cFT as measures of FT. Due to numerical differences, each test requires its own set of reference values.