We selected a random sample of 200 specimens obtained from persons 60 years or older who participated in the Third National Health and Nutrition Examination Survey (NHANES III), a large cross-sectional survey of the civilian non-institutionalized population conducted in 1988–1994 [8
]. This freeze–thaw study was conducted as part of a larger project to measure markers of kidney function in stored serum samples in NHANES III [9
]. Informed consent was obtained from all participants. Protocols for the conduct and implementation of this study were approved by the Institutional Review Boards of both the National Center for Health Statistics and the Johns Hopkins Bloomberg School of Public Health.
In 2009 the specimens, which had been frozen immediately after collection, were thawed and βTP and β2M were measured. These samples were refrozen and subsequently thawed for second βTP and β2M measurements 4 to 13 months later. Samples were stored at −70 °C until the time of βTP and β2M measurement. βTP and β2M were measured at the University of Minnesota using particle-enhanced immunonephelometric assays (N Latex β-trace protein assay and N Latex β2-microglobulin assay, Siemens Diagnostics, IL). The inter-assay coefficients of variation for the βTP and β2M assays were 5.7% (mean 0.594 mg/L) and 2.7% (mean 1.757 mg/L), respectively. Assays were performed from September 2009 through July 2010 and entailed 2 reagent lots and 2 calibrators for βTP, and 3 reagent lots and 2 calibrators for β2M. To monitor reagent performance, monthly means were calculated for both βTP and β2M, and controls were assayed twice daily during all assay runs. There were no progressive shifts in assay means for βTP or β2M.
For the two biomarkers, we excluded extreme outliers (difference>3 standard deviations from the mean; equivalent to 0.3% of observations from a normal distribution). We compared measurements before and after freeze–thaw using paired t-tests, Pearson’s and Spearman’s correlation coefficients, and the intra-class correlation coefficient using one-way analysis of variance (ANOVA). We also calculated the coefficient of variation of the difference between the two measurements (CVd), using the following equation:
- CVd= standard deviation/mean
where x is the value after freeze–thaw, y is the value before freeze–thaw, and n is the total number of pairs.
Confidence intervals for the CVd were obtained by bootstrapping 1000 replications with a correction for bias. We used scatterplots and Bland–Altman plots to visually compare the βTP and β2M measurements before and after freeze–thaw. We fit linear regression models using Deming regression to correct for measurement error.