A Method-bridging Study for Serum 25-hydroxyvitamin D to Standardize Historical Radioimmunoassay Data to Liquid Chromatography-Tandem Mass Spectrometry.

BACKGROUND:Serum concentrations of 25-hydroxyvitamin D (25OHD) were measured for the National Health and Nutrition Examination Survey (NHANES) over the 1988-2006 period using a radioimmunoassay (RIA). In 2010, the Centers for Disease Control and Prevention (CDC) reissued RIA-harmonized 25OHD for NHANES 2004 and 2006, and advised users to adjust the original RIA data from 1988-1994 by using an equation. Beginning with NHANES 2007-2008, a liquid chromatography-tandem mass spectrometry (LC-MS/MS) method measured 25OHD. METHODS:A method comparison (bridging) study was designed to convert original RIA 25OHD to LC–MS/MS-equivalents. This report compares the predictive ability of a competitor regression model (Model 2) to the equations that CDC publicly released in 2015 (Model 1). The models differ by time period variable and use of transformations. RESULTS:The two models provided similar adjusted R(2) (Model 1: 88.9%, Model 2: 90.4%) and root mean square error of prediction (plus or minus 9 to 10 nanomoles per liter [nmol/L]). Applying these models to NHANES 1988–2006 RIA 25OHD, the Pearson correlation of LC–MS/MS-equivalent concentrations was 0.99; the median difference between models was 0 nmol/L (interquartile range: –2.8 to 1 nmol/L). In contrast to declining RIA-harmonized 25OHD, both models showed little change in LC–MS/MS-predicted 25OHD over the 1988–2006 period. For 2001–2006, both models predicted similar prevalences of 25OHD less than 30 nmol/L, which were lower than the prevalence estimates based on RIA-harmonized data. Mean weighted LC–MS/MS-equivalent concentrations based on either model were about 3 nmol/L lower for the 1988–1994 survey and about 3 nmol/L higher for the 2001–2006 surveys, effectively smoothing out temporal trends observed using the harmonized RIA data. CONCLUSIONS:Given minimal differences between models, final selection was based on public availability of the regression data. The bridging equations provide a way to use the previous RIA results to obtain LC–MS/MS-equivalent concentrations and evaluate temporal trends in vitamin D status.