Accuracy-Based Vitamin D Survey: Six Years of Quality Improvement Guided by Proficiency Testing

Proficiency testing samples were prepared from pooled off-the-clot serum obtained from several donors, some of whom received oral ergocalciferol (vitamin D₂) at least 24 hours before their blood draw (under an Institutional Review Board [IRB]–approved protocol at the University of Washington, Seattle). Pools were created by using the C37-A guideline from the Clinical Laboratory Standards Institute, which describes the preparation and validation of homogenized commutable frozen human serum pools as secondary reference materials.²⁶  Briefly, blood is drawn into a plastic blood collection bag and kept on ice. Plasma is separated within 5 minutes and transferred to a glass bottle. The plasma is allowed to clot at room temperature for 3 to 4 hours and the resulting serum is pooled and gently stirred for 18 hours at 4°C. The samples are then filtered, aliquoted, and frozen. The entire process is completed within 56 hours. Recently, in collaboration with the National Institutes of Health's Vitamin D Standardization Program, the specimens used in the ABVD Survey were demonstrated to be commutable and fit-for-purpose for proficiency testing of all clinical assays tested.⁷ 

Reference concentrations for 25(OH)D₂, 25(OH)D₃, and the C-3 epimer of 25(OH)D₃ were determined by the isotope-dilution LC-MS/MS reference method procedure performed at the Fat-soluble Nutrients Laboratory, Nutritional Biomarkers Branch at the Centers for Disease Control and Prevention (Atlanta, Georgia).²² 

Each laboratory provided results for total 25(OH)D and, if measured, for 25(OH)D₂ and 25(OH)D₃.²⁵  Method means, standard deviations, and coefficients of variation were determined internally by the CAP. Summary data from the ABVD Survey were collected from participant summary reports spanning 2011–2017. Fractional deviation was calculated in Microsoft Excel 2010 (Microsoft, Redmond, Washington) as follows: (Observed Value – Reference Value) ÷ Reference Value. The data were analyzed by standard linear regression in Microsoft Excel and represent the results from up to 320 laboratories and included 9 different automated platforms and LC-MS/MS as a general methodology (Table 1), depending on the year. Multiple linear regression in the software package R (v.3.2.3) was used to predict the contribution of 25(OH)D₂, 25(OH)D₃, and the C-3 epimer of 25(OH)D₃ to the measurement of the concentration of total 25(OH)D observed for each method. If the contribution of the epimer was determined to be statistically significant when 25(OH)D₂ and 25(OH)D₃ were included in the model, models with and without the C-3 epimer are presented. The multiple linear regression equations were then used to predict the concentrations that would be reported by each method for ABVD-08, which was included in the 2018-A mailing.

During the first 6 years of the ABVD Program, there were 43 samples distributed to participants. Data were provided to participants via participant summary reports and an example of these data is presented in Figure 1. The reports provided participants with descriptive data for each of the methods so that laboratories could review their results in comparison to other laboratories running the same method. The Accuracy-Based Surveys also provide reference target values, which have been established by a reference method procedure.

To evaluate each method, simple linear regression was performed for each methodology compared with the reference method procedure (Figure 2, A through H). The correlation coefficients (Pearson r²) of the mean total 25(OH)D measurements for each methodology (ie, the mean across all laboratories submitting data) versus the reference method were in the range of 0.9021 to 0.9962. The slopes of the linear regression varied from 1.00 to 1.17 and the y-intercepts varied from 0.50 to 4.7 ng/mL.

To test the hypothesis that immunoassays were variably able to detect 25(OH)D₂, we obtained IRB approval to administer 100,000 units of ergocalciferol to participants with vitamin D deficiency (defined as a plasma 25(OH)D concentration of less than 20 ng/mL). Samples drawn from these research subjects were included in pools sent to survey participants such that approximately 30% of the samples sent out had low to moderate concentrations of vitamin D₂ metabolites [4.4–23.5 ng/mL, making up 24%–57% of the total 25(OH)D₂]. When evaluating the relative bias for each sample versus the concentration of 25(OH)D₂ (Figure 3, A through H), it is apparent that the Abbott (Figure 3, A), Beckman (Figure 3, B), Diasorin (Figure 3, C), and Roche (Figure 3, F) methods failed to fully detect 25(OH)D₂.

We also tested the hypothesis that C-3 epimeric 25(OH)D₃ had effects on vitamin D results by comparing the relative and absolute bias with the amount of C-3 epimeric D₃ in each sample (Figure 4, A through F). Somewhat surprisingly, the Abbott (Figure 4, A and D) and Siemens (Figure 4, C and F) methods appear to be affected by the epimer. Less surprisingly, the LC-MS/MS (Figure 4, B and E) methods also appear to be significantly affected by the D₃ epimer, with the mean bias of the LC-MS/MS methods matching the mean proportion of epimer across the samples [5.9% of the total 25(OH)D, on average].

When the ABVD Program was initiated, we hypothesized that instrument manufacturers and those performing laboratory-developed procedures would use the results of the proficiency testing survey to modify their calibration when necessary so that laboratories would provide more relevant concentrations for patient care. To test this hypothesis, we examined the fractional deviation of each measurement from the reference method for each methodology during a 6-year period and 43 samples (Figure 5, A through H). The results illustrate different sample-specific biases across the methods, with a consistently positive bias for the LC-MS/MS (Figure 5, E) methods and a consistently negative bias for Roche (Figure 5, F) and Diasorin (Figure 5, C). Importantly, it was obvious that there were 2 instrument manufacturers that made changes to their assay, which greatly improved the agreement between their assays and the reference measurement procedure, namely, Siemens (Figure 5, G) and IDS iSYS (Figure 5, D).

Finally, we hypothesized that from the results of the 43 samples, we would be able to predict the results of a proficiency testing sample with a significant amount of 25(OH)D₂. To test this hypothesis, we used multivariable linear regression to determine the influence of 25(OH)D₂, 25(OH)D₃, and epimeric 25(OH)D₃ concentrations on the measured concentration by each method (Table 2). We then used the resulting β-coefficients and intercepts to predict the mean concentration that would be reported for each method for the ABVD-08 sample that was sent out in the 2018-A mailing. For at least 1 model for each method, the predicted 25(OH)D results were within 15% of the observed 25(OH)D results, suggesting that the β-coefficients for 25(OH)D₂, which are estimates of the recovery of each assay for 25(OH)D₂, are approximately correct.

Harmonization of clinical laboratory results is important to maintain consistency of care from one institution to the next and to enable proper use of concentration guidelines established when using other methods. When available, reference method procedures and reference materials can be used to standardize assays, helping laboratories and manufacturers achieve the highest level of accuracy. By providing commutable samples (ie, samples that react in exactly the same manner as patient samples) with reference method procedure–defined concentrations, accuracy-based proficiency testing programs have delivered an important resource that allows laboratories and manufacturers to ensure that their results can be confidently used in the context of guidelines established elsewhere. Even in the absence of reference method procedures, accuracy-based proficiency testing can help bolster the harmonization of measurements over time.

By leveraging the results from many different laboratories around the world for each proficiency testing sample, we demonstrated sample-specific differences between each method and the reference method procedure. While some of the differences were explained by suboptimal cross-reactivity with 25(OH)D₂ or unexpected cross-reactivity with epimeric 25(OH)D₃, there are clearly other interferences in each of the assays that likely include different lipids and 24,25-dihydroxyvitamin D.¹⁴  In the future, it might be possible to use targeted or global metabolomics to identify what other molecules specifically cause the observed sample-specific matrix effects in this and similar accuracy-based proficiency testing surveys.

The underrecovery of vitamin 25(OH)D₂ could be clinically significant in some populations in the United States, where the only pharmaceutical-grade preparation of vitamin D continues to be ergocalciferol. Unlike other proficiency testing programs, this survey was able to carefully evaluate the influence of endogenous 25(OH)D₂ by using an IRB-approved protocol that administered pharmaceutical-grade ergocalciferol to vitamin D–deficient subjects. The results confirm previous findings²⁷  and owing to the number of samples sent out over many years, allowed us to calculate the average recovery of 25(OH)D₂ in patient samples. Most of the ligand-binding assays were found to underrecover this important analyte.

Another important lesson learned during the first 6 years of the ABVD Survey is that mass spectrometry is not perfect. In fact, many laboratories using LC-MS/MS failed to report values within the acceptable range (within 25% of the target value). There are 3 likely explanations for this; first, many laboratories that use LC-MS/MS make their own calibrators in-house, which can lead to significant bias with the reference method procedure if made improperly; second, LC-MS/MS assays have variable lower limits of quantification for 25(OH)D₂, which can lead to a significant bias in samples with low total 25(OH)D and a high percentage of 25(OH)D₂, as is commonly seen in patients receiving ergocalciferol therapy; lastly, insufficient chromatographic separation of the C-3 epimer of 25(OH)D₃ leads to overrecovery of 25(OH)D₃, which is most significant in neonates and infants, but will also lead to bias when compared to the reference method procedure in these pooled adult serum samples. The data from the ABVD Survey will hopefully be used by laboratories to refine their methods to reduce bias.

The most encouraging lesson learned might be the fact that 2 manufacturers made significant changes to their immunoassays, which substantially reduced the bias observed between each assay and the reference method procedure. These efforts are to be applauded by the laboratory community and will hopefully stand out as something other manufacturers could model as we attempt to improve patient care.

In conclusion, the first 6 years of the ABVD Survey have provided an important window into the quality of patient care surrounding vitamin D status, have allowed us to characterize the cross-reactivity of each assay for endogenous 25(OH)D₂, and may have helped 2 manufacturers improve their assays. It is anticipated that once laboratories and manufacturers fully embrace the importance of harmonized or standardized clinical biomarker results, other accuracy-based proficiency testing programs will have similar success.