Calibration of our RMPs was based on a previously described procedure [3] with modifications described in this section, namely, choice of calibration matrices (serum or solvent) and a narrower mass ratio range. Individual working internal standard solutions (ISTD) were prepared from concentrated internal standard stock solutions (prepared from solid material) with a targeted mass concentration of 25 ng/mL (32 nmol/L) for d₆-25(OH)D₃ and 4 ng/mL (10 nmol/L) for d₃-25(OH)D₂; the exact mass concentrations were determined at the time of gravimetric preparation. For solvent-based calibration, 2 independent ethanolic working solutions were gravimetrically prepared from SRM 2972a [10] using calibrated balances under environmentally controlled conditions. The targeted mass concentration was 40 ng/mL (100 nmol/L) for 25(OH)D₃ and 8 ng/mL (20 nmol/L) for 25(OH)D₂; the exact mass concentrations were determined at the time of gravimetric preparation. Alternatively, for serum-based calibration, we used certified reference materials, e.g., SRM 972a [11] or RM prepared in-house from pooled human serum with concentrations provided by the JCTLM-accepted RMPs at Ghent University [2]. For each analyte, 2 independent calibration curves (3 calibration levels each) were prepared, 1 from each independent ethanolic working solution (solvent-based) or from 2 different RMs (serum-based). To prepare each calibration level, we gravimetrically added individual working ISTD solutions (0.500 mL) to a pre-calculated amount (0.125–0.555 mL) of serum- or solvent-based calibration level to obtain a mass ratios of unlabeled to labeled analyte (analyte/ISTD) of approximately 0.7, 1.0, and, 1.3. Solvent-based calibrators (containing the standard and ISTD) were evaporated (in a vacuum centrifuge at 45 °C or at room temperature under N₂), reconstituted in 75% methanol/water, and injected in the instrument. For serum-based calibration, 1 mL of deionized water was added to each calibrator (containing the serum RM and ISTD). After 1 hour (room temperature, dark) we adjusted the pH to ~10 with 0.1 g/mL Na₂CO₃ (200 μL) to release the metabolites from vitamin D binding protein. After thorough mixing, the analytes of interest and their ISTDs were extracted twice with hexane, following the previously described procedure for liquid-liquid extraction from serum [3]. We evaporated the combined extracts (vacuum, 45 °C), followed by reconstitution with 75% methanol/water (0.300 mL). The serum extracts were filtered (0.45 μm filter plate) and injected for LC-MS/MS analysis. To assess the trueness of measurement for each calibration matrix, we analyzed 2 Ghent RMs (RM 001 and RM 003) using solvent- and serum-based calibrations in 3 independent campaigns. The assigned reference mass concentrations for RM 001 and RM 003 were 16.3 (40.7) and 11.8 (29.5) ng/mL (nmol/L) for 25(OH)D₃ and 1.24 (3.0) and 5.43 (13.2) ng/mL (nmol/L) for 25(OH)D₂, respectively.

For the routine LC-MS/MS procedure, we used serum-based calibrators prepared similarly to previously described calibrators in PBS-4% albumin [9]. We prepared 7 calibration levels by mixing ethanolic stock solutions of 25(OH)D₃ and 25(OH)D₂ with low analyte baseline serum. The concentration range of the calibration levels was from 2.9 to 57.7 ng/mL (7.2 to 144 nmol/L) and from 0.9 ng/mL to 24.7 ng/mL (2.2 to 60 nmol/L) for 25(OH)D₃ and 25(OH)D₂ respectively. For samples with high concentrations, we used one additional high calibrator per analyte at 117 and 47 ng/mL (292 and 114 nmol/L), respectively. The value assignments for 25(OH)D₃ and 25(OH)D₂ calibrators were confirmed by the CDC RMPs. ISTD solutions in 67% ethanol/water were used at 30 ng/mL (75 nmol/L) d₆-25(OH)D₃ and 8 ng/mL (19 nmol/L) d₃-25(OH)D₂.

All samples for 25(OH)D used in our RMPs were prepared according to a previously described procedure for liquid-liquid extraction from serum, with the modifications described in this section, namely a mass ratio of approximately 1 (analyte/ISTD) [3]. Each sample was spiked gravimetrically with pre-determined amount of ISTDs to get approximately a 1:1 mass ratio of analyte to ISTD for each analyte; wider mass ratios were used previously. All samples were initially screened for 25(OH)D₃ and 25(OH)D₂ using our routine method to obtain orientation values for each analyte. We gravimetrically added pre-calculated amounts of isotopically labeled ISTDs to serum (0.200–0.750 g, typically 0.500 g). After 1 h equilibration with the ISTD, the pH was adjusted (0.2 mL of 0.1 g/mL Na₂CO₃), the samples were mixed, and the analytes of interest were extracted with hexanes, using a previously described procedure [3]. Our RMP protocol included in each independent measurement series (run) the analysis of 1 NIST RM in singleton and 2 different Ghent RMs and all CDC candidate RMs (unknowns, typically 10) in duplicate. We carried the analysis of this set of samples typically through 3 runs, which provided the final campaign result, as a mean of all daily measurements for each sample. A diagram of the process is presented in Figure 1.

For routine measurements, we mixed 0.100 mL of serum sample with ISTD solution (0.075 mL) and 72% MeOH/water (0.1 mL) followed by liquid-liquid extraction with hexanes (1.5 mL) according to a previously published procedure [9]. We analyzed all samples used for imprecision studies with the RMP design protocol shown in Fig. 1.

The RMPs featured isocratic (Ascetis Express HPLC column) [3] or alternatively gradient chromatographic separation (Acquity HPLC column) with atmospheric pressure chemical ionization (APCI) in positive ion mode and tandem mass-spectrometric detection. The routine method was isotope dilution LC-MS/MS with isocratic separation and positive APCI ionization detection mode [9].

We used data generated with our RMPs for RMs and CDC candidate RMs to develop quality specifications for imprecision. We selected 20 specimens each for 25(OH)D₃ (no specific criteria regarding concentration) and 25(OH)D₂ (targeting samples with mass concentrations of 25(OH)D₂ above the limit of quantitation, LOQ (0.6 ng/mL, 1.5 nmol/L) and below 15 ng/mL (36 nmol/L)). The mass concentrations in the study samples ranged from 9.0 to 54 ng/mL (from 22.5 to 135 nmol/L) for 25(OH)D₃ and from 0.64 to 13.6 ng/mL (from 1.6 to 33 nmol/L) for 25(OH)D₂. Mass concentrations (X₁ and X₂) from daily duplicate preparation were averaged to obtain the daily mass concentration (X_D); to assess the daily variability we calculated the relative pair difference as a percent [abs(X1-X2)/X_D x 100] (Fig. 1). Daily mass concentrations over 3 days were used to calculate the campaign mass concentration (X_C) for each material. We calculated the campaign imprecision CV_C from the daily mass concentrations (X_D). We also selected 20 specimens from the routine LC-MS/MS method, using similar concentration requirements for 25(OH)D₂ (above the limit of detection, LOD (0.84 ng/mL, 2.0 nmol/L) to assess imprecision compared to the RMPs. The 20 specimens were not the same across methods or analytes. The mass concentrations in the study samples ranged from 6.8 to 41 ng/mL (17 to 102 nmol/L) for 25(OH)D₃ and from 0.89 to 8.7 ng/mL (2.2 to 21 nmol/L) for 25(OH)D₂. We calculated the mean and median relative pair differences from all daily relative pair differences (n=60) and campaign CV_C (n=20) for each analyte from reference and routine measurements.

We then applied the quality performance specifications for imprecision to reference data for CDC candidate RMs obtained during 3 recent campaigns, which we will call “training set.” The training set included 31 samples for 25(OH)D₃ (4.8–42 ng/mL, 12–105 nmol/L); only 9 of these samples had reportable concentrations for 25(OH)D₂ (0.62–14.2 ng/mL, 1.5–34 nmol/L).

Typically we analyze at least 1 NIST RM material, e.g., SRM 972a (prepared in singleton) [11], and at least 2 Ghent RMs (prepared in duplicate) with each reference run. The detailed data evaluation procedure is outlined in Table 1. Typically if one or multiple CDC candidate RM violate the imprecision specification, those reference measurement results are rejected and the samples are re-analyzed in another measurement series (repeat CDC candidate RM). If one Ghent RM or the NIST SRM violate the accuracy or imprecision criteria, the reference run is rejected and all samples are re-analyzed in another measurement series (repeat run). We used data from reference measurements of 2 materials, namely, SRM 972a level 2 and RM 003 from 7 recent campaigns, to demonstrate the accuracy of our RMPs. We selected these 2 RMs because both of them were analyzed in these campaigns and because the target concentrations were assigned by different reference methods, NIST and Ghent University.

We opted to use zeta-scores (ζ) for quantitative assessment of accuracy. This standardized parameter shows how well a measurement matches the assigned target concentration relative to the combined uncertainty of the measurement and the target concentration [12, 13]. A zeta-score of 0 indicates a perfect match. In proficiency testing programs, a zeta-score of ±2 indicates acceptable participant result. The zeta-score is defined as follows: (equation 1),ζ=(x−y)/u(x)2+u(y)2

where x is the CDC RMP result and y is the NIST or Ghent RMP result; u(x) is the standard uncertainty of the CDC RMP [3] and u(y) is the reported standard uncertainty for the Ghent and NIST RMs. For SRM 972a level 2, we used the target concentrations (y) listed in the Certificate of Analysis (COA) and the reported standard uncertainties u(y) of 1.1% and 3.8% for 25(OH)D₃ and 25(OH)D₂, respectively [11]. The standard uncertainties were derived by dividing the reported expanded uncertainty [U=2 u(x)] from the certificate by the coverage factor k=2. For RM 003, we used the target concentrations (y) of 11.8 ng/mL (29.5 nmol/L) for 25(OH)D₃ and 5.43 ng/mL (13.2 nmol/L) for 25(OH)D₂ and the reported standard uncertainties u(y) of 1.5% and 1.8% for 25(OH)D₃ and 25(OH)D_2, respectively, as provided in the report of analysis. The estimated standard uncertainties u(x) for CDC RMPs were 1.6% and 1.8% for 25(OH)D₃ and 25(OH)D_2, respectively [3].

To evaluate the daily accuracy, we calculated the zeta-score_D as described above, using the certified result and uncertainties from the NIST COA for the SRM material [11], and the Ghent assigned target value and reported uncertainties for the Gent RM material, listed above. In both cases, we used our daily reference measurement (x: X_D) and the uncertainty u(x) for our RMP. To evaluate the campaign accuracy, we calculated the zeta-score_C from the campaign mass concentration (x: X_C). Similarly, we determined the daily and campaign systematic deviation from the target concentration (bias, %) for each material. We developed a macro in SAS version 9.3 to evaluate daily and campaign reference measurements and to automatically apply the numerical quality performance specifications for accuracy.

In our laboratory, we have well-established methods for analysis of serum 25(OH)D metabolites, namely, LC-MS/MS routine and reference measurement procedures [9, 3]. For the RMPs, we studied 2 types of calibrations, solvent- and serum-based. For solvent-based calibration, ethanolic working solutions were prepared by gravimetric dilution of SRM 2972a with absolute ethanol. This is the most direct way to establish traceability to the highest order of commercially available accuracy-based NIST RM and SI units. All working solutions were stable when stored in tightly closed glass containers at −20 °C ± 2 and used within 1 month of preparation. To prepare serum-based calibrators, we used SRM 972a level 1 or level 4 for 25(OH)D₃ and SRM 972a level 3 for 25(OH)D₂. These 3 levels were chosen because they had the highest certified concentrations for each respective analyte, and therefore 1 vial each was sufficient to prepare 1 independent calibration curve. The second calibration curve for each analyte was prepared from Ghent RMs. We used both curves for quantitation. To compare the calibration matrices (solvent vs. serum) with regards to method trueness, 2 RMs (RM 001 and RM 003) were processed independently with each calibration matrix in 3 independent campaigns. We evaluated the calculated mass concentrations from each curve against the target concentrations for accuracy, expressed quantitatively with the zeta-score. Using 2 RMs over 3 campaigns, the mean daily zeta-score_D (SD) from serum- vs. solvent-based calibrations was −0.93 (0.66) vs. 0.01 (0.94) for 25(OH)D₃ and 0.43 (1.15) vs. 0.11 (1.25) for 25(OH)D₂, respectively. The overall daily bias from the serum-based curve was −2% and 1% for 25(OH)D₃ and 25(OH)D₂, respectively. The overall daily bias from the solvent-based curve was −0.1% and 0.3% for 25(OH)D₃ and 25(OH)D₂, respectively. Accuracy from solvent-based calibration showed nearly perfect agreement, as expressed by a mean zeta-score of approximately 0 and a mean bias ≤0.3%. Our preferred RMP calibration approach is therefore solvent-based.

For reference measurements, all samples were prepared by accurate weighing of serum and pre-calculated amounts of isotopically labeled ISTD solutions to obtain approximately 1:1 mass ratios of analyte to ISTD. In our earlier studies, the amount of working ISTD was not closely matched to the amount of measurand, resulting in wide mass ratios, 0.25–2.50. The impact of the use of the different mass ratios (wide vs narrow) on overall accuracy of all valid results has not been shown to be significantly different in our laboratory (data not shown). We noted improvement with imprecision violations using the narrow mass ratio from 4% and 17% for 25(OH)D₃ and 25(OH)D₂ down to 0% and 10%, respectively. Daily zeta-score violations were marginally improved for 25(OH)D₃ (3% to 2%) and greatly improved for 25(OH)D₂ (18% to 7%). Considering the low throughput (10 samples/run) and laborious sample preparation for RMPs, switching to a mass ratio of analyte to ISTD of approximately 1:1 was beneficial.

During the development and validation of the RMPs [3] we used the pre-defined proposed quality specifications for accuracy, i.e., CV<5% and bias <1.7% [7]. Our RMPs showed similar performance characteristics to those of other established RMPs [1, 2]. We used our RMPs to analyze the same sample over multiple days (campaign), which involves complex data review with independent daily and campaign data assessment from multiple RM. Because the accuracy specifications for total 25(OH)D are not applicable to the individual metabolites occurring at very different concentrations, we developed our in-house numerical quality performance imprecision and accuracy goals, based on years of experience with the method. We had to consider the complex multiday data analysis protocol of our RMPs (Fig. 1). We evaluated all daily data for imprecision and accuracy. If the run passed our criteria, the exact same analyses were repeated on 2 more days and the same daily data review was conducted. Data from all 3 valid runs comprise the campaign data; the combined data were evaluated for overall campaign imprecision and accuracy. For a summary of our quality performance specifications procedure, acceptability criteria, and actions taken upon data evaluation, see Table 1.