Vitamin D plays an important role in bone metabolism by regulating calcium and phosphorus homeostasis in organisms; therefore, vitamin D deficiency is a risk factor for osteoporosis and bone fracture [1, 2]. In addition to bone metabolism, vitamin D is involved in the functional regulation of various systems, such as immune, cardiovascular, reproductive, and endocrine systems. Thus, multiple studies are being actively conducted [2-4].

Vitamin D exists in two primary forms in the human body: vitamin D₃ (cholecalciferol) and vitamin D₂ (ergocalciferol), which are synthesized via exposure to ultraviolet light and from plants and absorbed as dietary intake, respectively. Vitamin D in the human body is primarily hydroxylated in the liver and exists as 25-OH vitamin D (25-OH-D) in the systemic circulation. A portion of the 25-OH-D is hydroxylated in the kidney and then transformed to 1,25-(OH)₂ vitamin D (1,25-[OH]₂-D), the biologically active form. Vitamin D status in the human body is generally monitored using 25-OH-D, which has a long half-life and is stored in high concentrations [3, 4].

Recently, assays using various test principles, including mass spectrometry (MS), which is currently the reference measurement procedure (RMP) for measuring 25-OH-D, or commercial immunoassays, are used to clinically test 25-OH-D [5, 6]. The National Health Insurance in Korea provides medical coverage for one test to initially diagnose vitamin D deficiency, one test to determine the therapeutic response 3–6 months after medication, and follow-up tests twice annually during the continuous medication period. However, only immunoassays excluding MS are covered for the initial diagnostic test. Significant discrepancies in 25-OH-D measurement results between testing methods and laboratories have previously been reported [7-10]. Therefore, the National Health Insurance in Korea that does not cover MS having higher accuracy at the initial diagnosis, making the early diagnosis of vitamin D deficiency difficult.

Because the difference in test results between methods using the same sample substantially affects medical decision-making and diagnosing and treating vitamin D deficiency [10], external quality assessment programs for 25-OH-D testing, such as the Accuracy-based 25OHD College of American Pathologists (CAP) survey, Vitamin D External Quality Assessment Scheme (DEQAS), and Vitamin D Metabolites Quality Assurance Program, are performed internationally. The Vitamin D Standardization Certification Program (VDSCP) run by the Centers for Disease Control and Prevention (CDC) certifies the 25-OH-D test as standardized when meeting the analytical performance criteria of bias ≤5% compared to the CDC RMP and ≤10% imprecision using human serums [5, 6]. Although most commercial assays have been certified via VDSCP, the results of measurements performed by clinical laboratories and academic researchers can be inaccurate because the certification criterion of bias is applied only for the mean percent bias of all samples, not for each sample. The individual sample pass rates ranged from 36 to 45% for immunoassays included in the present study [5]. Moreover, there are differences between the well-controlled environment of the manufacturers when conducting external quality assessments and the routine test environment of the clinical laboratories.

The present study evaluated four commercial 25-OH-D immunoassays. Regarding quantitative correlation and qualitative decisions, the commercial immunoassay results were compared with those of the laboratory-developed test (LDT) MS, which was assumed as a reference method, using residual serum samples.

1. Preparation of samples for method comparison

This study was approved by the institutional review board (IRB No. KUMC 2021-10-032), and the requirement for informed consent was waived. The remnant serum samples with a residual volume of ≥1.6 mL were collected as study samples from July to October 2021. According to the recommendation of the Clinical & Laboratory Standards Institute (CLSI) guideline EP09-A3: Measurement Procedure Comparison and Bias Estimation Using Patient Samples [11], 150 samples, namely 50 samples with a 25-OH-D concentration of 5–15 ng/mL, 50 samples with a concentration of 15–25 ng/mL, and 50 samples with a concentration of >25 ng/mL, were obtained to cover the analytical measurement interval. All collected samples were aliquoted and stored frozen at -80°C until measurement. On the measurement day, the frozen test samples were delivered to each laboratory within 3 hours. Measurements were taken immediately after thawing 50 samples daily for 3 days as scheduled.

2. Standard materials

The National Institute of Standards & Technology (NIST; MD, USA) standard reference materials (SRMs) 972a were measured on two MS systems to evaluate the accuracy of MS in assessing the 25-OH-D value in the study samples [12]. The SRMs comprised four levels of lyophilized serum which had diverse assigned 25-OH-D values: “normal” human pooled serum (level 1), human serum diluted with horse serum to achieve a lower 25-OH-D value (level 2), and spiked human serum to enrich 25-OH-D₂ (level 3) or 3-epi-25-OH-D₃ (level 4) [13].

3. Development and validation of reference measurement procedure

The two LDT MS tests were performed in 1) Seoul National University Bundang Hospital according to the protocol previously published [14], except for changing the instruments to a Nexera X2 Series HPLC/UHPLC (Shimadzu Corporation, Kyoto, Japan) and SCIEX Triple Quad 6500+ LC-MS/MS System (SCIEX, Framingham, MA, USA), and in 2) Green Cross Laboratories using protein precipitation, solvent extraction, and derivatization steps using reagents from the PerkinElmer (Turku, Finland) MSMS Vitamin D kit, on the ACQUITY UPLC I-Class System (Waters Corporation, Milford, MA, USA) equipped with a Waters ACQUITY UPLC BEH C18 column (2.1×50 mm, 1.7 μm; Waters Corporation), and Xevo TQD (Waters Corporation) MS/MS system. Both systems have been certified by CAP and DEQAS, and the system in Green Cross Laboratories obtained VDSCP certification. To validate the performance of MS tests as RMP, the average values for NIST SRMs 972a obtain ed by the two systems have been compared with the assigned value. The standard materials were reconstructed on the measurement day at Seoul National University Bundang Hospital and were immediately measured after being delivered to each laboratory. Subsequently, the average values for study samples on MS systems were used as references to evaluate the immunoassays.

4. Evaluation of four commercial 25-OH-D immunoassays

Four commercial immunoassays, the Roche Elecsys Vitamin D Total II for Cobas e801 analyzer (Roche; Roche Diagnostics GmbH, Mannheim, Germany), Siemens Atellica IM Vitamin D Total assay for Atellica IM analyzer (Siemens; Siemens Healthineers, Forchheim, Germany), Abbott Alinity 25-OH Vitamin D for Alinity i system (Abbott; Abbott Laboratories, Abbott Park, IL, USA), and Beckman Coulter Access 25-OH Vitamin D Total for Beckman Coulter Access immunoassay analyzer (Beckman; Beckman Coulter, Brea, CA, USA), were evaluated for measuring 25-OH-D in the study samples. All immunoassays are traceable to a Joint Committee for Traceability in Laboratory Medicine-approved isotope dilution MS RMP, which is traceable to the NIST SRM 2972. The Beckman test was performed at Soonchunhyang University Bucheon Hospital, while the other methods were performed at Konkuk University Medical Center. The 150 serum samples stored frozen at -80°C were delivered to each laboratory in November 2021. Their 25-OH-D concentrations were measured over 3 days using a preplanned schedule, with 50 samples daily. The measured results were reported and collected as ng/mL in all methods; 1 ng/mL corresponded to 2.5 nmol/L [6].

5. Statistical analysis

The immunoassay results were compared with the mean values from two MS tests, the MS1 and MS2 methods performed at Seoul National University Bundang Hospital and Green Cross Laboratories, respectively, using the Passing–Bablok regression analysis. The slope and intercept of regression with their 95% confidence interval (CI) and Pearson correlation coefficient (r) were determined. The mean percent bias (%bias) was calculated using a Bland–Altman plot. The mean %bias from each test platform was compared to the 5.0% VDSCP-claimed acceptable bias limit for 25-OH-D measurement; the acceptability was then evaluated. Three 25-OH-D concentrations, 10, 20, and 30 ng/mL, are currently considered the medical decision limits (MDLs), dividing the vitamin D status into four categories: severe deficiency (25-OH-D <10 ng/mL), which requires correction based on the criteria of all organizations; mild deficiency (10≤ 25-OH-D <20 ng/mL), for which most organizations recommend correction; insufficiency (20≤ 25-OH-D <30 ng/mL), whose supposed necessity of supplementation is controversial; and sufficient level (30 ng/mL≤ 25-OH-D), in which supplementation is rarely recommended [10, 15]. The regression estimate with 95% CI at each MDL on each test system was then calculated. In addition, using the three MDLs as cut-off values of the qualitative decision of vitamin D status for the study samples, the four categories were determined on each system. The sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) of decisions for vitamin D status that required correction were calculated by regarding the decision of MS at the highest MDL of 30.0 ng/mL as the standard. The prevalence of insufficient vitamin D status, 85.3% in the Korean population, published by the Korea Association of Health Promotion [16], was adopted to determine PPVs and NPVs. The statistical analyses were performed using SPSS statistic version 26.0 (IBM Corp., Armonk, NY, USA) and MedCalc version 14.8.1 (MedCalc Software, Maria kerke, Belgium).

1. Validation of reference measurement procedures

1) Traceability of two MS systems

The NIST SRM 972a was measured on two MS systems; the mean values are shown in Supplementary Fig. 1. The assigned values of 25-OH-D in each material concentration, 31.15 ng/mL, 20.19 ng/mL, 34.17 ng/mL, and 55.95 ng/mL, and their acceptable lower and upper limits by applying the 5.0% allowable bias, are presented. The mean values for all NIST materials measured in the MS systems were within the acceptable range: 30.61 ng/mL (32.01, 29.21 ng/mL on each system) for level 1 with -1.73 %bias, 19.81 ng/mL (19.93, 19.69 ng/mL) for level 2 with -1.88 %bias, 33.15 ng/mL (31.30, 34.99 ng/mL) for level 3 with -3.00 %bias, and 53.96 ng/mL (54.90, 53.01 ng/mL) for level 4 with -3.57 %bias, considering the known interference of 3-epi-25-OH-D₃ [17].

2) Equivalence of two MS systems

In the Passing–Bablok regression for two MS systems, the slope and 95% CI were 1.010 and 0.980–1.045, the intercept and 95% CI were 0.049 and -0.691–0.689, and the Pearson correlation coefficient (r) was 0.982, which was >0.975. The 25-OH-D concentrations in the collected residual samples ranged from 5.86 to 67.75 ng/mL in the MS tests. The mean %bias for the test samples between the MS systems was 0.9% (Supplementary Fig. 2). The results on the two MS systems were considered equivalent because the mean %bias was not only within 5.0% the VDSCP criterion, but also within 3.4% the optimum specification suggested by European Federation of Clinical Chemistry and Laboratory Medicine database [18]. Subsequently, the mean values on the MS systems were used as RMP values for evaluating immunoassays.

2. Evaluation of four commercial 25-OH-D immunoassays

The second MS test and Beckman immunoassay were not performed for sample N’3 due to an insufficient sample volume. Sample N’11, which had a high I-index and showed measurement %differences >100% in the Siemens and Beckman tests, was considered an outlier via Tukey’s method and excluded from the Bland–Altman analysis.

1) Correlation of the four immunoassay values with the RMP values

Regression analysis was conducted using the RMP values as a comparative procedure, and those from the other assays as candidate procedures; the Pearson correlation coefficient (r) in the immunoassays ranged from 0.890 to 0.973, which was highly correlated with the MS tests; however, it did not meet 0.975, the current recommendation of the CLSI EP09-A3 guideline about the use of ordinary linear regression. Thus, the Pearson correlation coefficient (r) and slope and intercept with their 95% CI were calculated using Passing–Bablok regression analysis. As shown in Table 1, in four immunoassays, the slope and intercept range from 0.855 to 1.043 and -2.107 to 2.635, respectively. No immunoassay was statistically equivalent to the MS systems, as their 95% CI of the slope and intercept deviated from 1 and 0, respectively.

Table 1 . The parameters of the Passing–Bablok regression analysis for each immunoassay compared with RMP values

	Roche	Siemens	Abbott	Beckman
r (r2)	0.973 (0.947)	0.911 (0.830)	0.957 (0.916)	0.890 (0.792)
Slope (95% CI)	0.924 (0.884–0.969)	0.855 (0.776–0.932)	1.043 (0.984–1.098)	0.914 (0.847–0.979)
Intercept (95% CI)	-2.107 (-3.040–-1.344)	0.621 (-0.495–2.086)	-2.049 (-3.304–-0.885)	2.635 (1.387–3.968)

Abbreviations: CI, confidence interval; Roche, Roche Elecsys Vitamin D Total; Siemens, Siemens Atellica IM Vitamin D Total assay; Abbott, Abbott Alinity 25-OH Vitamin D; Beckman Coulter, Beckman Coulter Access 25-OH Vitamin D Total.

As presented in Fig. 1, the mean %bias and its 95% CI of each test system were -17.6% (-36.6–1.5%), -7.8% (-38.3–22.8%), -9.6% (-41.8–22.7%) in Siemens, and 6.7% (-29.6–43.1%) in the Roche, Abbott, Siemens, and Beckman tests, respectively. The mean %bias of each immunoassay was out of the 5.0% VDSCP criterion, showing significant differences compared to MS.

Figure 1. The distribution of %bias for the study samples determined using mass spectrometry and immunoassays. (A) Roche, (B) Siemens, (C) Abbott, and (D) Beckman. The solid blue and dashed brown lines represent the mean %bias of each immunoassay and ±5.0% and the VDSCP-claimed mean %bias acceptable limit, respectively, whereas the dotted orange line represents zero bias.

2) Differences between the four immunoassays values at the MDL with the RMP value

The measurement difference at the concentration near the MDL can affect the qualitative decision of vitamin D status. The negative and positive biases can increase the false positive and negative decisions for vitamin D deficiency, respectively. The regression estimates, their 95% CI, and %difference on each platform for three levels of MDL samples with 25-OH-D concentrations of 10, 20, and 30 ng/mL in the reference method were analyzed; the results are shown in Table 2. The regression estimates and 95% CI for the immunoassays were commonly out of the VDSCP criteria, in contrast to those between the MS systems.

Table 2 . The regression estimates and 95% CI on each test system for three levels of samples with concentrations at the MDL

Method	Expected value (95% CI) in each MDL (ng/mL)
	10	20	30
Roche	7.13 (6.59–7.59)	16.37 (15.95–16.78)	25.61 (25.01–26.28)
Siemens	9.17 (8.62–10.02)	17.73 (17.28–18.66)	26.28 (25.15–27.82)
Abbott	8.38 (7.62–9.04)	18.80 (18.12–19.23)	29.23 (28.44–29.92)
Beckman Coulter	11.78 (10.99–12.60)	20.92 (20.24–21.47)	30.06 (28.93–31.02)
MS2*	10.15 (9.76–10.55)	20.24 (19.97–20.53)	30.34 (29.90–30.84)

*Regression estimates for MS2 were determined based on a correlation with MS1, whereas those for other immunoassays were calculated via a correlation with the RMP value.

Abbreviations: MS, mass spectrometry; Roche, Roche Elecsys Vitamin D Total; Siemens, Siemens Atellica IM Vitamin D Total assay; Abbott, Abbott Alinity 25-OH Vitamin D; Beckman Coulter, Beckman Coulter Access 25-OH Vitamin D Total.

3) Qualitative decision of four immunoassays values

Compared to the decision based on the MS systems, the agreement of the vitamin D status that correction would be required (<30 ng/mL) was determined by each immunoassay (Table 3). The diagnostic accuracy for the decision ranged from 90.0 to 94.0%. The immunoassays, except the Roche test, yielded a low NPV ranging from 62.4 to 79.0%, indicating that vitamin D deficiency diagnosis can be easily missed due to ‘false negative’ results. Alternatively, the Roche test showed a trend of values consistently lower than that of the RMP, which can overestimate vitamin D deficiency via ‘false positive’ results. The comparison of quantitative values and qualitative decisions between the MS system and other immunoassays are visualized as scatter plots in Supplementary Fig. 3 and Fig. 2. When the vitamin D statuses were divided into the four groups using the three MDL levels as cut-off values, the decisions in immunoassays showed a discordance with those of the MS system, ranging from 21.3% (32/150) to 36.0% (54/150) on each test. The Cohen’s weighted kappa coefficient for the immunoassays, determined via an inter-rater agreement analysis, ranged from 0.762 to 0.852 (Supplementary Table 1).

Table 3 . Agreement of vitamin D status between mass spectrometry and other immunoassays at the 30 ng/mL cut-off level

		MS		Total	Parameter (%)	Range
		Not sufficient	Normal
Roche	Not sufficient	108	19	127	Sn	100.0 (96.6–100.0)
	Normal	0	23	23	Sp	54.8 (38.7–70.2)
	Total	108	42	150	PPV	92.8 (90.2–94.7)
					NPV	100.0 (100.0–100.0)
					Accuracy	93.4 (88.1–96.8)
Siemens	Not sufficient	104	18	122	Sn	96.3 (90.8–99.0)
	Normal	4	24	28	Sp	57.1 (41.0–72.3)
	Total	108	42	150	PPV	92.9 (90.2–94.9)
					NPV	72.7 (49.5–87.8)
					Accuracy	90.5 (84.7–94.7)
Abbott	Not sufficient	104	8	112	Sn	96.3 (90.8–99.0)
	Normal	4	34	38	Sp	81.0 (65.9–91.4)
	Total	108	42	150	PPV	96.7 (94.0–98.2)
					NPV	79.0 (58.7–90.9)
					Accuracy	94.0 (89.0–97.3)
Beckman Coulter	Not sufficient	98	8	106	Sn	91.6 (84.6–96.1)
	Normal	9	34	43	Sp	81.0 (65.9–91.4)
	Total	107	42	149	PPV	96.5 (93.7–98.1)
					NPV	62.4 (46.6–75.9)
					Accuracy	90.0 (84.1–94.3)

The prevalence of each vitamin D status in the Korean population, wherein 85.3% of the participants had vitamin D levels lower than 30 ng/mL, was adopted from a previous Korean study [16] to calculate positive and negative predictive values for each test platform.

Abbreviations: MS, mass spectrometry; Roche, Roche Elecsys Vitamin D Total; Siemens, Siemens Atellica IM Vitamin D Total assay; Abbott, Abbott Alinity 25-OH Vitamin D; Beckman Coulter, Beckman Coulter Access 25-OH Vitamin D Total; Sn, sensitivity; Sp, specificity; PPV, positive predictive value; NPV, negative predictive value.

Figure 2. The quantitative values of 25-OH-D and decision in study samples determined using mass spectrometry and immunoassays. (A) Roche, (B) Siemens, (C) Abbott, and (D) Beckman, presented as scatter plots. The dotted lines are medical decision limits, blue dots represent concordant results, and red dots represent discordant results.

Because vitamin D in the body participates in the functional regulation of various organs, standardizing 25-OH-D measurement tests to diagnose vitamin D deficiency or confirm the therapeutic response is essential for ensuring clinical utility. However, many studies have continuously reported significant discrepancies between test methods or laboratories. In addition to the effects of this measurement error on the clinical use of tests, it can be an obstacle to establishing an international standard for determining vitamin D status in humans [3, 10]. This intrinsic limitation of the 25-OH-D measurement test has been presented continuously since its implementation. Although several international quality assessment programs for a 25-OH-D test have been developed and introduced [5, 13], many clinical laboratories and researchers still perform tests regardless of standardization; thus, there could be confusion in the interpretation of the test results.

Although MS is the RMP for the 25-OH-D measurement, multiple laboratories use immunoassays as routine tests for easier maintenance and simple handing of the instrument. Additionally, for the initial diagnosis of vitamin D deficiency, the National Health Insurance in Korea provides coverage only for immunoassays and not for MS. Although immunoassays have practical advantages and clinical utility, various cross-reactivities with substances such as 1,25-(OH)₂-D, 24,25-(OH)₂-D, and 3-epi-25-OH-D have also been reported [17] and have different acceptable limits for interference materials, which results in interference within the allowable range. Therefore, changing the test system during the follow-up in diagnosed and treated patients can confuse medical decision-making. Thus, comparing 25-OH-D assays that encompass different test principles is necessary.

In the present study, four immunoassays had statistically significant differences in quantitative results compared to the MS results. The values were not equivalent in the regression analysis, and the mean %difference in each immunoassay at the three MDL levels exceeded the VDSCP criterion of 5.0%. In the concordance of qualitative decisions for vitamin D status, it was confirmed that discordances, such as false positives or negatives, can occur depending on the imprecision or bias of each assay, and the decision errors were considered to affect the diagnosis and treatment of vitamin D deficiency.

Although the immunoassays included in the present study, except for the Beckman test [5], achieved certification via VDSCP, the mean %differences exceeding the certification criteria were observed. It is considered to be caused by the difference in metrological traceability between the central laboratories of manufacturers using master calibrators with small uncertainty and the manufacturer’s standing procedure and clinical laboratories using end-user calibrators with greater uncertainty and routine laboratory procedures. This can represent the actual test environment of clinical laboratories that are not strictly controlled, and the bias of specific test methods in diverse papers is reported differently depending on the study [19]. Thus, the error must be considered even when using the tests certified by the standardization program.

This study had certain limitations. First, the values used for analysis were obtained from a single institution by measuring once. The between-laboratory variation was not included, whereas the effect of the possible random error was not entirely excluded. Second, the precision evaluation of each test, which could have affected the comparative analysis, was not performed for this study but was replaced by method evaluation data for routine tests. Third, the prevalence of the four vitamin D statuses in the study samples determined via MS systems (4.7% [7/150] of severe deficiency; 34.7% [52/150] of mild deficiency; 32.7% [49/150] of insufficiency; 28.0% [42/150] of sufficient level) was different to previous Korean rese-arch [16], because test samples were selected to cover a wide analytical measurement range to meet the CLSI EP09-A3 guideline recommendation. Therefore, it should be considered that the study design did not fully re˝ect the practical test performance in a clinical setting.

Despite these limitations, this study is expected to provide basic knowledge for interpreting and utilizing 25-OH-D tests by evaluating the methods commonly used in clinical laboratories, including MS and immunoassays.

None declared.

This study was supported and funded by the Korean Society of Clinical Chemistry (KSCC) in 2021.