Metabolism Clinical and Experimental 89 (2018) 1–7

Contents lists available at ScienceDirect

Metabolism Clinical and Experimental

j ourna l homepage: www.metabo l i smjourna l .com

Commentary

Limitations of platform assays to measure serum 25OHD level impact onguidelines and practice decision making☆,☆☆

Maya Rahme a, Laila Al-Shaar a, Ravinder Singh f, Rafic Baddoura e, Georges Halaby d, Asma Arabi a,Robert H. Habib b, Rose Daher c, Darina Bassil a, Karim El-Ferkh a, Maha Hoteit a, Ghada El-Hajj Fuleihan a,⁎a Department of Internal Medicine, Division of Endocrinology, Calcium Metabolism and Osteoporosis Program, WHO Collaborating Center for Metabolic Bone Disorders, American University ofBeirut Medical Center, Beirut, Lebanonb Department of Internal Medicine, American University of Beirut Medical Center, Beirut, Lebanonc Department of Pathology & Laboratory Medicine, American University of Beirut Medical Center, Beirut, Lebanond Department of Endocrinology, Hotel Dieu de France, St Joseph University, Beirut, Lebanone Department of Rheumatology, Hotel Dieu de France, St Joseph University, Beirut, Lebanonf Division of Clinical Biochemistry and Immunology, Mayo Clinic Laboratories, Mayo Clinic, Rochester, MN, USA

☆ All authors state that they have no conflict of interes☆☆ Trial Registration ClinicalTrials.gov Identifier: NCT013

⁎ Corresponding author at: Calcium Metabolism andCollaborating Center for Metabolic Bone Disorders, AMedical Center, P.O. Box: 113-6044/C8, Beirut, Lebanon.

E-mail address: gf01@aub.edu.lb (G. El-Hajj Fuleihan)

https://doi.org/10.1016/j.metabol.2018.09.0030026-0495/© 2018 Elsevier Inc. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 12 July 2018Accepted 13 September 2018

Context: Liquid Chromatography Mass Spectroscopy (LC-MS/MS) is the preferred method to measure 25hydroxyvitamin D (25OHD) levels, but laboratories are increasingly adopting automated platform assays.Objective: We assessed the performance of commonly used automated immunoassays, with that of LC-MS/MS,and the National Institute of Standards and Technology (NIST) reference values, to measure 25OHD levels.Methods/Setting:We compared serum 25OHD levels obtained from 219 elderly subjects, enrolled in a vitamin Dtrial, using the Diasorin Liaison platform assay, and the tandem LC-MS/MSmethod. We also assessed the perfor-mance of the Diasorin and Roche automated assays, expressed asmean % bias from the NIST standards, based onthe vitamin D External Quality Assessment Scheme (DEQAS) reports, from 2013 to 2017.Results: Serum 25OHD levels were significantly lower in the Diasorin compared to LC-MS/MS assay at baseline,18.5 ± 7.8 vs 20.5 ± 7.6 ng/ml (p b 0.001), and all other time points. Diasorin (25OHD) = 0.76 × LC-MS/MS(25OHD) + 4.3, R2 = 0.596. The absolute bias was independent of 25OHD values, and the pattern unfit forany cross-calibration. The proportion of subjects considered for vitaminD treatment based on pre-set cut-offs dif-fered significantly between the 2 assays. There also was wide variability in the performance of both automatedassays, compared to NIST reference values.Conclusion: The performance of most widely used automated assays is sub-optimal. Our findings underscore thepressing need to re-consider current practices with regard to 25OHD measurements, interpretation of resultsfrom research studies,meta-analyses, the development of vitaminD guidelines, and their relevance to optimizinghealth.

© 2018 Elsevier Inc. All rights reserved.

1. Introduction

Hypovitaminosis D is quite prevalent worldwide, and when chronic,may have negative, all-be-it subtle, repercussions on skeletal healthacross the life course [1–3]. The effects of this condition on a variety ofacute (asthma, infection) and chronic, non-communicable diseases(cancer, diabetes, and cardiovascular) are more debatable [1–3].

t.15366.Osteoporosis Program, WHOmerican University of Beirut

.

Serum 25-hydroxyvitamin D (25OHD) reflects vitamin D nutritionalstatus [4]. With the recognition of the high prevalence of vitamin D in-sufficiency in the general population, testing for 25OHD levels has be-come an increasingly common practice. The Vitamin D ExternalQuality Assessment Scheme (DEQAS), reported that in the Fall 2014,26 different assay methods were used in over 1100 laboratories from53 countries worldwide [5]. The most common types of assays proto-types used by various manufacturers today are the antibody-basedmethods (kit-based), the increasingly popular rapid high-output auto-mated methods (platform-based), and the high-performance liquidchromatography (HPLC)-based methods with either UV or mass spec-trometric (MS) detection [5]. The increasing high-test volume requiresa simple, reproducible, high-throughput method, which accuratelymeasures total circulating 25OHD (25OHD2 and 25OHD3), thus the

2 M. Rahme et al. / Metabolism Clinical and Experimental 89 (2018) 1–7

increasing popularity of platform assays in many commercial orhospital-based laboratories. In contrast, many academic researchcenters, and clinical centers of excellence are favoring the moreestablished “gold standard” of LC-MS/MS.

Data from theNational Health AndNutritional Epidemiologic Survey(NHANES) reveals that the mean serum 25OHD levels increased by 5–6 nmol/l (2–2.4 ng/ml) between the periods of 1998–2006 and 2007–2010, findings that may be in part explained by the use of differentassay methods, and possibly increasing vitamin D supplementation[6]. Our group has reported a similar temporal trend in Lebanesepatients [7]. Potential sources of variability between assays includedifferences in vitaminD assaymethods, vitaminDmetabolite extractionfrom vitamin D binding protein (DBP); differences in the detection of25OHD2, cross-reactivity with 3-epi-25OHD3 and other vitamin Dmetabolites, and matrix interferences, to name a few [8–11]. Thedifferences in measured serum 25OHD levels, within and also betweenassays, result in underestimation or overestimation of the actual 25OHDlevel depending on the specific assay used [5,9,10,12], and on the type ofsupplement used by the patient (Vitamin D2 or vitamin D3) [13,14]. Aspart of our quality assurance programwhen transitioning from a radio-immunoassay to an automated platform assay, we had compared re-sults from sera run in parallel on both assays on 494 samples, andshowed significant differences between the radio-immuno-assay andthe platform assay, differences of−38 to+19 ng/ml, a bias that was in-dependent of the serum 25OHD level [12].

The overall objective of this study is to investigate the variability inthe performance of vitamin D assays, and its impact on the assessmentof vitamin D status of subjects tested, and on interventions based onpre-set vitamin D guideline based cut-offs.

2. Methods

2.1. Study Design

We systematically assessed the performance of 2 of the mostcommonly used automated platform immunoassays to-date, tomeasure serum 25OHD, using two approaches.

First, we compared the performance of DiaSorin LIAISON® 25OHDtotal assay (DiaSorin, Stillwater, MN, USA) against the established liquidchromatography–tandemmass spectrometry (LC–MS/MS) assay, basedon samples obtained from elderly subjects enrolled in a vitamin D ran-domized trial, run in parallel using both assays.

Second,we assessed the performance of both theDiaSorin LIAISON®and the Roche Elecsys COBAS® (Roche Diagnostics International,Rotkreuz, Switzerland) platform assays, based on the calculated mean% bias obtainedwith both assays for the highest and the lowest NIST ref-erence standard value, using the quarterly reports obtained fromDEQAS, from 2013 to 2017.

2.2. Study Population and Trial Design

Elderly, overweight, ambulatory subjects with a serum 25OHDbetween 10 and 30 ng/ml (using the Diasorin platform assay) at screen-ing, were recruited from the greater Beirut Area. In total 257 elderlysubjects were randomly assigned to receive high dose (3750 IU/day)or low dose (600 IU/day) of vitamin D, and all subjects received1000 mg of elemental calcium citrate (Europharm); 35 subjects (14%)did not complete the study with no vitamin D data available poststudy discontinuation (NCT01315366) [15]. Briefly, 219 subjects hadbaseline 25OHD levels measured in parallel using both assays at studyentry. Subjects had a mean age (range) of 70 (67–74) years, a meanBMI of 29.2 (26.8–32.4) kg/m2, and 55% were females. Overall, 77%had one or more co-morbidity, and 69% met the definition of pre-diabetes but were not on oral hypoglycemic drugs [15].

2.3. Serum 25OHD Measurements and Assay Methodology

Samples weremeasured in parallel using Diasorin Liaison total rapidautomated assay chemiluminescent (Diasorin, Stillwater, MN) atAmerican University of Beirut (AUB), and the LC-MS/MS method atthe Mayo Clinic laboratories, Mayo Clinic, Rochester, Minnesota, USA.The LC-MS/MS at the Mayo Clinic Laboratories is the reference methodtraceable to the National Institute of Standards and Technology (NIST).Additional parallel measurements using both methods were done insubsets of patients at 6 months (N = 171) and 12 months (N = 126).Hormonal assays were run in batches and samples for levels drawn se-rially on the same patient (0, 6, 12 months) were included within thesame assay run, as singlets, both for Diasorin and LC-MS/MS assays.Serum 25OHD values are reported in nanograms per milliliter, androunded to one decimal, (to convert to nanomoles per liter (SI units),multiply by 2.496).

Both institutions are participants in DEQAS for several years (www.deqas.org). The Clinical Chemistry Laboratory at AUB also receivesquality assurance, evaluation, and accreditation from the College ofAmerican Pathologists (CAP) (www.cap.org).

2.4. Diasorin Liaison Platform Assay

Diasorin Liaison uses a chemiluminescent immunoassay whereserum 25OHD concentration is inversely proportional to the obtainedchemiluminescent signal. The manufacturer reports an analyticalrange of 4–150 ng/ml, an intra-assay precision (variability) of7.7–12.7% for values between 5.8 and 35 ng/ml, and an inter-assayprecision of 11.6–25% for values between 5.8 and 35 ng/ml. Cross reac-tivity is reported by the manufacturer to be 104% to 25 OHD2 and 100%to 25OHD3 (www.diasorin.com). However, according to the College ofAmerican Pathologists (CAP) Accuracy Based Vitamin D survey 2011,it was estimated to be 90% for 25OHD3 and 67% for 25OHD2 in ourlaboratory. At AUB, the intra-assay precision (CV) for duplicate analysiswas 4.61 ± 3.2%, and the inter-assay precision (CV) was 9.68 ± 5.35%,(n = 30 patients); and using manufacturer's quality control samplesthe inter-assay precisionwas 7.1% and 5.2% formean 25OHD concentra-tions of 16.6 and 55.4 ng/ml, respectively (n = 55).

2.5. Liquid Chromatography Mass Spectroscopy (LC-MS/MS)

Liquid Chromatography-Tandem Mass Spectrometry (LC-MS/MS)(ThermoFisher Scientific, Franklin, Massachusetts 02038 and AppliedBiosystems-MDS Sciex, Foster City, CA 94404) assays were performedat the Mayo Clinic Laboratories. Intra-assay CVs were 3.8%, 2.4%, and4.7% and inter-assay CVs were 6.4%, 6.8%, and 5.0%, at 24, 52, and140 ng/ml, respectively.

2.6. Assessment of Method Specific Assay Bias Based on Quarterly DEQASReports

Each quarter, DEQAS sends 5 unknown 25OHD samples to over 1100laboratories partaking in its quality assurance program. These unknownsamples are run by each laboratory based on their specific method andthe results are mailed back to DEQAS. Once all results are received,DEQAS generates a general report with details on method specificmean 25OHD values obtained for each method separately and specifiesthe NIST target value for each unknown sample. We used the quarterlyreports received fromDEQAS to calculate themethodmean % bias [calcu-lated as (Method specific mean total 25(OH)D level − NIST target 25(OH)D value ∕ NIST 25(OH)D target value) × 100], for the 2 mostcommonly used automated platform assays to-date, Diasorin Liaisonand Roche Elecsys, and for LC-MS/MS methods, for the period 2013–2017. We also further compared the performance of the Roche assaycalculating the laboratory individual % bias, obtained when DEQASquarterly samples were also assayed, independently and in parallel, in

Fig. 1. Panel A: Bland-Altman plot of the baseline difference between DiasorinLiaison and LC-MS/MS 25-OHD levels vs. baseline LC-MS/MS 25OHD in 219subjects (symbols). Solid horizontal line, mean difference between the two assays(or bias); dashed horizontal lines, 95% confidence interval. Panel B. Bland-Altmanplot of the baseline percentage difference [(LC-MS/MS − Diasorin Liaison) ∕ LC-MS/MS ∗ 100]. Solid horizontal line, mean difference between the two assays (orbias); dashed horizontal lines, 95% confidence interval.

3M. Rahme et al. / Metabolism Clinical and Experimental 89 (2018) 1–7

2 laboratories using this samemethodology over the same period at ourinstitution.

2.7. Statistical Analysis

Serum 25OHD levels were rounded to first decimal and mean orpercent differences were rounded to nearest integer. Continuousvariables were summarized using mean and standard deviation (SD),or median and range, as applicable based on normality of data assessedby the Shapiro-Wilk test. Difference in the mean 25OHD levels wastested using paired t-test or Wilcoxon signed rank test. Measurementsof these assays were regressed against each other, and a linearregression model fit was generated accordingly. Subjects were alsocategorized according to their 25OHD levels [Low (b10ng/ml), interme-diate (10–19.99ng/ml), desirable (≥20ng/ml)] based oneach assay, andagreement of patient grouping based on the comparison assays wastested using McNemar Test. Bland-Altman plots of the difference (orpercentage difference) in 25OHD levels between both assays versusthe LC-MS/MS value, considered the best method, were generated forall subjects at baseline, 6, 12 months, and for all time points combined.Analyseswere done using SigmaPlot 11.0 (Systat Software Inc., San Jose,CA) and IBM SPSS software version 24.0 (SPSS, Chicago, Illinois). Twotailed p-value b0.05 was considered significant.

3. Results

3.1. Comparison Between LC-MS/MS and Diasorin Liaison Assays

Baseline: Themean serum25OHD levels obtained on 219 subjects atbaseline using Diasorin Liaison were systematically and significantlylower 18.5±7.8 ng/ml (mean±SD), from the corresponding levels ob-tainedwith LC-MS/MS, 20.5±7.6 ng/ml, (p b 0.0001). The absolute biasassessed with Bland Altman graphs, averaged 2 ± 6 ng/ml, [95% CI =−10, 14] and was independent of the LC-MS/MS 25OHD value(Fig. 1A), and the percent bias averaged 6.9±29% (Fig. 1B). Comparableresults were obtained with serum 25OHD levels at 6 or 12 months, andin the complete dataset (Table 1, and Appendix Figures).

Results of linear regression analyses of serum 25OHD levels byDiasorin as a function of levels obtained by LC-MS/MS, at various timepoints, were as follows:

Complete data set (N = 516): 25OHD Liaison = 0.756 LC-MS/MS+ 4.26, R2 = 0.596

Baseline data set (N = 219): 25OHD Liaison = 0.693 LC-MS/MS+ 4.36, R2 = 0.4606 months data set (N = 171): 25OHD Liaison = 0.722 LC-MS/MS+ 6.16, R2 = 0.52812 months data set (N = 126): 25OHD Liaison = 0.653 LC-MS/MS+ 8.36, R2 = 0.490

3.2. Impact of Assay Methodology (LC-MS/MS and Diasorin Liaison) on Pa-tient Classification

There was a 10–15% difference in the classification of study subjectsas having less than desirable levels of serum 25OHD levels whether oneconsidered a cut-off of 20 ng/ml Institute of Medicine (IOM) [16], or30 ng/ml (Endocrine society guidelines) [17].

Baseline 25OHD values: selecting a cut-off at ≥20 ng/ml, 64% ofsubjects would then be classified as having less than desirable 25OHDlevels using the Diasorin Liaison, compared to 50% using the LC-MS/MS assay (p b 0.001). For a cut-off at ≥30 ng/ml, 91% of subjects wouldbe considered for treatment using Diasorin Liaison, compared to 89%with LC-MS/MS (p = 0.44).

Complete dataset: Using a desirable 25 OHD level of ≥20 ng/ml, 38%of subjectswould then be classified as having less than desirable 25OHD

levels using Diasorin Liaison, compared to 27% using the LC-MS/MSassay (p b 0.001) (Fig. 2A). Similarly, for a desirable 25OHD level of≥30 ng/ml, 74% of subjects would be considered for treatment usingDiasorin Liaison, compared to 68% according to results obtained by LC-MS/MS (p = 0.001) (Fig. 2A).

The distribution curve of 25OHD levels at baseline by LC-MS/MS hada more normal distribution, and was shifted to the right for low values,and to the left for high values, as compared to the one obtained usingthe Diasorin assay (Fig. 2B).

3.3. Concordance Rates Between the LC-MS/MS and Diasorin Liaison AssayResults

Concordance of proportions defined depends on the selected cate-gory approach. The overall concordance rate for classification betweenthe two assays was 80%, and for values below 10 ng/ml it was 67%(Fig. 3). The weighted k coefficients at baseline, 6, 12 months, and forall data points were 0.52, 0.43, 0.48, and 0.60 respectively, p b 0.04 forall coefficients.

To further evaluate differences between assays, the difference in25OHD levels obtainedwithboth assayswas calculated for each subject:the proportions of subjects that varied by b20% was 55%, between 20and 40% it was 34%, and by N40% it was 11% (Fig. 4).

Table 1Summary of 25OHD levels obtained in elderly RCT using LC-MS/MS and Diasorin Liaison 25OHD assays.

Test N Mean ± SD Median Minimum Maximum p-Value

Baseline AUB (Diasorin Liaison) (ng/ml) 219 18.5 ± 7.8 16.6 4 45.9 b0.0001Mayo (LC-MS/MS) (ng/ml) 219 20.5 ± 7.6 19 6 45Difference (ng/ml)a 219 2 ± 6.2 1.7 −30.9 19.2Difference (%)b 219 6.9 ± 29.3 10 −206 61.9

6 months AUB (Diasorin Liaison) (ng/ml) 171 28.1 ± 9.9 26.8 8.6 61 b0.0001Mayo (LC-MS/MS) (ng/ml) 171 30.4 ± 10 29 11 68Difference (ng/ml)a 171 2.3 ± 7.4 2.7 −17.5 28.4Difference (%)b 171 5.8 ± 23.2 10 −76.1 67

12 months AUB (Diasorin Liaison) (ng/ml) 126 28.5 ± 9.5 26.5 9.7 54.2 0.001Mayo (LC-MS/MS) (ng/ml) 126 30.9 ± 10.2 28 14 65Difference (ng/ml)a 126 2.4 ± 7.6 2.7 −26.8 28.3Difference (%)b 126 5.1 ± 26.4 9.5 −167.5 60.9

All AUB (Diasorin Liaison) (ng/ml) 516 24.1 ± 10.2 23.3 4 61 b0.0001Mayo (LC-MS/MS) (ng/ml) 516 26.3 ± 10.4 25 6 68Difference (ng/ml)a 516 2.2 ± 6.9 2.2 −30.9 28.4Difference (%)b 516 6.1 ± 26.7 9.8 −206 67

a Difference = 25OHD (LC-MS/MS)− (Diasorin Liaison).b %Difference = (LC-MS/MS− Diasorin Liaison) ∕ LC-MS/MS ∗ 100.

4 M. Rahme et al. / Metabolism Clinical and Experimental 89 (2018) 1–7

3.4. Performance of the Roche Elecsys PlatformAssay at AUBMC 2013–2017

The mean method specific % bias for the 2 most commonly usedautomated platform assays to-date, compared to NIST target valuesvaried widely overt the 5 year period. Indeed, the mean % bias variedfrom −18% to 11% for the Diasorin-Liaison assay, for the lowest NISTmean target value of 33.6 ± 8.1 nmol/l (13.5 ± 3.2 ng/ml), and from

Fig. 2. Proportion (Panel A) or frequency (Panel B) of subjects with 25OHD levels in thetotal cohort (n 516) above or at indicated assay specific 25OHD level. Diasorin Liaison(gray) and the LC-MS/MS (black) assays.

−17% to 8% for the Diasorin assay, for the highest NIST mean targetvalue of 100.1 ± 16.5 nmol/l (40 ± 6.6 ng/ml) (Fig. 5, Panels A and B).The variations were wider for the Roche Elecsys assay, between −27%to 24% for the lowest NIST mean target value, and from −3% to 35%for the highest NIST mean target value. Conversely, the accuracy of LC-MS/MS was superior, and although it had a consistent positive biascompared to target NIST values, it was, nevertheless, much closer tothe desirable range of ±5% bias, set by the vitamin D StandardizationProgram (VDSP) [9,10,18]. The individual % bias from NIST referencevalues obtained from running the unknown DEQAS samples in 2laboratories at our institution that use the same Roche Elecsys platformassay, also varied widely, from −30% to 11% and −37% to 35% for the

Fig. 3. Comparison of Diasorin Liaison vs. LC-MS/MS25OHD levels in the overall data set (n516) subjects. Symbols represent individual data points using both assays. Gray lightcircles represent concordant points at increasing 25OHD cut-offs [A (b10 ng/ml), B (10–20 ng/ml), C (N20 ng/ml)]. Dashed line, Line of identity; solid line, linear regressionmodel fit with its equation.

Fig. 4. Frequency distribution of the percentage difference in 25-OHD levels in the overalldata set 516 subjects.

5M. Rahme et al. / Metabolism Clinical and Experimental 89 (2018) 1–7

mean lowest NIST target value, and between −9% to 35% and −4% to22% for the mean highest NIST target value, over the same study period(Fig. 5, Panels C and D).

4. Discussion

We demonstrate a substantial difference in serum 25OHD levelsobtained with a commonly used platform automated assay, DiasorinLiaison, as compared to LC-MS/MS, at baseline and post-intervention,in a large sample of elderly overweight ambulatory individuals whocompleted a one-year vitamin D trial. Compared to NIST referencevalues, both the LC-MS/MS based assay, and the automated Diasorin Li-aison and Roche Elecsys platform assays showed variations in assay

Fig. 5. (A and B): Method specific mean % bias from NIST lowest (panel A) and highest (pamanufacturers and for LC-MS/MS methods. The number of laboratories varied by DEQAS repfrom NIST lowest (Panel C) and highest (Panel D) 25OHD target values, at 2 AUB laboratoriesmean using the lowest 25(OH)D NIST target value for each cycle, over the DEQAS report periovalue (panels B and D) obtained by deriving the mean using the highest 25(OH)D NIST targe6.6 ng/ml).

performance, that were however larger and more unpredictable forthe Roche Elecsys platform. Roche has recently released the ElecsysTotal II 25OHD assay, with an anticipated superior performance, andthat is certified by the CDC VDSP [19].

In our study, the platform Diasorin assay resulted on average inlower values than LC-MS/MS, with a concordance rate of only 67%with LC-MS/MS at values below 10 ng/ml values, and of 80% at highervalues. Considering the whole sample, almost ½ of the paired sampleshad a mean difference exceeding 20%. The distribution of the absolutedifference for individual paired values also spanned a wide range,between−30 to 19 ng/ml for baseline values, and −27 to 28 ng/ml, at12 months post-supplementation. The difference was random, thusruling out a systematic difference in calibration, and making anyattempt to correction to the more established LC-MS/MS valuesimpossible.

There is a multitude of methods to measure serum 25(OH) assays,and the wide variations noted between methods are not new[12,20–26]. Collection methods with additives (gel and clot activator)and presence of 25OHD2 and 3-epi 25OHD3 all affected results obtaineddepending on assay type [25]. Cavalier et al. demonstrated that re-standardization of the various commercial immunoassays will reducevariations noted, but that substantial differences remain if sera with adifferent matrix (pregnant women, or patients with renal failure) areused [26]. It is nevertheless increasingly worrisome as this situationpersists despite major efforts for standardization, such as launchedwith the VDSP effort almost 8 years ago [11,27]. Janssen et al. also com-pared the performance of various assays to the accepted LC-MS/MS, andshowed a similar modest concordance rate between the Diasorin andLC-MS/MS of 70% [14]. A study from Germany based on testing a totalof 300 surplus single EDTA plasma samples revealed regression equa-tions between the LC-MS/MS and Diasorin assays that were similar tothose we describe in our study [22].

nel B) 25OHD target values, for the 2 most commonly used platform assay methods/ort cycle, as shown in legend, depending on the cycle. (C and D): Method specific % bias. The mean lowest 25(OH)D NIST target value (panels A and C) obtained by deriving thed 2013–2017: 33.6 ± 8.1 nmol/l (13.4 ± 3.2 ng/ml). Mean highest 25(OH)D NIST targett value for each cycle, over the report period 2013–2017: 100.07 ± 16.5 nmol/l (40.0 ±

6 M. Rahme et al. / Metabolism Clinical and Experimental 89 (2018) 1–7

The positive bias of LC-MS/MS assay in our study has been noted byothers and seems to be well recognized. [28,29] The DEQAS programaffords further insights into the performance over time of various assaysincluding LC-MS/MS, comparing values obtained each quarter forunknown samples received by each laboratory to their correspondingNIST target values, in a large sample of laboratories using a spectrumof technologies worldwide [29]. Based on data obtained from quarterlycycles from 2013 to 2017, we demonstrate that themean bias values forLC-MS/MS compared to theNIST standards are consistently positive, buton average closest to the target 5%mean bias set by the VDSP. However,mean bias values are much more variable and unpredictable for theplatform assays. Indeed, they are most consistently negative for theDiasorin platform assay, as had been previously reported, but neverthe-less with wide variations, which are even larger and less predictablewith the Roche automated assay. Others have also reported on thelarge variations expressed as mean % bias from target NIST at anycycle, and that also varywidely across cycles for the same assaymethod,including the Diasorin platform assay [5,11].

DEQAS has allowed further systematic insight into sources of 25OHDassay variations as being, in part, due to interference from other circu-lating vitamin D metabolites, such as 24,25 (OH)2D metabolites (R orS isomers), and the 3-epimer 25(OH)D [30]. Indeed, while cross-reactivity with 24 hydroxylated metabolites is a major limitation withligand-binding assays, it is not with LC-MS/MS [30]. Conversely,whereas the 3-epimer shows cross-reactivity of 56% with competitiveprotein binding assays, it is not resolved by all LC-MS/MS assays, andis not detected in any of the antibody based methods [30].

Our study has some limitations. It did not systematically evaluatepre-analytical, analytical (matrix), or post-analytical factors that couldaccount for observed differences in 25OHD between assays. But someof these limitations are inherent to vitamin D measurements per se(for example skipping the extraction step in automated assays) anddo not reflect on the quality of work performed in our laboratories,both of which are DEQAS proficient. Although trial samples were notassayed within the same time frame at both institutions, blood sampleswere processed, sera separated, and immediately frozen at −70°, untilassay performance at AUB and were shipped on dry ice to the MayoClinic laboratories. Vitamin D is a steroid that is very stable as opposedto peptide hormones. We consider the LC-MS/MS assay to be the“gold standard” in our comparisons of samples obtained from theclinical trial, and although LC-MS/MS is the best option availabletoday, it is a heterogeneous group of methods with their own limita-tions. They nevertheless still do perform better considering the %mean bias data compared with NIST standards as compared to othertechnologies.

Our study has many important contributions. It is based on a largesample of specimens, spanning a wide spectrum of vitamin D levels,and the results obtained are therefore quite robust. We capture andcompare assay performance based on trial specimens over a protractedtime period (over 4 years, 2011–2015), thus confirming the on-goingnature of the problem of assay variation, in spite of the launch andcontinued efforts of the VDSP initiative. Based on our assessment of %bias from NIST standards across methods and manufacturers, ourstudy also confirms that the problem of assay variation is widespreadacross methodologies, and also on-going (2013–2017). Finally, we alsodemonstrate wide variation in % bias from NIST reference values withthe use of an identical methodology at 2 DEQAS certified laboratories,within the same institution.

In contrast to older days when Diasorin RIA was almost universallyused in research and practice and a single transposable cut-off wasused worldwide, the landscape is quite different and most challengingtoday. The increase in the flurry of assays over the last 2 decades, andthe above described assay variations havemajor implications for clinicalpractice, in terms of diagnosis, intervention thresholds, and target levelsreached post-supplementation [12,18]. Indeed, in our study classifica-tion using the same IOM recommended 25OHD cut-off of 20 ng/ml

varied by 10–15%, and this variation was lower using the Endocrinesociety desirable level of 30 ng/ml. The overall concordance ratebetween paired samples was only 80% and as low as 67% for 25OHDvalues below 10 ng/ml, resulting in projected large differences innumber of individuals who would be recommended supplementationat the public health level. Suh variations will continue to constitute amajor obstacle to harmonization of international vitamin D guidelines[11,12]. These assay limitations have also probably contributed to theincreasing number of reference and research laboratories that haveswitched to LC-MS/MS assays over time [31]. Although labor intensive,they still provide a better option if available, as we do not anticipatesubstantial further improvement in platform assays over the nearbyfuture [9,10,12,27,31,32].

In conclusion, the variability in 25 OHD assay performance thatplaguesmost platform automated technologies incursmajor limitationsto the quality of the evidence obtained from vitamin D randomizedtrials and meta-analyses, formulation of vitamin D guidelines, andthus ultimately affects interventions to be implemented at the publichealth level or and clinical practice. Considering the above, it would bemost advisable to abide by international recommendations, and refrainfrommeasuring serum 25OHD levels in apparently healthy ambulatoryindividuals [16,17,33]. Serum 25 OHD measurements should be per-formed only in high risk individuals (metabolic bone disorders, liverfailure, chronic kidney disease, certain medications, etc.) and in the set-ting of clinical trials. Measurements are ideally performed at centersthat consistently partake and fulfill the most rigorous vitamin D profi-ciency programs, such as VDSP [27,34]. Such approach would pave theway forward to optimize the quality of the evidence from research stud-ies, allow the determination of robust vitamin D intervention cut-offs,and harmonize public health and clinical practice strategies worldwide.

Funding

AUB, St Joseph University, LNCSR, NIH, and Mayo Clinic.

Acknowledgements

The authors are grateful to study subjects for their participation, andthank administrators at the LebaneseMinistry of Social Affairs dispensa-ries, local dispensaries, senior pharmacist at AUB, Mrs. Ulfat Usta, re-search staff, Mrs. Carmen Hajj Shahine and laboratory technicians fortheir time and dedication and making the study possible. The authorsthankEuro-PharmCanada for providing the vitaminD/identical placebotablets and calcium citrate supplements. The authors greatly appreciatethe time and input of members of the data safety monitoring board,Heike Bischoff-Ferrari MD, DrPH (University of Zurich, Switzerland),Christopher Gallagher, MD (Creighton University, USA) and ReinholdVieth PhD, FCACB (Mt Sinai Hospital, Montreal, Canada). We would liketo thank the expert input on assayperformance and interpretation of re-sults provided by Mr. Graham Carter, DEQAS, London, England, Drs.Hubert Vesper and Julia Botelho, Center of Disease Control, VDSP andDr. Chris Sempos Office of Dietary Supplements, National Institutes ofHealth. The study findings were presented by Dr. Neil Binkley, on behalfof study investigators, at theNIHOffice of Dietary Supplements, VitaminD Standardization meeting, on Nov 29, 2017.

The study was funded by the American University of Beirut, St. Jo-seph University, the Lebanese National Council for Scientific Research,and the National Institutes of Health, Fogarty International Center andthe Office of Dietary Supplements D43 TW009118.

Appendix A. Supplementary Data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.metabol.2018.09.003.

7M. Rahme et al. / Metabolism Clinical and Experimental 89 (2018) 1–7

References

[1] Holick MF. Vitamin D deficiency. N Engl J Med 2007;357(3):266–81.[2] Lips P, van Schoor NM. The effect of vitamin D on bone and osteoporosis. Best Pract

Res Clin Endocrinol Metab 2011;25(4):585–91.[3] Bouillon R, Van Schoor NM, Gielen E, Boonen S, Mathieu C, Vanderschueren D, et al.

Optimal vitamin D status: a critical analysis on the basis of evidence-basedmedicine.J Clin Endocrinol Metab 2013;98(8):E1283–304.

[4] El-Hajj Fuleihan G, Bouillon R, Clarke B, Chakhtoura M, Cooper C, McClung M, et al.Serum 25-hydroxyvitamin D levels: variability, knowledge gaps, and the conceptof a desirable range. J Bone Miner Res 2015;30(7):1119–33.

[5] Carter GD, Berry J, Durazo-Arvizu R, Gunter E, Jones G, Jones J, et al. Quality assess-ment of vitamin Dmetabolite assays used by clinical and research laboratories. J Ste-roid Biochem Mol Biol 2017;173:100–4.

[6] Schleicher RL, Sternberg MR, Lacher DA, Sempos CT, Looker AC, Durazo-Arvizu RA,et al. A method-bridging study for serum 25-hydroxyvitamin D to standardize his-torical radioimmunoassay data to liquid chromatography-tandem mass spectrome-try. Natl Health Stat Report 2016(93):1–16.

[7] Hoteit M, Al-Shaar L, Yazbeck C, Bou Sleiman M, Ghalayini T, Fuleihan Gel H.Hypovitaminosis D in a sunny country: time trends, predictors, and implicationsfor practice guidelines. Metabolism 2014;63(7):968–78.

[8] Phinney KW, Tai SS, BednerM, Camara JE, Chia RRC, Sander LC, et al. Development ofan improved standard reference material for vitamin D metabolites in humanserum. Anal Chem 2017;89(9):4907–13.

[9] Binkley N, Sempos CT. Standardizing vitamin D assays: the way forward. J BoneMiner Res 2014;29(8):1709–14.

[10] Sempos CT, Betz JM, Camara JE, Carter GD, Cavalier E, ClarkeMW, et al. General stepsto standardize the laboratory measurement of serum total 25-hydroxyvitamin D. JAOAC Int 2017;100(5):1230–3.

[11] Binkley N, Carter GD. Toward clarity in clinical vitamin D status assessment: 25(OH)D assay standardization. Endocrinol Metab Clin North Am 2017;46(4):885–99.

[12] Barake M, Daher RT, Salti I, Cortas NK, Al-Shaar L, Habib RH, et al. 25-Hydroxyvitamin D assay variations and impact on clinical decision making. J ClinEndocrinol Metab 2012;97(3):835–43.

[13] Ong L, Saw S, Sahabdeen NB, Tey KT, Ho CS, Sethi SK. Current 25-hydroxyvitamin Dassays: do they pass the test? Clin Chim Acta 2012;413(13–14):1127–34.

[14] Janssen MJ, Wielders JP, Bekker CC, Boesten LS, Buijs MM, Heijboer AC, et al. Multi-center comparison study of current methods to measure 25-hydroxyvitamin D inserum. Steroids 2012;77(13):1366–72.

[15] El-Hajj Fuleihan G, Baddoura R, Habib RH, Halaby G, Arabi A, RahmeM, et al. Effect ofvitamin D replacement on indexes of insulin resistance in overweight elderly indi-viduals: a randomized controlled trial. Am J Clin Nutr 2016;104(2):315–23.

[16] Ross AC, Manson JE, Abrams SA, Aloia JF, Brannon PM, Clinton SK, et al. The 2011 re-port on dietary reference intakes for calcium and vitamin D from the Institute ofMedicine: what clinicians need to know. J Clin Endocrinol Metab 2011;96(1):53–8.

[17] Holick MF, Binkley NC, Bischoff-Ferrari HA, Gordon CM, Hanley DA, Heaney RP, et al.Evaluation, treatment, and prevention of vitamin D deficiency: an Endocrine Societyclinical practice guideline. J Clin Endocrinol Metab 2011;96(7):1911–30.

[18] Thienpont LM, Stepman HC, Vesper HW. Standardization of measurements of 25-hydroxyvitamin D3 and D2. Scand J Clin Lab Invest Suppl 2012;243:41–9.

[19] Vitamin D Standardization-Certification Program (CDC VDSCP) Certified Total 25-Hydroxyvitamin D Procedures. https://www.cdc.gov/labstandards/pdf/hs/CDC_Certified_Vitamin_D_Procedures-508.pdf. [Last Accessed August 2018].

[20] Lai JK, Lucas RM, Banks E, Ponsonby AL. Variability in vitamin D assays impairs clin-ical assessment of vitamin D status. Intern Med J 2012;42(1):43–50.

[21] Glendenning P, Taranto M, Noble JM, Musk AA, Hammond C, Goldswain PR, et al.Current assays overestimate 25-hydroxyvitamin D3 and underestimate 25-hydroxyvitamin D2 compared with HPLC: need for assay-specific decision limitsand metabolite-specific assays. Ann Clin Biochem 2006;43:23–30.

[22] Roth HJ, Schmidt-Gayk H,Weber H, Niederau C. Accuracy and clinical implications ofseven 25-hydroxyvitamin D methods compared with liquid chromatography-tandem mass spectrometry as a reference. Ann Clin Biochem 2008;45:153–9.

[23] Binkley N, Krueger D, Cowgill CS, Plum L, Lake E, Hansen KE, et al. Assay variationconfounds the diagnosis of hypovitaminosis D: a call for standardization. J ClinEndocrinol Metab 2004;89(7):3152–7.

[24] De Koning L, Al-Turkmani MR, Berg AH, Shkreta A, Law T, Kellogg MD. Variation inclinical vitamin D status by DiaSorin Liaison and LC-MS/MS in the presence of ele-vated 25-OH vitamin D2. Clin Chim Acta 2013;415:54–8.

[25] Yu S, Cheng X, Fang H, Zhang R, Han J, Qin X, et al. 25OHD analogues and vacuumblood collection tubes dramatically affect the accuracy of automated immunoassays.Sci Rep 2015;5:14636.

[26] Cavalier E, Lukas P, Crine Y, Peeters S, Carlisi A, Le Goff C, et al. Evaluation of automatedimmunoassays for 25(OH)-vitamin D determination in different critical populationsbefore and after standardization of the assays. Clin Chim Acta 2014;431:60–5.

[27] Sempos CT, Vesper HW, Phinney KW, Thienpont LM, Coates PM. Vitamin D status asan international issue: national surveys and the problem of standardization. Scand JClin Lab Invest Suppl 2012;243:32–40.

[28] Berry DJ, Dutton J, Fraser WD, Jarvelin MR, Hypponen E. Harmonization study be-tween LC-MS/MS and Diasorin RIA for measurement of 25-hydroxyvitamin D con-centrations in a large population survey. J Clin Lab Anal 2017;31(3).

[29] DEQAS (Vitamin D External Quality Assessment Scheme) website. http://www.deqas.org/. [Last Accessed August 2018].

[30] Carter GD, Jones JC, Shannon J, Williams EL, Jones G, Kaufmann M, et al. 25-Hydroxyvitamin D assays: potential interference from other circulating vitamin Dmetabolites. J Steroid Biochem Mol Biol 2016;164:134–8.

[31] Carter GD, Berry J, Durazo-Arvizu R, Gunter E, JonesG, Jones J, et al. HydroxyvitaminDas-says: an historical perspective from DEQAS. J Steroid BiochemMol Biol 2018;177:30–5.

[32] Binkley N, Dawson-Hughes B, Durazo-Arvizu R, Thamm M, Tian L, Merkel JM, et al.Vitamin D measurement standardization: the way out of the chaos. J SteroidBiochem Mol Biol 2017;173:117–21.

[33] Hanley DA, Cranney A, Jones G, et al. Vitamin D in adult health and disease: a reviewand guideline statement from Osteoporosis Canada. CMAJ 2010;182(12):E610–8.

[34] Cashman KD, Kiely M, Kinsella M, Durazo-Arvizu RA, Tian L, Zhang Y, et al. Evalua-tion of Vitamin D Standardization Program protocols for standardizing serum 25-hydroxyvitamin D data: a case study of the program's potential for national nutritionand health surveys. Am J Clin Nutr 2013;97(6):1235–42.