British Journal of Nutrition (2024), 131, 248–255 doi:10.1017/S0007114523001782
© The Author(s), 2023. Published by Cambridge University Press on behalf of The Nutrition Society. This is an Open Access article, distributed
under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use,
distribution and reproduction, provided the original article is properly cited.
Comparison of test performance of two commonly used multiplex assays to
measure micronutrient and inflammatory markers in serum: results from a
survey among pregnant women in South Africa
Tsitsi Letwin Chimhashu1,2*, Hans Verhoef1, Elizabeth A. Symington3, Lizelle Zandberg2,
Jeannine Baumgartner2,4, Linda Malan2, Cornelius Marius Smuts2, Edith J. M. Feskens1 and
Alida Melse-Boonstra1
1Division of Human Nutrition and Health, Wageningen University & Research, Wageningen, The Netherlands
2Centre of Excellence for Nutrition, North-West University, Potchefstroom, South Africa
3Department of Life and Consumer Sciences, University of South Africa, Johannesburg, South Africa
4Department of Nutritional Sciences, King’s College, London, UK
(Submitted 23 March 2023 – Final revision received 7 July 2023 – Accepted 7 August 2023 – First published online 10 August 2023)
Abstract
The combined sandwich-ELISA (s-ELISA; VitMin Lab, Germany) and the Quansys Q-Plex™ Human Micronutrient Array (7-Plex) are multiplex
serum assays that are used to assess population micronutrient status in low-income countries. We aimed to compare the agreement of five
analytes, α-1-acid glycoprotein (AGP), C-reactive protein (CRP), ferritin, retinol-binding protein 4 (RBP4) and soluble transferrin receptor (sTfR)
as measured by the 7-Plex and the s-ELISA. Serum samples were collected between March 2016 and December 2017. Pregnant women (n 249)
were recruited at primary healthcare clinics in Johannesburg, and serum samples were collected between March 2016 and December 2017.
Agreement between continuousmeasurements was assessed by Bland–Altman plots and concordancemeasures. Agreement in classifications of
deficiency or inflammation was assessed by Cohen’s kappa. Strong correlations (r> 0·80) were observed between the 7-Plex and s-ELISA for
CRP and ferritin. Except for CRP, the 7-Plex assay gave consistently higher measurements than the s-ELISA. With the exception of CRP (Lin’s
ρ= 0·92), there was poor agreement between the two assays, with Lin’s ρ< 0·90. Discrepancies of test results difference between methods
increased as the serum concentrations rose. Cohen’s kappa for all the five analytes was< 0·81 and ranged from slight agreement (vitamin A
deficiency) to substantial (inflammation and Fe deficiency) agreement. The 7-Plex 1.0 is a research and or surveillance tool with potential for use
in low-resource laboratories but cannot be used interchangeably with the s-ELISA. Further optimising and validation is required to establish its
interchangeability with other validated methods.
Keywords: α-1-acid glycoprotein: C-reactive protein: Ferritin: Quansys 7-Plex assay: Retinol-binding protein: Soluble transferrin
receptor
Public health programmes in low- and middle-income countries the Quansys 5-Plex (2014) and 7-Plex (2017)(3,4) can measure
urgently need accurate, reproducible and cost-efficient methods several biomarkers in a single run, using only a small blood
to assess micronutrient status, with a view that these methods sample volume.
can be used to identify populations at risk, to determine the The s-ELISA was developed by Dr Juergen Erhardt of the
appropriate interventions and tomonitor programme impacts. In VitMin Lab (Willstätt, Germany) and has been widely used in
low-resource settings, representative population data on micro- studies and national surveys for micronutrient status determi-
nutrient status remain scarce(1). A huge barrier to this is a lack of nation(3–7). It has scored well in several critical assessments(8).
accurate, affordable and fast laboratory surveillance methods The s-ELISA simultaneously measures three micronutrient
that utilise a small blood sample volume to effectively assess a markers and two inflammatory markers in a single small volume
population’s prevalence and severity of micronutrient deficien- of sample (5 μl) and at a low cost: α-1-acid glycoprotein (AGP),
cies. The VitMin Lab combined sandwich-ELISA (s-ELISA)(2) and C-reactive protein (CRP), ferritin, retinol-binding protein 4 (RBP)
Abbreviations: AGP, α-1-acid glycoprotein; CRP, C-reactive protein; LoA, limits of agreement; NuPED, Nutrition during Pregnancy and Early Development; RBP4,
retinol-binding protein 4; s-ELISA, sandwich-ELISA; sTfR, soluble transferrin receptor.
* Corresponding author: Tsitsi Letwin Chimhashu, email tsitsi.chimhashu@wur.nl
https://doi.org/10.1017/S0007114523001782 Published online by Cambridge University Press
Multiplex assays to assess nutrient status 249
and soluble transferrin receptor (sTfR). The s-ELISA assay was Study design, setting and participants
shown to have good agreement with the ferritin RIA from Bio-
For the purpose of the current study, we used samples and data
Rad Laboratories (n 44), the sTfR ELISA assay from Ramco
collected from the NuPED study. The NuPED study was a
Laboratories (n 119) and the CRP ELISA assay from Immuno-
(2) prospective cohort study designed to follow women duringBiological Laboratories Inc. (n 17) . Furthermore, the s-ELISA
pregnancy and their infants until 12 months of age. Pregnant
assays produced mostly comparable results to the Roche
women were recruited at primary healthcare clinics in
reference-type assays for AGP, CRP and ferritin(9).
Johannesburg, whilst data and samples were collected at
TheQuansys 5-Plex (AGP, CRP, ferritin, RBP and sTfR) and 7-
RMMCH between March 2016 and December 2017 in
Plex (2017) assay also concurrently measures five or seven
Johannesburg, South Africa(13).
analytes: AGP, CRP, ferritin, RBP, sTfR, histidine-rich protein II
Pregnant women (n 250) were included in the study if aged
(HRP2; produced by malaria-causing Plasmodium falciparum
18–39 years,< 18weeks of gestationwith singleton pregnancies,
parasites) and thyroglobulin in a single sample aliquot (10 μl).
proficient in a local language, born in either South Africa or a
Researchers have described the Quansys machine as a non-
neighbouring country (Botswana, Lesotho, Zimbabwe), and
diagnostic research tool that offers assays that are accurate, easy-
resided in Johannesburg for at least 12 months. Women were
to-use, reliable and scalable. The Q-Plex assay is thought to have
excluded if they reported illicit drug use, smoked or were
the potential to overturn undesirable logistical hurdles such as diagnosed with a non-communicable disease such as hyper-
international shipping of samples for analysis and sometimes
(3) cholesterolemia and hypertension diabetes or renal disease.restrictive sample export/import legislation . They were also excluded when reporting to underlying and or
Over the years, researchers have compared the 7-Plex
serious illnesses such as tuberculosis, hepatitis or cancer. HIV
performance to that of the s-ELISA. Good agreement between a
status was not considered an exclusion criteria.
newly developed method to an established one is essential to
determine interchangeability or replacement of an established
method with a more advanced, faster and/or less expensive Blood collection and storage
newer method(10). Poor agreement between methods may lead Venous blood was drawn into 6-ml serum separator tubes (gold
to over- and or underestimation of deficiency prevalence tiger top tube BD Vacutainer) from the antecubital vein,
estimates that undermine appropriate policy development and centrifuged and processed within 1 h after blood draw. Serum
effective targeting of national interventions(11). Contrasting was stored at −20°C; for a maximum of 14 d until transportation
results, that is, good(3,4,12) and poor(7) agreement between the on dry ice (–80°C;) to the Centre of Excellence for Nutrition
7-Plex and s-ELISA from various studies, have led researchers to laboratories in Potchefstroom, South Africa, for storage at until
conclude that further studies are required to evaluate the validity analysed by Quansys 7-Plex (2018). Thereafter, frozen samples
of the Q-Plex assays against well-established methods. were shipped on dry ice to Wageningen University,
Furthermore, uncertainties with the precision of the 7-Plex Wageningen, the Netherlands (July 2020), where they were
may be of interest to most researchers in low-resource settings. stored at −80°C;, thawed for s-ELISA sample preparation,
These may be a result of blood sample type, laboratory refrozen and shipped on dry ice to the VitMin Lab, Willstaett,
equipment, experience of technicians and laboratory conditions Germany (May 2021).
such as ambient temperature and effect of freeze–thaw cycles. In
this study, we therefore aimed to compare the level of agreement
Description of the ELISA methods
between the Quansys 7-Plex and s-ELISA array results for AGP,
CRP, ferritin, RBP and sTfR with the use of blood serum samples Q-Plex Human Micronutrient (7-Plex) kits (Quansys
from pregnant women living in South Africa. Biosciences) were used according to kit instructions. Full details
on the Quansys Q-Plex Human Micronutrient ELISA can also be
found elsewhere(3,4). In our study, pooled samples were utilised
to assess quality control parameters during analysis. These
Subjects and methodology pooled samples served as control samples andweremeasured in
Ethical considerations duplicate for accuracy.
Details on the s-ELISA assay can be found elsewhere(2).
Written informed consent was obtained from all the women at Capture antibodies used were ferritin (Code A0133, Dako), sTfR
the first visit before data collection. Ethical approval was (Cat. No 4Tr26; Clone 23D10, Hytest), RBP (Code A0040, Dako)
obtained from the Human Research Ethics Committees of the and CRP (Code A0073, Dako). Detection antibodies were anti-
North-West University, Potchefstroom (NWU-00186-15-A1 and ferritin-horseradish peroxidase (Code P0145, Dako), anti-sTfR-
NWU-00456-19-A1) and the University of the Witwatersrand, horseradish peroxidase (Cat. No. 4Tr26-c; Clone 13E4, Hytest),
Johannesburg (M150968 and M161045). Approval was also anti-RBP-anti-ferritin-horseradish peroxidase (Code P0304,
given by The Gauteng Health Department, City of Johannesburg Dako) and anti-CRP-anti-ferritin-horseradish peroxidase(2).
District Research Committee and Clinical Manager of Rahima Two calibration curves were used for each analyte to cover a
Moosa Mother and Child Hospital (RMMCH). The Nutrition wider range (online Supplementary Fig. 1). Samples were
during Pregnancy and Early Development (NuPED) study was measured in duplicate. A serum quality control sample with a
conducted according to the guidelines laid down in the medium content of the five proteins was used to minimise the
Declaration of Helsinki. differences between plates and different measurement days on
https://doi.org/10.1017/S0007114523001782 Published online by Cambridge University Press
250 T. L. Chimhashu et al.
each 384-well plate at ten different positions. The absorption of with regression lines to indicate potential trends in the relation-
the study samples was adjusted based on themean absorption of ship between the differences and the magnitude of the
this quality control. Additionally, on each plate, a quality control measurements(10). Bias was defined as the difference in log-
from Biorad (Liqicheck Immunology Control) with low, medium transformed means of the two methods’ measures. Thereafter,
and high content of the five proteins was used to check the we back-transformed the bias to produce a geometric mean and
calibration curve in the low, medium and high range. The geometric SD. The geometric SD is a dimensionless, multiplicative
absolute values of all these quality controls were compared with factor such that dividing or multiplication of the geometric mean
results from The Vitamin A Laboratory-External Quality by this ratio indicates a variation that is equivalent to subtraction
Assessment (VITAL-EQA). The VITAL-EQA is a quality assurance or addition of 1 SD on a log-transformed scale. Limits of
programme that was set up by the CDC in 2003 to standardise the agreement (LoA) derived from log-transformed data were also
measurements of serum vitamins and micronutrient in back-transformed to yield limits for the ratio of actual
international studies(14). measurements(21). The LoA was calculated as the mean differ-
ence ± 1·96 SD of the differences, and it provides a range where
Definitions 95 % of the differences (when normally distributed) between the
two methods should fall(10,20). Furthermore, the bias and
Commonly used classificationmethods in literature were used to
agreement limits of CI can determine the sampling error in
estimate inflammation or deficiency between the two methods.
relation to the dimension of the sample(20). The level of
Fe deficiency in the absence of inflammation was defined as
μ (15) agreement and disagreement of the same variable by twoserum ferritin concentration< 15 g/l . Since it was measured
instruments on a continuous scale was determined by log-
with a Ramco kit, Fe deficiency erythropoiesis was defined as
transformed derived Lin’s concordance coefficient (ρ). The Lin
sTfR> 8·3 mg/l. Elevated CRP and AGP were defined as
(16) coefficient combinesmeasures of both precision and accuracy toconcentrations> 5 mg/l and> 1 g/l , respectively.
determine how meaningful the distance from the line of perfect
Inflammatory status was defined as no inflammation (CRP
≤ concordance. This 45° line quantifies the dis(agreement)concentration 5 mg/l and AGP concentration≤ 1 g/l) v. any
between the set of analytes that were measured by two different
inflammation (elevated CRP and/or AGP). The internal correc-
assays (see refs. 22 and 23 for further details). We interpreted
tion factor (CF) approach proposed by Thurnham et al. (2012)
P< 0·90 as an indication of poor agreement, and 0·90–0·94,
was used to adjust Fe and vitamin A indicators for inflammation
0·95–0·99 and> 0·99 as moderate, substantial and almost perfect
before categorising them into one of the four categories: (1)
levels of agreement, respectively(24).
reference (CRP concentration≤ 5mg/l and AGP concentration≤
Lastly, to evaluate agreement in categorical test results, we
1 g/l); (2) incubation (CRP concentration> 5 mg/l and AGP
≤ used Cohen’s kappa coefficient (κ) with its 95 % confidence andconcentration 1 g/l); (3) early convalescence (CRP concen-
percent agreement interval. The kappa coefficient provides the
tration> 5 mg/l and AGP concentration> 1 g/l); and (4) late
≤ ability of the two assays to define and classify deficiency andconvalescence (CRP concentration 5 mg/l and concentration
inflammation the same, adjusted for how often they agree by
AGP> 1 g/l)(15). With the assumption that RBP occurs in a 1:1
chance. The benchmark scale was used to estimate the kappa
ratio with retinol with no variability in this ratio, vitamin A
statistics:< 0·00 was interpreted as poor, 0·00–0·20 as slight;
deficiency was defined as RBP< 0·70 μmol/l(17,18).
0·21–0·40 as fair; 0·41–0·60 as moderate; 0·61–0·80 as good
and≥ 0·81 as very good(25).
Statistical analysis
Stata software version 16 (StataCorp LLC.) and SPSS version 26
(IBM Corp) was used for analysis. As all the outcome variables
Results
were non-normally distributed, analyte concentrations are
described as median and 25th–75th percentile. All analyte The inter-assay CV for the control sample from the 7-Plex
concentrations were unadjusted for inflammation. analysis were as follows: AGP 11·6 %, CRP 21·1 %, ferritin 14·0 %,
Several statistical evaluation measures were used to compare RBP 10·9 % and sTfR 25·5 %. Whilst the inter-assay CV for the
the reproducibility of results between the 7-Plex and s-ELISA. control sample were from the s-ELISA analysis were AGP 8·1 %,
First, after natural log-transformation, to quantify the extent of CRP 5·8 %, ferritin 2·3 %, RBP 3·6 % and sTfR 3·6 %.
linearity between the two measurements, correlations between One subject had no sample analysed for the s-ELISA and was
the two methods were determined by Pearson’s correlation therefore removed from the 7-Plex analysis, leaving 249 data
coefficients (r) and scatterplots with the line of identity. Second, points for comparison of AGP, CRP, ferritin and RBP measure-
Bland–Altman analysis (difference plots and statistics(19)) were ments. For sTfR, we excluded a further five values of
used to assess agreement between analyte measurement results measurement for the s-ELISA because they were below the
from the two assays. We did this because high r correlation and limit of quantification, and one 7-Plex outlier (7-Plex: 45·8 mg/l;
scatter of points lying near the line of identity do not s-ELISA: 3·60 mg/l) that was highly influential in determining the
automatically imply that there is good agreement between the slope of agreement in the Bland–Altman plot. Thus, the final
two methods(20). To address the relationship between the non- sample size was 243.
constant difference, or the assumption of homoscedastic Table 1 shows the characteristics of the study population. The
variance (non-constant variance) violations, we produced mean age was 27 years, and most women were in the second
Brand–Altman plots again with log-transformed variables, and trimester of pregnancy. The distributions of analyte
https://doi.org/10.1017/S0007114523001782 Published online by Cambridge University Press
Multiplex assays to assess nutrient status 251
Table 1. Descriptive characteristics of study population (n 249) differences between the 7-Plex and s-ELISA measurements
increased as the magnitude of their average concentration
Values
increased (Fig. 2).
Median 25th–75th percentile
Age (years) 27 24·0–32·0
Gestational age, by ultrasound (wks) 14 12·0–16·0
Body weight (kg) 67·5 58·0–77·8 Discussion
BMI (kg/m2) 26·3 23·0–30·6
MUAC (cm) 29·9 27·2–33·1 Our results showed very strong correlations for CRP and ferritin
n % between results obtained by Quansys Human Micronutrient Q-
HIV-positive 64 25·7 Plex (7-Plex) version 1.0 and the s-ELISA from VitMin Lab, and
MUAC, mid-upper arm circumference. moderate concordance for CRP, whereas the other analytes
performed less well. Furthermore, the 7-Plex showed largely
comparable prevalence estimates with the s-ELISA, except for
concentrations as measured by the 7-Plex and s-ELISA assays are vitamin A deficiency.
presented in Table 2. The prevalence of inflammatory status and Because neither the s-ELISA nor the 7-Plex are reference
Fe deficiency was similar between the two methods. However, standard methods and thus not known to be superior to the
prevalence estimates for vitamin A deficiency (5 % v. 0·8 %) other, the comparative study presented in this paper should not
differed substantially between the two methods, as also be seen as a method validation, but rather as an evaluation to
indicated by the corresponding κ values and % agreement determine if results from the 7-Plex can be used interchangeably
presented in Table 3. Overall, using Cohen’s kappa, inter- with those from an established laboratory or assay(7). It should
definitional agreement between the two assays was slight to also be noted that newer versions of Quansys Biosciences are
substantial (Table 3), with inflammation definitions having being developed (A. Nelson, personal communication, 2021),
substantial agreements across three different evaluation which may result in different test performance.
approaches. The percentage of pregnant women classified To our knowledge, only one other study(3) compared the 7-
differently between the two assays was 8 % for ID, 11 % for Plex and s-ELISA in a population at high risk of micronutrient
elevated CRP, 13 % for elevated AGP, 11 % for Fe-deficient deficiencies. In that study, agreement between assay measure-
erythropoiesis and 39 % for vitamin A deficiency. ments was quantified using the absolute differences between
Scatter graphs based on log-transformed values showing observations made using the two methods on the same subjects,
correlations between the two methods are presented in Fig. 1. and determining 95 % reference intervals for these
The correlations between the 7-Plex and s-ELISA results were differences(21). By contrast, we used an extension of this
moderate for AGP (r= 0·58), RBP (r= 0·70) and sTfR (r= 0·67) statistical technique by using logarithmic transformation and
and strong for CRP (r= 0·93) and ferritin (r= 0·89). regression approach for non-uniform differences, which is more
Bias, LoA, and correlation coefficients of AGP, CRP, ferritin, appropriate given the distributional characteristics of and the
sTfR and RBP concentrations comparing the 7-Plex and the s- associations between the pairs of variables investigated(10).
ELISA assays are described in Table 4. For all the analytes, the A serum retinol (vitamin A) concentration of≤ 0·70 μmol/l is
log-transformed constant variance model was selected. recommended by theWHO as a marker to assess the population
Geometric mean values obtained by 7-Plex were 9 %–34 % burden of vitamin A deficiency(26,27). Its measurement requires
higher than values obtained by s-ELISA assays. We observed HPLC, which is expensive, technically demanding and rarely
moderate concordance for CRP (Lin’s ρ = 0·92), but poor available in developing countries. Instead, serum RBP concen-
agreement for AGP, ferritin, RBP and sTfR. For all the biomarkers tration is used as a proxy measure. Because international
assessed, the 7-Plex produced higher values than the s-ELISA, reference material for RBP is currently unavailable, researchers
with the largest difference in analytes measured being largest for have used different reference methods to evaluate the
ferritin readings. performance of RBP assays(12). For example, in one study,
Visual inspection of these plots on the original scales RBP was compared with retinol as a reference assay, under the
appeared to violate more than one of the assumptions for the assumption that RBP and retinol circulate in plasma at one-to-
Bland–Altman LoA approach: either there was a relationship one molar ratio. The RBP values obtained from the VitMin Lab s-
between the difference and the mean (non-constant difference), ELISA are adjusted in the analysis to retinol equivalents by
or the assumption of homoscedastic variance (non-constant simultaneous measurement of control samples with known
variance) was violated. Visual inspection of the Bland–Altman retinol concentration, whereas those from the Q-Plex 7 assay are
plots showed that, of the five analytes, the non-constant unadjusted for retinol. Quansys Biosciences encourages the user
difference for AGP and sTfR improved the most (online to make their own inferences of the generated RBP values and
Supplementary Table 2). In Fig. 2, good agreement for CRP, advises that care should be taken when interpreting the RBP4
ferritin and sTfR analyte concentrations as well as upward trend, value as the results obtained will not provide an estimation of
that is, the difference between the measures is a function of the retinol in the sample. Thus, it is advised that RBP values are
average of the measures, is shown. For all analytes but calibrated by determining retinol concentration with HPLC on a
sTfR,< 5 % of the values were outside the LoA, with the majority subsample of the same blood samples (A. Nelson, Quansys
lying above the upper limit. In addition, the variability of Biosciences, personal communication, 2021).
https://doi.org/10.1017/S0007114523001782 Published online by Cambridge University Press
252 T. L. Chimhashu et al.
Table 2. Micronutrient and inflammatory marker concentrations as measured by 7-Plex and s-ELISA
7-Plex s-ELISA
Median 25th–75th percentile Median 25th–75th percentile
AGP (g/l) 0·75 0·63–0·86 0·69 0·58–0·83
CRP (mg/l) 6·48 3·08–14·36 6·92 3·12–13·1
Ferritin (μg/l) 54·4 28·4–116·4 49·1 26·7–79·0
RBP (μmol/l) 1·54 1·27–1·80 1·11 0·89–1·41
sTfR (mg/l) 4·82 3·77–6·61 4·68 3·87–6·03
n % n %
AGP > 1 g/l 28 11·2 24 9·60
CRP > 5 mg/l 148 59·4 155 62·2
Ferritin < 15 μg/l 32 12·9 35 14·1
RBP< 0·70 μmol/l 2 0·80 14 5·60
sTfR> 8·3 mg/l 38 15·3 23 9·40
AGP, α-1-acid glycoprotein; CRP, C-reactive protein; RBP, retinol-binding protein; sTfR, soluble transferrin receptor.
AGP, CRP, ferritin, RBP (n 249), and sTfR (n 243; see text).
Table 3. Agreement between 7-Plex and s-ELISA in classification of inflammation or deficiency categories
n % Agreement Kappa, κ 95% CI
Serum AGP concentration > 1 g/l 249 87·2 0·31 0·13, 0·50
Serum CRP concentration > 5 g/l 249 89·2 0·77 0·69, 0·85
Inflammatory status* 249 87·6 0·74 0·65, 0·82
Categorical correction factor† 249 78·2 0·62 0·54, 0·71
Fe deficiency‡ 249 91·6 0·64 0·50, 0·78
Fe-deficient erythropoiesis§ 243|| 88·5 0·47 0·31, 0·63
Vitamin A deficiency¶ 249 60·6 0·12 0·05, 0·19
s-ELISA, sandwich-ELISA; AGP, α-1-acid glycoprotein; CRP, C-reactive protein; CF, correction factor.
* Inflammatory status was defined as no inflammation (serum concentration of both CRP≤ 5 mg/l and AGP≤ 1 g/l) v. any inflammation (elevated CRP and/or AGP).
† Categorical CF approach as proposed by Thurnhamet al.(15) divided inflammation into four categories: (a) reference (CRP concentration≤ 5mg/l and AGP concentration≤ 1 g/l); (b)
incubation (CRP concentration >5 mg/l and AGP concentration≤ 1 g/l); (c) early convalescence (CRP concentration >5 mg/l and AGP concentration> 1 g/l); and (d) late
convalescence (CRP concentration≤ 5 mg/l and AGP concentration > 1 g/l.
‡ Fe deficiency: serum ferritin concentration< 15 μg/l.
§ Fe-deficient erythropoiesis: serum-soluble transferrin concentration> 8·3 mg/l.
|| Six values were excluded (see text).
¶ Vitamin A deficiency: serum retinol-binding protein concentration< 0·70 μmol/l.
Calibration of RBP concentration to retinol concentration and comparing the 7-Plex with s-ELISA in pregnant Nigerian women,
subsequent dichotomisation of RBP concentration to values≤ as well as poor agreement for elevated AGP. Contrary to our
0·70 μmol/l is problematic for two reasons. First, surveys in study, the NiMaNu cohort study found strong correlations and
humans indicate that the molar concentration of retinol in serum good agreement between methods for AGP, ferritin, RBP and
can differ from that of RBP so that the molar ratio can differ from sTfR(3). We found good agreement between the two ELISA
1:1 depending on inflammation, protein-energy malnutrition, methods within lower concentration ranges. In our study, the
obesity, vitamin A status, Fe status and pregnancy(28). Second, discrepancies in values were at the higher end of the
the selection of a cut point of 0·70 μmol/l for RBP concentrations distributions. We, however, observed that all analyte concen-
can lead to biased estimates of the prevalence of vitamin A trations in our sample were in the upper range of what was
deficiency as defined by serum retinol concentration of≤ 0·70 reported NiMaNu cohort study. This might be due to the
μmol/l(18). Thus, we recommend that RBP concentrations are difference in pregnancy stages of the sampled women. Pregnant
measured and reported without prior calibration to retinol women (n 206) included in the comparison study of the 7-Plex
concentration, and that cut points for dichotomisation of RBP and s-ELISA were randomly selected from the original NiMaNu
values are selected depending on diagnostic aims of the study. In study sample pool of 654 plasma samples, and majority of these
a study among Kenyan children, it was also shown that the women were mostly in the third trimester(3,29). Analyte concen-
diagnostic performance of RBP concentration in assessing trations are known to be lowered by physiological haemodilu-
vitamin A deficiency is good, but it can be improved by adding tion during the progression of pregnancy(30).
serum transthyretin concentration(18). Further studies are needed There can be large differences among sTfR assays in the cut-
to confirm this finding, with a view to potentially incorporate offs used to define Fe-deficient erythropoiesis that is due to a lack
transthyretin as a target marker into multiplex micronutrient of a common referencematerial, differences between antibodies
assays. For more extensive discussion of these issues, we refer to used in various assays(2,3,31), so that the results of different assays
a previous paper(18). are not directly comparable(7,9).
Similar to our results, the NiMaNu cohort study(3) also found A factor that could have affected our results was the extra
strong relationships and good agreement for CRP when freeze–thaw cycle that the s-ELISA samples had to undergo
https://doi.org/10.1017/S0007114523001782 Published online by Cambridge University Press
Multiplex assays to assess nutrient status 253
ρ = 0·56
r= 0·58
ρ = 0·92
r= 0·93
ρ = 0·83
r= 0·89
ρ = 0·50
r= 0·70
ρ = 0·67
Fig. 2. Regression-based Bland–Altman plots showing differences between
r= 0·67 results from 7-Plex and s-ELISA immunoassay (y-axes) plotted against average
concentrations (x-axes). AGP, CRP, ferritin, RBP (n 249), and sTfR (n 243; see
text). Differences based on log-transformed variables. Solid blue line: linear
regression line indicating the non-constant difference; solid red line: 95% limits of
Fig. 1. Comparisonof serumconcentrationsof selected analytesmeasured by 7- agreement (meandifference ± 2 SD) calculated from linear regression; dotted green
Plex v. the s-ELISA methods. All variables are in log scale. AGP, CRP, ferritin, line: line of identity (perfect concordance). AGP, α-1-acid glycoprotein; CRP, C-
RBP (n 249) and sTfR (n 243; see text). Solid red line: prediction line; dotted green reactive protein; RBP, retinol-binding protein, sTfR, soluble transferrin receptor.
line: line of identity (perfect concordance). AGP, 1-acid-glycoprotein; CRP, C-
reactive protein; RBP, retinol-binding protein; sTfR, soluble transferrin receptor.
before analysis. However, Esmaeili et al. (2018) demonstrated reported that, except for sTfR, undiluted serum samples can
that the 7-Plex assay has good stability up until five freeze–thaw undergo several freeze–thaw cycles without marked changes in
cycles for all five analytes. In addition, Erhardt et al. (2004) analyte concentrations.
https://doi.org/10.1017/S0007114523001782 Published online by Cambridge University Press
254 T. L. Chimhashu et al.
Additionally, we also observed high CV for the 7-Plex assay.
This could have beendue tomanual pipetting into platewells. In a
recent 7-Plex cross-lab analysis, the equipment in a laboratorywas
found to be a source of imprecision(12). Another limitation of our
study was that acceptable values for the LoA (95% of the
differences to lie between± 2 SD) were not established a priori(20).
We found that compared with other comparison study results
of the 7-Plex that used plasma samples from pregnant Nigerian
women, there was a lack of high concordance in all five analytes
in our study(3). This could have been due to our use of serum
samples. It has been suggested that 7-Plex results may be due to
different sample preparation, that is, the use of serum instead of
plasma samples(7). This is further corroborated by a report that
EDTA plasma produced 74% higher Q-Plex sTfR concentrations
compared with serum. However, although stated by Quansys
Biosciences that the assays are accurate for measuring micro-
nutrients in both serum and plasma, future work could investigate
whether the difference in sample preparation affect results.
In conclusion, values observed for all five analytes from the
7-Plex were within the expected ranges at community level in
low- and middle-income countries. Except for CRP concen-
trations, however, the 7-Plex assay gave consistently higher
readings than the s-ELISA and the difference between methods
increased as the serum concentrations rose. Thus, the 7-Plex
assay cannot be used interchangeablywith the s-ELISAmethod.
We concur with the manufacturer and several earlier studies
that the 7-Plex should be used only as a research or as a
community surveillance tool(12) and not for clinical diagnostic
purposes in individuals.
Acknowledgements
The authors thank the pregnant women who participated in the
NuPED study. Dr Juergen Erhardt for conducting the VitMin
ELISA analyses and Cecile Cooke (Centre of Excellence for
Nutrition, North-West University, South Africa) for sample
management and laboratory analysis. The authors would also
like to thank the following people for their contribution to the
NuPED primary study: Professor A Coovadia, Professor H
Lombaard, Dr AJ Wise, Dr E Loock, Dr R Adams, Professor M
Faber, Dr O Sotunde, Dr M Rothman, Dr L Siziba, the
sonographers and nurses of Rahima Moosa Mother and Child
Hospital, the nurses at the primary healthcare clinics and the
fieldworkers.
The project was financially supported by PepsiCo. The
NuPED study was supported in part by the National Research
Foundation of South Africa, Unique Grant 99374 (EAS) (https://
www.nrf.ac.za/) and the South African Medical Research
Council under a Self-Initiated Research Grant (CMS) (http://
www.mrc.ac.za/).
Conceptualisation of this study: A. M. B. and T. L. C.;
Methodology: T.L. C. and A. M. B.; Statistical analyses: T. L. C.;
Data analysis support and interpretation: H. V.; Execution of
primary study and data collection: E. A. S., L. Z., J. B., L. M. and C.
M. S.; s-ELISA analysis: E. J. M. F.; Writing – review and editing: T.
L. C., H. V., L. M., E. J. M. F., C. M. S. and A. M. B.; all authors read
and edited the manuscript.
Table 4. Performance indicators of 7-Plex compared with s-ELISA for measurement of serum ferritin, sTfR, AGP, CRP and RBP concentrations†
Lin’s concord- % Outside
Original scale data P Ln-scale data P Original scale data P Ln-scale data P Pearson’s correlation coefficient§ Bias|| ance§ LoA|| LoA
Parameter Non-constant difference‡ Non-constant variance§ r Mean SD rho 95% CI 95% CI n %
AGP 0·004 0·688 0·995 0·115 0·58 1·09 1·28 0·56 0·47, 0·64 0·67, 1·78 12 4·82
CRP < 0·001 0·027 0·003 0·992 0·93 1·09 1·65 0·92 0·90, 0·94 0·41, 2·91 11 4·42
Ferritin < 0·001 < 0·001 0·991 0·999 0·89 1·25 1·70 0·83 0·80, 0·87 0·44, 3·54 11 4·42
RBP 0·010 0·001 < 0·001 0·994 0·70 1·34 1·30 0·50 0·43, 0·57 0·80, 2·23 10 4·02
sTfR* 0·004 0·211 0·998 0·998 0·67 0·99 1·42 0·67 0·60, 0·74 0·49, 1·98 15 6·17
s-ELISA, sandwich-ELISA; sTfR, soluble transferrin receptor; AGP, α-1-acid glycoprotein; CRP, C-reactive protein; RBP, retinol-binding protein; LoA, limits of agreement.
*sTfR (n 243) (six values excluded as they were outside of the limit of quantification range or they were influential outliers; see text).
† AGP, CRP, ferritin, RBP (n 249).
‡ The P value indicates the probability of finding the values for the coefficient for slope as extreme as observed or more extreme, assuming that in fact no association between the results from the two assays (coefficient for slope is zero).
§ The P value indicates the probability of finding the values for the coefficient for slope as extreme as observed or more extreme of the absolute residuals. The residuals are computed from a regression of the difference on the averages. Data
analysis based on ln-scale data.
|| Mean relative difference (bias) and LoA have been back-transformed from the natural log scale.
https://doi.org/10.1017/S0007114523001782 Published online by Cambridge University Press
Multiplex assays to assess nutrient status 255
None of the researchers had a conflict of interest. None of the South Africa: a study protocol for a prospective cohort. BMC
study funders had any involvement in the study design, analysis, Pregnancy Childbirth 18, 308.
or interpretation of data, the writing of the report, or the decision 15. Chaudhary-Webb M, Zhang M & Pfeiffer CM (2023) CDC’s
Vitamin A laboratory-external quality assessment program
to submit the article for publication.
shows little change in laboratory performance for multiple
nutritional biomarkers over a 10-year period (2008–2017).
J Nutr 153, 373–384.
Supplementary material 16. World Health Organization (2011) Serum Ferritin
For supplementary material/s referred to in this article, please Concentrations for the Assessment of Iron Status and Iron
Deficiency in Populations. Geneva: World Health
visit https://doi.org/10.1017/S0007114523001782
Organization.
17. Thurnham DI & McCabe GP (2012) Influence of Infection
and Inflammation on Biomarkers of Nutritional Status with
References an Emphasis on Vitamin A and Iron. World Health
1. Osendarp SJM, Brown KH, Neufeld LM, et al. (2020) The Organization Report: Priorities in the Assessment of
double burden of malnutritionfurther perspective. Lancet 396, Vitamin A and Iron Status in Populations, Panama City,
813. Panama. pp. 15–17. https://cdn.who.int/media/docs/
2. Erhardt JG, Estes JE, Pfeiffer CM, et al. (2004) Combined default-source/micronutrients/background-paper4-report-
measurement of ferritin, soluble transferrin receptor, retinol assessment-vitaandiron-status.pdf?sfvrsn=cc563dde_2 (accessed
binding protein, and C-reactive protein by an inexpensive, January 2023).
sensitive, and simple sandwich enzyme-linked immunosorbent 18. World Health Organization (2009) Global Prevalence of
assay technique. J Nutr 134, 3127–3132. Vitamin A Deficiency in Populations at Risk 1995–2005:
3. Brindle E, Lillis L, Barney R, et al. (2017) Simultaneous WHO Global Database on Vitamin A Deficiency. Geneva:
assessment of iodine, iron, vitamin A,malarial antigenemia, and World Health Organization.
inflammation status biomarkers via a multiplex immunoassay 19. Talsma EF, Verhoef H, Brouwer ID, et al. (2015) Proxy markers
method on a population of pregnant women from Niger. PLoS of serum retinol concentration, used alone and in combination,
One 12, e0185868. to assess population vitamin A status in Kenyan children: a
4. Brindle E, Stevens D, Crudder C, et al. (2014) A multiplex cross-sectional study. BMC Med 13, 1–9.
immunoassay method for simultaneous quantification of iron, 20. Steichen TJ & Cox NJ (2002) A note on the concordance
vitamin A and inflammation status markers. PLoS One 9, e115164. correlation coefficient. Stata J 2, 183–189.
5. Fu Q, Zhu J & Van Eyk JE (2010) Comparison of multiplex 21. Giavarina D (2015) Understanding Bland Altman analysis.
immunoassay platforms. Clin Chem 56, 314–318. Biochem Med 25, 141–151.
6. Tighe PJ, Ryder RR, Todd I, et al. (2015) ELISA in the multiplex 22. Bland JM & Altman DG (1986) Statistical methods for assessing
era: potentials and pitfalls. Proteomics Clin Appl 9, 406–422. agreement between two methods of clinical measurement.
7. Esmaeili R, Zhang M, Sternberg MR, et al. (2019) The Quansys Lancet 327, 307–310.
multiplex immunoassay for serum ferritin, C-reactive protein, 23. Lin LI (1989) A concordance correlation coefficient to evaluate
and α-1-acid glycoprotein showed good comparability with reproducibility. Biom 45, 255–268.
reference-type assays but not for soluble transferrin receptor 24. Liao JJ & Lewis JW (2000) A note on concordance correlation
and retinol-binding protein. PLoS One 14, e0215782. coefficient. PDA J Pharm Sci 54, 23–26.
8. Karakochuk CD, Henderson AM, Samson KLI, et al. (2018) 25. McBride GB (2005) A proposal for strength-of-agreement
Comparison of a newmultiplex immunoassay formeasurement criteria for Lins Concordance Correlation Coefficient. NIWA
of ferritin, soluble transferrin receptor, retinol-binding protein, Client Rep: HAM2005–062 45, 307–310.
c-reactive protein and α¹-acid-glycoprotein concentrations 26. Landis JR & Koch GG (1977) The measurement of observer
against a widely-used s-ELISA Method. Diagn 8, 13. agreement for categorical data. Biom 33, 159–174.
9. Karakochuk CD, Whitfield KC, Rappaport AI, et al. (2017) 27. World Health Organization (2011) Serum Retinol
Comparison of four immunoassays to measure serum ferritin Concentrations for Determining the Prevalence of Vitamin A
concentrations and iron deficiency prevalence among non- Deficiency in Populations. Geneva: WHO.
pregnant Cambodian women and Congolese children. Clin 28. World Health Organization (1996) Indicators for Assessing
Chem Lab Med 55, 65–72. Vitamin A Deficiency and their Application in Monitoring and
10. Fischer CM, ZhangM, SternbergMR, et al. (2021) TheVitMin lab Evaluating Intervention Programmes. Geneva: World Health
Sandwich-ELISA assays for iron and inflammation markers Organization.
compared well with clinical analyzer reference-type assays in 29. Engle-Stone R, Haskell MJ, Ndjebayi AO, et al. (2011) Plasma
subsamples of the Nepal national micronutrient status survey. J retinol-binding protein predicts plasma retinol concentration in
Nutr 152, 350–359. both infected and uninfected Cameroonian women and
11. Bland JM & Altman DG (1999) Measuring agreement in method children. J Nutr 141, 2233–2241.
comparison studies. Stat Methods Med Res 8, 135–160. 30. Wessells KR, Ouedraogo CT, Young RR, et al. (2017)
12. Laillou A, Yakes E, Le TH, et al. (2014) Intra-individual double Micronutrient status among pregnant women in zinder, niger
burden of overweight and micronutrient deficiencies among and risk factors associated with deficiency. Nutrients
Vietnamese women. PLoS One 9, e110499. 9, 430.
13. Brindle E, Lillis L, Barney R, et al. (2021) Amulticenter analytical 31. Bock JL (2005) Handbook of clinical laboratory testing during
performance evaluation of a multiplexed immunoarray for the pregnancy. Ann M. Gronowski, editor. Totowa, NJ: Humana
simultaneous measurement of biomarkers of micronutrient Press, 2004, 474 pp. ISBN 1–58829–270–3. Clin Chem 51,
deficiency, inflammation and malarial antigenemia. PLoS One 1321–1322.
16, e0259509. 32. Pfeiffer CM & Looker AC (2017) Laboratory methodologies for
14. Symington EA, Baumgartner J, Malan L, et al. (2018) Nutrition indicators of iron status: strengths, limitations, and analytical
during pregnancy and early development (NuPED) in urban challenges. Am J Clin Nutr 106, 1606s–1614s.
https://doi.org/10.1017/S0007114523001782 Published online by Cambridge University Press