Study setting and population
The present study made use of samples that were collected at baseline in a randomised placebo-controlled trial to show non-inferiority of home fortification with 3 mg iron as NaFeEDTA compared with 12.5 mg iron as encapsulated ferrous fumarate. The main results of this trial will be reported elsewhere [5]. The study was conducted in children aged 12–36 months from January–December 2014 in Kisumu-West District, Kenya, an area that is located at around 1350 m above sea level. To recruit the children, community health workers compiled a list of parents with children within the eligible age range in a predefined study area and invited parents to bring all of them for screening to the research clinic, where parents were asked to sign an informed consent form.
Collection of data and samples
We determined weight and height using Salter Scale (UNICEF, catalogue 0145555, Copenhagen, Denmark) and height/recumbent length boards (UNICEF, catalogue 0114500, Copenhagen, Denmark) within 100 g and 1 mm, respectively. Phlebotomists collected 4 mL venous blood in tubes containing Li-heparin. An aliquot of blood was centrifuged, plasma was transferred to a microtube, centrifuged, and stored immediately in liquid nitrogen (−196 °C). The erythrocyte sediment was washed and centrifuged three times with isotonic phosphate-buffered saline. We assessed haemoglobin concentration (HemoCue 301, Ängelholm, Sweden) in duplicate, and zinc protoporphyrin: haem ratio (AVIV haematofluorometer, model 206D, Lakewood NJ, USA) in whole blood and in erythrocytes, each in triplicate. For quality control of the haematofluorometer, we used erythrocyte controls for low, medium and high ZPP values from the manufacturer (Aviv) and as per manufacturer’s instructions. Measurements were within the acceptable range throughout the study.
ZPP in washed erythrocytes is considered a more valid measure of iron-deficient erythropoiesis when compared to ZPP in whole blood because the washing process removes substances dissolved in plasma such as bilirubin and riboflavin that can fluoresce at a wavelength similar to that of ZPP.
We used two rapid diagnostic tests to detect Plasmodium antigenaemia. CareStart G0151 (AccessBio, USA; http://www.accessbio.net/) can detect lactate hydrogenase (pLDH) produced by either P. falciparum or Plasmodium species other than P. falciparum (i.e. P. ovale, P. malariae or P. vivax). CareStart G0171 can detect histidine-rich protein-2 (HRP2), which is produced exclusively by P. falciparum.
Plasma iron indicators (concentrations of ferritin and soluble transferrin receptor), inflammation indicators (concentrations of C-reactive protein and α-1-acid glycoprotein) and other nutritional markers (concentrations of albumin and vitamin B12) were measured on an Abbott Architect C16000 and i2000 SR analyser at Meander Medical Centre, Amersfoort, The Netherlands, with reagents from and as per instructions of the manufacturer.
Eligibility criteria
After data and blood samples were collected but before randomisation to intervention, children were given pre-medication (3-day courses of dihydroartemisinin-piperaquine and albendazole; a single dose of praziquantel). Children were eligible for enrolment in the trial and the present study if: aged 12–36 months, resident in the study area; parental consent form signed by both parents; not acutely sick or febrile (axillary temperature ≥ 37.5 °C) at the time of recruitment; absence of reported or suspected major systemic disorder (e.g. HIV infection, sickle cell disease); no use of antiretroviral drugs against HIV, rifampicin, carbamazepine, phenytoin or phenobarbital and no twin sibling. Children were excluded if: haemoglobin concentration < 70 g/L (for ethical reasons, because the trial had a placebo arm); severely wasted (weight-for-height z-score < −3 SD); known allergy to dihydroartemisinin-piperaquine, benzimidazole or praziquantel; parent-reported history of using antihelminthic drugs in the 1-month period before the screening date; not at risk of malaria (e.g. children who received chemoprophylaxis against malaria because of HIV infection or sickle cell disease); they received their first dose of dihydroartemisinin-piperaquine at the research clinic and did not complete the prescribed second and third doses at home.
Sample size determination
We included all children enrolled in the trial in the present study. Our sample size calculations were based on our primary aim to show non-inferiority of the haemoglobin concentration response to home fortification with 3 mg iron as NaFeEDTA compared with 12.5 mg iron as ferrous fumarate intention. Because this aim is irrelevant to the present study, these calculations are not reported here, although they are available elsewhere [5].
Statistical analysis
Definitions
Anthropometric indices were calculated by comparing measurements with the standards WHO Growth Standards [6]. For whole blood and erythrocyte ZPP, we used a cut-off value of 70 μmol/mol haem (2.7 μg/g haemoglobin), corresponding to the 95% upper limit of the reference values for women and children participating in the US National Health and Nutrition Examination Survey (NHANES) II, from which individuals with anaemia, low transferrin saturation and elevated blood lead concentrations had been excluded [1,2,3]. For erythrocyte ZPP, we also used a cut-off point of 40 μmol/mol haem, which is based on several small studies comparing iron-deficient and iron-replete individuals [7].
We defined iron deficiency as the absence or near-absence of storage iron, indicated by plasma ferritin concentration < 12 μg/L [8]. Because this definition is recommended by WHO to measure population iron status except where inflammation is prevalent [2], we considered it to be valid only in children without inflammation, Plasmodium infection, or HIV infection. In addition, we used the following definitions: anaemia: haemoglobin concentration < 108 g/L (i.e. 110 g/L reference range for children aged 6–59 months, at sea level, minus 2 g/L adjustment for an altitude at 1000–1500 m above sea level) [9]; inflammation: plasma concentrations of C-reactive protein concentration > 5 mg/L [10] and/or α
1-acid glycoprotein concentration > 1 g/L [11]; being stunted or wasted: height-for-age or weight-for-height z-score < −2 SD; [6] P. falciparum infection: presence in blood of HRP2 or P. falciparum-specific pLDH; any Plasmodium infection: presence of HRP2 or pLDH specific to either P. falciparum or human Plasmodium species other than P. falciparum; low vitamin B12 status: plasma vitamin B12 concentration < 150 pmol/L.
Description of the study population
We calculated prevalence values for binary variables, means with corresponding SDs for variables with an approximately normal distribution, and quartiles for continuous variables that were not normally distributed. Because some of the plasma markers used (ferritin, albumin, vitamin B12) can act as acute phase reactants we also described these characteristics in children without inflammation or Plasmodium infection.
Factors associated with ZPP
We explored associations between ZPP and personal characteristics (age, sex), inflammation markers, iron markers, Plasmodium infection and other plasma markers (albumin and vitamin B12 concentrations). Groups were compared assuming t-distributions of ZPP values that were normalised by log-transformation. Exponentiation of results yielded group differences that were expressed as relative differences.
We inspected scatterplots and used simple linear regression analysis to assess associations between ZPP (log-transformed) and explanatory variables with continuous outcomes. Some explanatory factors were untransformed, with the implicit assumption that ZPP values can increase or decrease exponentially with an absolute increment in the explanatory variable; in other cases, log-transformation of the explanatory factor yielded a better model fit, indicating that ZPP values and explanatory variables change at rates that are proportional to their current values. In such cases, variation in the independent variable was expressed as geometric standard deviation, i.e. a dimensionless, multiplicative factor such that dividing or multiplication of the geometric mean by this ratio indicates a variation that is equivalent to subtraction or addition of one standard deviation on a log-transformed scale [12].
We subsequently used multiple linear regression analyses to identify factors that were independently associated with ZPP. Given a linear association between continuous variables, dichotomisation generally results in loss of statistical precision [13]. Thus we preferred to use continuous variables that were shown to be linearly associated with ZPP in the bivariate analyses. Our analysis started with a full model that included haemoglobin concentration, Plasmodium infection, and plasma concentrations of ferritin, soluble transferrin receptor, C-reactive protein, α
1-acid glycoprotein, albumin, vitamin B12, sex (binary) and age class (binary). All plasma markers except albumin and vitamin B12 were log-transformed. Factors were manually eliminated using a backward elimination process with a removal criterion of p > 0.05.
Diagnostic performance of ZPP to detect iron deficiency
This part of the analysis was restricted to children without inflammation (i.e. plasma concentrations of C-reactive protein <5 mg/L and/or α
1-acid glycoprotein <1 g/L) and without Plasmodium infection. We used logistic discriminant analysis to model the probability of iron deficiency as a function of continuous explanatory variables, either alone or combined.
We used the pROC package [14] within R vs. 3.2.0 (www.r-project.org) to produce and analyse receiver operating characteristics (ROC) curves, with comparison of areas-under-the-curve (AUCs) by DeLong’s test for paired curves. Partial AUCs were computed with a correction to achieve a maximal value of 1.0 and a non-discriminant value of 0.5, whatever the range of specificity or sensitivity values. Confidence intervals of estimates for partial areas-under-the-curve (pAUCs) were computed by stratified bootstrapping with 10,000 replicates.
Diagnostic utility of ZPP to estimate prevalence of iron deficiency
First, we assessed the diagnostic performance of ZPP to estimate the prevalence of iron deficiency, using two commonly used cut-off points for ZPP, namely whole blood ZPP >70 μmol/mol haem and erythrocyte ZPP >40 μmol/mol haem (see preceding paragraphs). We used Wilson’s method to calculate confidence intervals around proportions [15, 16]; for sensitivity and specificity, and for the pair of predictive values, we calculated 97.5% univariate CIs. The cross-product of these univariate CIs, considered together, form a joint 95% confidence region for both population parameters [17].
Second, as an example, we used our data to explore the utility of using the combination of haemoglobin concentration and ZPP to screen for iron deficiency (Additional file 1), with cut-points chosen to ensure a high sensitivity so that most cases are detected, at the cost of false positives that could be eliminated by further diagnostic tests. Children can be excluded from further testing if negative test results correctly identify children without iron deficiency in the vast majority of cases. Such a strategy may be desirable in community-based surveys with relatively low prevalence of iron deficiency, but also in medical practice with higher prevalence values (because of self-selection). Thus we estimated the proportion of children who could be eliminated from further testing in settings with a prevalence range for iron deficiency of 0%–50%, which probably covers the vast majority of community settings, with arbitrarily selected sensitivity values and negative predictive values of >90%. Because of its ease of measurement, we limited this assessment with ZPP being measured in whole blood.