Impact of small-quantity lipid-based nutrient supplement on hemoglobin, iron status and biomarkers of inflammation in pregnant Ghanaian women

We examined hemoglobin (Hb, g/L), iron status (zinc protoporphyrin, ZPP, µmol/mol heme, and transferrin receptor, TfR, mg/L) and inflammation (C-reactive protein, CRP and alpha-1 glycoprotein, AGP) in pregnant Ghanaian women who participated in a randomized controlled trial. Women (n = 1320) received either 60 mg Fe + 400-µg folic acid (IFA); 18 micronutrients including 20-mg Fe (MMN) or small-quantity lipid-based nutrient supplements (SQ-LNS, 118 kcal/d) with the same micronutrient levels as in MMN, plus four additional minerals (LNS) daily during pregnancy. Intention-to-treat analysis included 349, 354 and 354 women in the IFA, MMN and LNS groups, respectively, with overall baseline mean Hb and anemia (Hb <100) prevalence of 112 and 13.3%, respectively. At 36 gestational weeks, overall Hb was 117, and anemia prevalence was 5.3%. Compared with the IFA group, the LNS and MMN groups had lower mean Hb (120 ± 11 vs. 115 ± 12 and 117 ± 12, respectively; P < 0.001), higher mean ZPP (42 ± 30 vs. 50 ± 29 and 49 ± 30; P = 0.010) and TfR (4.0 ± 1.3 vs. 4.9 ± 1.8 and 4.6 ± 1.7; P < 0.001), and greater prevalence of anemia (2.2% vs. 7.9% and 5.8%; P = 0.019), elevated ZPP (>60) [9.4% vs. 18.6% and 19.2%; P = 0.003] and elevated TfR (>6.0) [9.0% vs. 19.2% and 15.1%; P = 0.004]. CRP and AGP concentrations did not differ among groups. We conclude that among pregnant women in a semi-urban setting in Ghana, supplementation with SQ-LNS or MMN containing 20 mg iron resulted in lower Hb and iron status but had no impact on inflammation, when compared with iron (60 mg) plus folic acid (400 µg). The amount of iron in such supplements that is most effective for improving both maternal Hb/iron status and birth outcomes requires further evaluation. This trial was registered at ClinicalTrials.gov as: NCT00970866. The iLiNS‐DYAD study in Ghana was conducted in several adjoining semi‐urban communities in the Yilo Krobo and the Lower Manya Krobo Districts about 70 km north of Accra, Ghana. Details of the study setting, participants, design, randomization and masking schemes, and other key procedures have been reported elsewhere (Adu‐Afarwuah et al. 2015). In brief, the study was designed as a partially double‐blind, parallel, individually randomized, controlled trial with three equal‐size groups. Pregnant women attending usual ante‐natal clinics in four main health facilities in the area between December 2009 and December 2011 completed a screening questionnaire if they were ≥18 years old, ≤20‐week gestation (as determined by the antenatal clinics mostly by fundal height), and had an antenatal card complete with history and examination. Informed consent for the screening was obtained by trained study workers at the antenatal clinics. Following screening, women were excluded if the antenatal card indicated HIV infection, asthma, epilepsy, tuberculosis or any malignancy. Additional exclusion criteria were known milk or peanut allergy, not residing in the area, intention to move within the next 2 years, unwillingness to receive field workers or take study supplement, participation in another trial or gestational age (GA) >20 weeks before completion of the enrolment process. Women who passed the screening were visited in their homes, where details of the study were provided, and those willing to participate were recruited, after signing or thumb‐printing informed consent. Recruited women remaining eligible underwent a baseline laboratory assessment after consent, and were immediately randomized to receive one of three treatments daily: (a) 60 mg iron plus 400‐µg folic acid (hereafter, IFA supplement or group); (b) multiple micronutrient capsule containing 18 vitamins and minerals (including 20 mg iron) (hereafter, MMN supplement or group); and (c) SQ‐LNS with similar micronutrients as the MMN supplement, plus other minerals and macronutrients (hereafter, LNS supplement or group). Group allocations were developed by the Study Statistician at UC Davis using a computer‐generated (SAS version 9.3) randomization scheme (in blocks of nine), and were placed in sealed, opaque envelopes. At each enrolment, a Study Nurse offered nine envelopes at a time, and the woman picked one to reveal the allocation. Allocation information was kept securely by the Field Supervisor and the Study Statistician only. The compositions of the 3 supplements were reported previously (Adu‐Afarwuah et al. 2015), as well as the considerations underlying the concentrations of the nutrients in the MMN and SQ‐LNS (Arimond et al. 2013). Apart from iron which was kept at 20 mg/day in the MMN and SQ‐LNS, the vitamin and mineral contents were either 1x or 2x the RDA for pregnancy, or in a few cases, the maximum amount that could be included in the supplement given technical and organoleptic constraints. The IFA and MMN supplements were provided as capsules in blister packs, and were intended to be consumed with water after a meal, one capsule per day throughout pregnancy. The LNS supplement was in 20‐g sachets, and was intended to be mixed with any prepared food, one sachet per day throughout pregnancy. To maintain blinding, two individuals independent of the study placed color‐coded stickers behind the blister packs (three different colors for IFA and three for MMN supplements) so that the capsules were known to the study team and participants only by the colors of the stickers. Laboratory staff and data analysts had no knowledge of group assignment until all preliminary analyses had been completed and the allocation codes were broken. The study was registered on http://ClinicalTrials.gov (Identifier: {“type”:”clinical-trial”,”attrs”:{“text”:”NCT00970866″,”term_id”:”NCT00970866″}}NCT00970866) and was approved by ethics committees of the University of California, Davis, the Ghana Health Service and the University of Ghana Noguchi Memorial Institute for Medical Research. We collected socio‐demographic information at baseline, and determined GA mostly by ultrasound biometry (Aloka SSD 500, Tokyo, Japan). During follow‐up, field workers visited women in their homes every 2 weeks, whereupon they delivered a fresh supply of supplement and monitored supplement intakes. At each of laboratory assessments at baseline and at 36 GW, women’s weight (Seca 874) and height (Seca 217) were measured, and peripheral malaria parasitemia (Clearview Malarial Combo, Vision Biotech, South Africa), Hb (HemoCue AG, Wetzikon, Switzerland) and zinc protoporphyrin, ZPP (hematofluorometer, Aviv Biomedical Co. NJ, USA), were determined using venous blood (Adu‐Afarwuah et al. 2015). We used the original Aviv cover‐slides and three‐level control material for the ZPP measurements, after red blood cells were washed three times with normal saline. Plasma samples obtained after blood was centrifuged at 1252 ×g for 15 min were stored in Ghana at −20°C, before being air‐freighted on dry ice to UC Davis, where soluble transferrin receptor (TfR, mg/L), CRP (mg/L) and AGP (g/L) concentrations were determined using a Cobas Integra 400 plus Automatic Analyzer (Roche Diagnostic Corp., Indianapolis, IN). At 36 GW, the continuous outcome measures were Hb (g/L), ZPP (µmol/mol heme) and plasma TfR (mg/L), CRP (mg/L) and AGP (g/L) concentrations, while the binary outcome measures were the percentages of women with low Hb, high Hb and elevated ZPP, TfR, CRP and AGP. For the Ghana iLiNS‐DYAD Study, an effect size (Cohen’s d: difference between group means divided by the pooled standard deviation) of 0.3 (considered a small‐to‐moderate effect size) (Cohen, 1988) was the basis for sample size calculation. Thus, our sample size was based on detecting an effect size of 0.3 between any two groups for any continuous variable at 36 GW, with a two‐sided 5% test and 80% power. As described previously (Adu‐Afarwuah et al. 2015), we enrolled 1320 pregnant women into the study, but after excluding 177 who received both IFA and MMN supplements during pregnancy because of a temporary mislabeling of supplements, as well as 86 in the LNS group who were pregnant during the same time period, 1057 women were included in the current analysis. Based on a sample size of 827 women (~275 per group) for whom data were available at 36 GW, we had 94% power to detect an effect size of 0.3 between any two groups for Hb, ZPP or TfR. This would allow a difference of 3.4 g/L in Hb, 8.9 µmol/mol heme in ZPP and 0.5 mg/L in TfR (given SD of 11.0, 30.0 and 2.0, respectively) to be detected between any two groups. We posted the statistical analysis plan (http://www.ilins.org) before analysis. Statistical analysis, by intention‐to‐treat, was performed using SAS for Windows Release 9.3 (Cary, NC, USA). Background socio‐demographic characteristics were summarized as mean ± SD for continuous variables, or number of participants and percentages for categorical variables. As done previously (Adu‐Afarwuah et al. 2015), we used two indices, namely assets index and housing index as proxy indictors for socioeconomic status, and calculated household food insecurity access (HFIA) score (Coates et al. 2007) as a measure of degree of household food insecurity. Higher values of the assets and housing indices represented higher socioeconomic status, and higher values of the food insecurity index represented higher food insecurity. We calculated adherence to treatment as percentage of days from enrolment to the home visit closest to the laboratory assessment at 36 GW, when women reported consuming the supplement. We used Hb <100 g/L as our primary definition for low Hb (representing anemia). This was based on previous WHO (WHO/UNICEF/UNU 2001; WHO 2007) and International Nutritional Anemia Consultative Group, INACG (Nestel & INACG Steering Committee 2002) documents that suggest lowering the standard 110 g/L cut‐off by 10 g/L for pregnant women of African extraction to achieve adequate sensitivity and specificity for screening purposes (WHO/UNICEF/UNU 2001). In addition, we defined low Hb using the standard cut‐off of Hb <110 g/L, based on a recent WHO recommendation (WHO 2011) to maintain that cut‐off (110 g/L) without any adjustment, because of scarce evidence to support the adjustment. A meta‐analysis (Haider et al. 2013) revealed that Hb cut‐offs ranging from <100 g/L to 115 g/L have been used in studies to define anemia in pregnant women. We defined high Hb as >130 g/L (Pena‐Rosas et al. 2012), elevated ZPP (proxy for iron deficiency) as >60 µmol/mol heme (Walsh et al. 2011) and elevated TfR (proxy for tissue iron deficiency) as >6.0 mg/L (Pfeiffer et al. 2007; Vandevijvere et al. 2013). Because there is no generally accepted cut‐off value for TfR, we derived the 6.0 mg/L cut‐off based on the evidence that TfR values obtained using the Automatic Analyzer assay (as used in this study) were on average 30% lower than values obtained with the ELISA assay (Pfeiffer et al. 2007). Therefore, we reduced by 30% the 8.5 mg/L cut‐off value used when TfR was determined using ELISA (Vandevijvere et al. 2013) to obtain the cut‐off of approximately 6.0 mg/L for our analysis. Because we used two cut‐offs to define anemia, we also defined IDA in two ways: first as Hb <100 g/L and at least one marker of iron deficiency (ZPP >60 (µmol/mol heme) or TfR >6.0 mg/L (Pfeiffer et al. 2007; Vandevijvere et al. 2013)), and second, as Hb <110 g/L and at least one marker of iron deficiency. We defined elevated inflammatory markers using the cut‐off values of >5.0 mg/L for CRP and >1.0 g/L for AGP (Thurnham and McCabe, 2012), and categorized women with inflammation as either elevated CRP only (indicative of incubation phase of infection), elevated CRP and AGP (indicative of early convalescence) or elevated AGP only (indicative of late convalescence) (Thurnham and McCabe, 2012). At 36 GW, we calculated overall mean (±SD) values and percentages for Hb and markers of iron status and inflammation. We compared groups by using general linear models (continuous outcomes) and logistic regression models (binary), with Tukey–Kramer adjustment for multiple comparisons. Along with the group comparisons, we calculated pairwise mean differences (continuous outcomes, ANOVA) and relative risks (binary outcomes, Logistic regression) with their 95% CI and P‐values. Relative risks were calculated using Poisson regression (Spiegelman & Hertzmark 2005). In addition, we analyzed changes in the prevalence of anemia, high Hb and elevated ZPP, TfR, CRP and AGP from enrolment using mixed model logistic regression (SAS PROC GLIMMIX). Where the mixed model logistic regression failed to converge because of sparse data, we used generalized estimating equations model (SAS PROC GENMOD). We analyzed each outcome twice, first without any covariate adjustments, and then with adjustment for covariates significantly associated (P < 0.10) with the outcome in a bivariate analysis. Because ZPP, TfR, AGP and CRP are not normally distributed, we calculated the group means (±SD or SE), group percentages and pair‐wise mean differences and relative risks with their 95% CI based on untransformed data, but generated the P‐values for group or pair‐wise comparisons using logarithmically transformed data. To investigate the possible effect of group differences in adherence to treatment, we performed a per‐protocol analysis, which was restricted to women with adherence ≥70%. We evaluated potential interaction of treatment group with pre‐specified baseline variables for maternal characteristics, anemia and iron status. These variables were: age, years of schooling, BMI, gestational age at enrolment, household assets index, housing index, food insecurity access score, season at enrolment (dry or wet), primiparous, anemia and elevated ZPP, TfR, and AGP or CRP. Where an interaction was significant (alpha <0.10), we performed subgroup analysis by including an interaction term between treatment and the effect modifier in the ANCOVA or logistic regression model. For continuous effect modifiers, we used data from all participants to create a linear regression model to predict the values of the outcome at the 10th and 90th percentile of the effect modifier distribution. Each effect modifier was considered separately in the models to avoid collinearity. In a sensitivity analysis aimed at correcting for the effect of inflammation (CRP and AGP) on the Hb and iron status outcomes, we repeated the above analyses using values of Hb and iron status markers corrected for inflammation (WHO 2007). These corrected values were calculated by grouping women into three inflammation categories, estimating the correction factor (CF) for each inflammation category, and multiplying the Hb and iron status values of each woman by the inflammation category‐specific CF (Grant et al. 2012). The three inflammation categories were: reference (normal CRP and AGP), incubation (raised CRP and normal AGP) and early (raised CRP and AGP) or late (normal CRP and raised AGP) convalescence [these two phases of convalescence were combined because of small sample sizes and little indication of differences]. For ZPP at 36 GW, women were grouped into two inflammation categories (normal vs. any inflammation), because the three‐category grouping did not yield consistent results.