Study design and participants
Data were drawn from the Growing Up in Singapore Towards healthy Outcomes (GUSTO) pregnancy cohort study (www.clinicaltrials.gov, NCT01174875), detailed elsewhere . This study was conducted according to the guidelines laid down in the Declaration of Helsinki. Ethical approval was obtained from the Domain Specific Review Board of Singapore National Healthcare Group (reference D/09/021) and the Centralised Institutional Review Board of SingHealth (reference 2009/280/D).
Pregnant women attending antenatal visits (< 14 weeks’ gestation) in KK Women’s and Children’s Hospital (KKH) and National University Hospital (NUH) were recruited into the GUSTO study between June 2009 and September 2010. KKH and NUH are the two major public maternity units in Singapore. Recruited women were Singapore citizens or permanent residents between 18 and 50 years of age with biparentally homogeneous ethnicity (Chinese, Malay or Indian). Those who conceived naturally were included in this study. Women receiving chemotherapy or psychotropic drugs and those with type 1 diabetes mellitus were excluded. Informed written consent was obtained from all women prior to recruitment.
Recruited women returned to the hospitals at 26–28 weeks’ gestation for a follow-up study visit. Detailed interviews were conducted in the clinics by trained staff. Data on maternal socio-demographics, educational attainment, obstetric history, smoking status, iron-containing supplementation and anaemia history were collected. Women were asked about the highest level of education attained. Number of previous pregnancies and their outcomes were recorded to determine parity, which included all live- and stillbirths occurring at or after 24 weeks’ gestation, to classify women as nulliparous or parous. Positive smoking status was defined as any cigarette smoking in the current pregnancy. Data on iron-containing supplements, including those taken as part of a multivitamin and mineral supplement or prenatal supplement, was recorded if it was taken for more than once a week in the current pregnancy. Women were asked if they had any history of anaemia in previous pregnancies, either antenatally or post-partum. Data on maternal Hb concentration (g/dl) in early pregnancy (14 weeks’ gestation or less) was collected from the hospital medical records. Women were classified as anaemic if their Hb was less than 11 g/dL .
Self-reported pre-pregnancy weight and measured booking weight at the first antenatal clinic visit (≤14 weeks’ gestation) were recorded. Height was measured with a portable stadiometer (Seca 213, Hamburg, Germany) at 26–28 weeks gestation. Body mass index (BMI) was determined using the formula of weight (kg)/ height (m2). Since the early pregnancy BMI obtained at the first clinic visit was strongly correlated with self-reported pre-pregnancy BMI (r = 0.96, p < 0.001), was free from recall bias and had a lower percentage of missing values than pre-pregnancy BMI (5.7% vs. 7.7%), it was used for all study analyses. BMI status was categorized as < 23 versus ≥23 kg/m2 based on cut-off points for Asian populations .
Plasma ferritin and soluble transferrin receptor assessments
At 26–28 weeks’ gestation, maternal fasting blood samples were collected for the measurements of plasma ferritin and soluble transferrin receptor (sTfR). Plasma ferritin (μg/L) was measured using the sandwich enzyme-linked immunosorbent assay (ELISA) method (AssayMax Human Ferritin ELISA, AssayPro, United States) with an intra-assay coefficient of variation (CV) of 2.9%. The kit standard (AssayMax Human Ferritin Standard) was used as a control, which has been calibrated against WHO International Standard. Women were classified as having iron sufficiency, modest iron depletion and severe iron depletion based on plasma ferritin concentrations of ≥30, 15 to < 30 and < 15 μg/L, respectively [3, 5]. Both modest and severe iron depletion were defined as iron deficiency. Since ferritin is an acute phase protein whose concentration can increase markedly during infection and other inflammatory conditions , we quantified the levels of sTfR as an additional biomarker of iron deficiency, since it is considered to be less affected by acute-phase reactants [3, 21]. Elevated sTfR indicates the presence of functional iron deficiency. Plasma sTfR (nmol/L) was measured using an ELISA (Human sTFr ELISA, BioVendor, Czech Republic) with an intra- and inter-assay CV of 10.9 and 4.8%, respectively. Control human serum samples (BioVendor Quality Control) were run in each assay as an internal control.
Descriptive statistics are presented as percentages for categorical variables; means, standard deviations, medians and 25-75th percentiles for continuous variables. Comparisons of demographics and characteristics between women with iron sufficiency, iron depletion and severe iron depletion were performed using Fisher’s exact tests for categorical variables, ANOVA or Kruskal-Wallis tests for continuous variables. Spearman correlation was used to analyse the continuous association between maternal Hb in early pregnancy and plasma ferritin at 26–28 weeks’ gestation.
Ordinal logistic regression with three ordinal levels was performed for multivariable analyses to assess independent risk factors for iron depletion and severe iron depletion. Compared to a series of binary logistic regression or using multinomial logistic regression, the use of an ordinal logistic regression model helps to increase the power by making full use of the structure of an ordinal scale, producing a more stable estimate and summary with a broad interpretation, applicable across multiple dichotomizations of outcome . In determining variables to be included or excluded from the multivariable model, it has been shown that methods using pre-determined p-value criteria in the univariate analysis are inappropriate, and likewise for automated variable selection procedures (e.g. forward, stepwise) . This is because confounding effects and inter-correlations between independent variables are not being considered, which can lead to biased and distorted outcomes . A better way to determine which variables should be included in the multivariable model is by using clinical judgement , as done in other studies identifying risk factors of an outcome [24, 25]. In this analysis, we selected the potential risk factors and built the model based on a literature review [26,27,28], clinical knowledge and by using a directed acyclic graph. In multivariable ordinal logistic regression analysis, we entered the following potential risk factors simultaneously into the model: maternal age (< 25, 25–34 or ≥ 35 years), BMI (< 23 or ≥ 23 kg/m2), ethnicity (Chinese, Malay or Indian), education (below or at university levels), parity (nulliparous or multiparous), smoking status (no or yes), iron-containing supplementation (no or yes) and history of anaemia (no or yes). The fit of model and proportional odds assumption were checked and met. The proportional odds ratio as presented in this study could be viewed as independent from the degree of severity used to classify the iron status and was thus, valid over all cut-points simultaneously.
Missing values for maternal BMI (n = 6), education (n = 13), parity (n = 1), smoking status (n = 2), iron-containing supplementation (n = 97) and history of anaemia (n = 1) were imputed 100 times using multiple imputation analyses by chained equations. The results of the 100 analyses were pooled using Rubin’s rule. Complete-case analysis was performed as a sensitivity analysis (n = 871). All statistical analyses were two-sided with a 5% significance level and were performed using SPSS software, Version 20 (USA).