Setting
The study was carried out in Ha Nam province, located in North Vietnam. Ha Nam has a population of approximately 799,400 people as of 2014, with most residents still working in subsistence agriculture. No antenatal or child iron supplementation programs were in place at the time of the study.
Study Design and participants
A cohort study of infants born to women who had previously participated in a cluster randomized controlled trial (c-RCT) of antenatal micronutrient supplementation from enrolment (mean 12 weeks until 3 months postpartum) [5]. Women enrolled in the original c-RCT and their infants (ACTNR 12,610,000,944,033) were followed until 6 m post-partum in the original c-RCT (between September 2010 and Jan 2012) [5]. Women were randomised at the cluster (comune level) and received either (1) one tablet of iron-folic acid (IFA) taken daily (60 mg elemental iron /0·4 mg folic acid per tablet, 7 tablets per week); or (2) one capsule of IFA taken twice a week (60 mg elemental iron /1·5 mg folic acid per capsule; 2 capsules per week); or (3) one capsule of multiple micronutrients (MMN) taken twice a week (60 mg elemental iron, 1·5 mg folic acid, plus other micronutrients) [6]. The primary aim of the c-RCT was to compare the effect of twice weekly provision of antenatal IFA supplementation (either alone or in combination with other micronutrients) with daily provision of IFA supplementation, on maternal and infant outcomes during the first 6 months of life. Detailed information on the methodology used, including a table describing the composition of the supplements, has been previously published [5].
The cohort study followed the c-RCT described above. All 1171 infants live-born to women who were enrolled in the c-RCT and still alive at 6 m of age were eligible for enrolment in the cohort study. Children in the cohort study were followed from 6 months until 36 m of age (between May 2012 and May 2014). Written informed consent was obtained from carers of the children. The study protocol was approved by the Melbourne Health Human Research Ethics Committee and the Ha Nam Provincial Human Research Ethics Committee. Children were seen at the commune health station at 6, 12, 18, 24, 30 and 3 m of age.
Study measurements
Anthropometric measurements
Infant height/length (cm) was measured at 6-monthly intervals between 6 and 36 m of age using a portable Shorr Board (Shorr Productions) during a study-related visit to the clinic. Research staff recorded triplicate measurements of anthropometric measures, a second observer checked all measurements, and the median measurement was obtained. HAZ scores were calculated based on the child’s age and median measurement of height, using WHO anthro [7]. For participants less than 24 months recumbent length was measured, and for participants 24 months and over standing height was measured. Mild, moderate, severe stunting was defined as HAZ scores less than one, two, and three standard deviations below WHO growth standards, respectively.
Iron measurements
One ml of venous blood was collected for serum ferritin concentration at 6m of age during a study-related visit to the clinic. Samples were frozen at –20 °C and transported to the Alfred Hospital, Melbourne, Australia for testing. Serum ferritin (ug/L) was analysed using a chemiluminescent microparticle assay (Architect ci16200; Abbott Diagnostics). Infant iron deficiency was defined as serum ferritin less than 12ug/L [8]. Infant haemoglobin (g/L) was measured using HemoCue (HemoCue AB, Angelholm, Sweden) during the same study-related visits. Infant anaemia was defined as haemoglobin less than 110 g/L [9].
Demographic measurements
Maternal sociodemographic factors were assessed using a standardized questionnaire administered by trained research staff at enrolment as part of c-RCT. The questionnaire included information on demographics, maternal age (years), maternal occupation, education (primary school, secondary school, university/college), and pregnancy history.
Statistical methods
The sample size was determined by the available data on the infants of mothers participating in the c-RCT. The analysis sample included all children with at least one valid HAZ score between 6 and 36 m of age. HAZ scores were considered valid if the data point was collected within ± 2·5m of the planned 6-monthly visit; data outside this range were discarded. Categorical data are presented as percentages with frequency, and continuous data as mean and standard deviation (SD) or median and 25th–75th percentile {Q1-Q3}, where appropriate.
Latent growth curve (LGC) modelling, which uses observed and latent variables, was used. Firstly, the trajectory of growth in HAZ scores over 6–36m was modelled unconditionally (without covariates) to identify its shape. The variance of the residual errors were initially independent and constrained to be equal over time. These assumptions of uncorrelated and homogeneous error variances over time were explored. The underlying assumption of multivariate normality of HAZ scores was justified by definition with natural reference being zero representing the average value of the reference population [7]. Secondly, an unadjusted conditional LGC model was used to evaluate the (linear) association of infant ferritin level at 6m (log base 2 transformed) and the latent growth parameters based on HAZ scores over 6–36m. Maternal iron intervention and clinically identified covariates (i.e., infant sex [female vs. male], maternal education [secondary school vs. primary school; university/college vs. primary school], birth weight (kg) [continuous], and child anaemia [yes vs no]) were added to obtain adjusted results.
The full information maximum likelihood (FIML) was chosen to obtain estimates of the model parameters and standard errors. The FIML handles missing data by the model and assumes missing data is at most missing at random. Nested models were compared using the likelihood ratio test (χ2) while non-nested models used the Akaike Information Criterion (AIC). The fit of the models was measured using the comparative fit index (CFI) and Tucker-Lewis Index (TLI) comparing the fitted model with a baseline model and root mean squared error of approximation (RMSEA) that penalises for excessive complexity. To determine goodness of the model fit, the following three fit indices were used: a CFI close to one (good fit > 0·95; adequate fit > 0·90), TLI close to one (> 0·95) and RMSEA (good fit < 0·05, adequate fit [0·05, 0·08], poor fit > 0·1) [10]. Data were analysed using Stata/IC Version 14 for Windows (StataCorp, 2015, College Station, TX, USA).