Study design and setting
The DCHS is an observational population-based birth cohort located in the peri-urban Drakenstein district of Cape Town, South Africa [32, 33]. This community is characterised by low socio-economic status (SES) and multiple health and psychosocial adversities with a high prevalence of risk factors such as maternal HIV, food insecurity, and malnutrition. However, more than 90% of the population have access to public health services, with TC Newman and Mbekweni Clinics being the two primary healthcare centres for this study. Pregnant women were recruited for the DCHS while attending antenatal clinic visits, and well-characterised mother-child dyads have been followed prospectively.
Participants
Between 2012 and 2015, 1125 pregnant women were enrolled in the DCHS and 1143 live births were included with good retention in postnatal follow-up care. In a nested neuroimaging study [34], a sub-group of children were invited for MRI as neonates (2012-2016), at 2-3 years (2015-2018), and at 6-7 years (2018-2022). Exclusion criteria included (1) medical comorbidity (genetic syndrome, neurological disorder, or congenital abnormality); (2) gestation less than 36 weeks; (3) low Apgar score (less than 7 at 5 minutes); (4) neonatal intensive care admission; (5) maternal use of illicit drugs; (6) MRI contraindications; and (7) child HIV infection. Of the eligible children, those who were scanned at birth were followed-up at subsequent imaging sessions alongside additional children who were selected at the 2-3 year timepoint based on known neurodevelopmental risk factors (maternal HIV and alcohol exposure) and a randomly selected control group. Previous DCHS research has demonstrated comparability between the nested neuroimaging sub-study sample and the full cohort [34].
Findings regarding the association between antenatal maternal anaemia and child brain structure at 2-3 years of age in this cohort have been published [4]. The current study is focused on the 6-7 year timepoint, at which 157 children had both useable structural neuroimaging data (see neuroimaging measures section below) and maternal haemoglobin data for inclusion in analysis. A study flowchart is available as supplementary information (Additional File 1, Figure S1).
Measures
Contextual measures. Demographic, environmental, psychosocial, and clinical data for mother-child dyads were collected antenatally and postnatally for descriptive purposes and for consideration as covariates. All mothers underwent repeat testing for HIV during pregnancy and infants were tested postnatally as per national guidelines. Birth details were obtained by study staff at delivery and child gestational age was calculated using ultrasonography in the 2nd trimester of pregnancy or, where this was unavailable, symphysis-fundal height or maternal report of the most recent menstrual cycle. Child anthropometry measures were observed and classified according to WHO guidelines [35] at routine study visits as well as neuroimaging sessions. Maternal tobacco smoking in pregnancy was self-reported and a dichotomous classification of antenatal alcohol use was retrospectively assessed using the Alcohol, Smoking, and Substance Involvement Screening Test (ASSIST) [36].
Anaemia and iron deficiency. Antenatal maternal and postnatal child anaemia was assessed based on serum haemoglobin measurements in pregnancy and childhood, respectively. Maternal haemoglobin measurements were acquired using rapid tests at antenatal clinic visits as per standard-of-care policy, and iron and folic acid supplementation was recommended as per national guidelines. This data was abstracted from clinical records by trained DCHS staff at study enrolment. Based on WHO guidelines [37], haemoglobin levels of <11g/dL during pregnancy were used to classify pregnant women as anaemic. Further classifications into mild (haemoglobin 10.0 – 10.9g/dL), moderate (haemoglobin 7.0 – 9.9g/dL), and severe anaemia (haemoglobin <7.0g/dL) were determined. Child haemoglobin was only available for children who presented at hospital with pneumonia between birth and the MRI scan, as part of a full blood count. Child anaemia was classified based on age-specific cut-offs using WHO guidelines for all measurements in children aged over 6 months and local guidelines (Groote Schuur Hospital/University of Cape Town Pathology Laboratory) for children under 6 months (Additional File 1, Table S1). A dichotomous variable for child anaemia status was created based on meeting anaemia criteria at least once between birth and 6-7 years of age.
Neuroimaging outcomes. Based on earlier findings [4], structural MRI was chosen as the most relevant imaging measure. Brain volume was acquired on a 3T Siemens Skyra MRI system at the Cape Universities Body Imaging Centre (CUBIC) using a 32 channel head-coil. All children were scanned awake while watching a movie inside the MRI machine. The neuroimaging protocol and MRI specifications are available as supplementary material (Additional File 1, Text S1).
All structural brain scans were processed with FreeSurfer Version 7.1.1 using an automated process of cortical reconstruction and volumetric segmentation [38, 39]. Regional brain volumes were extracted for analysis using the Desikan-Killiany atlas [40] and an inbuilt probabilistic atlas [38] for cortical parcellation and subcortical segmentation, respectively [40]. Given that maternal anaemia was associated with smaller caudate nucleus, putamen, and corpus callosum volumes in the DCHS analysis at 2-3 years of age [4], these brain regions were chosen as key apriori regions of interest (ROIs) for a targeted analysis at 6-7 years of age. However, based on broader literature [24-26, 28], other potentially vulnerable subcortical regions including the hippocampus, amygdala, thalamus, nucleus accumbens, and pallidum were identified for exploratory analyses. All subcortical structures were segmented into left and right hemispheres. The corpus callosum was segmented into posterior, mid-posterior, central, mid-anterior, and anterior regions. The total corpus callosum volume was computed by summing all individual subregions, and the body was defined as the sum of the mid-posterior, central, and mid-anterior volumes. Intracranial volume (ICV) was included as a covariate in analyses to account for normal interindividual variability in brain size.
All regional segmented output (n=204) was subject to a standardised quality control check using the ENIGMA Cortical Quality Control Protocol 2.0 [41]. This was conducted independently by two senior research staff with experience in neuroimaging processing and analysis. Subjects with consistent failures across all brain regions on internal and external quality control (QC) were excluded (n=36). Further decisions on inclusion in the dataset were made by identifying participants that emerged as statistical outliers in SPSS (using Tukey’s method) [42] for subcortical (n=0) and corpus callosum (n=1) ROIs. Overall, a sample of 167 participants passed the visual and statistical QC testing, of which 157 had maternal haemoglobin data (Additional File 1, Figure S1).
Statistical analysis
Sample characteristics. Demographic data and clinical characteristics were presented as means and standard deviations for continuous variables and frequencies for categorical variables. Sociodemographic and clinical (e.g., maternal exposures) group differences between children with antenatal maternal anaemia exposure and children without antenatal maternal anaemia exposure were calculated using unpaired t-tests for continuous data and chi-squared tests or Fisher’s exact tests for categorical data.
Maternal anaemia status. The exposure variable was antenatal maternal anaemia status (dichotomized as anaemic versus non-anaemic based on WHO haemoglobin cut-offs for pregnancy) and the outcomes were regional child brain volumes selected apriori. Between-group differences were investigated using multivariate analysis of variance (MANOVA) general linear models. Given that this analysis aimed to determine whether previously identified findings in 2-3 year-old children persist with age at 6-7 years in the same cohort, a similar statistical approach was conducted. This included a targeted analysis for key apriori ROIs, namely the corpus callosum, putamen, and caudate nucleus. However, a separate MANOVA for an exploratory analysis of other subcortical regions was run to ensure no emerging findings were missed.
A separate set of MANOVA models were performed for grey (left and right hemispheric volumes of subcortical structures) and white (individual corpus callosum regions) matter ROIs. Models were built using a hierarchical stepwise approach with 1) an unadjusted model assessing group differences without the inclusion of covariates, 2) a partially adjusted model including age at scan, sex, SES (represented by maternal education and total household income), and ICV as covariates known to affect brain volume apriori [1, 43], and 3) a fully adjusted model including maternal exposures with demonstrated group differences, placing a particular focus on antenatal alcohol exposure which has consistently been associated with smaller corpus callosum volumes in the broader literature [44] and is known to interact with iron metabolism at a physiological level [45, 46]. A series of fully adjusted post-hoc univariate analyses (ANOVAs) were additionally performed for each individual ROI, correcting for multiple comparisons using the False Discovery Rate (FDR) method [47]. Separate ANOVAs were conducted to assess the association between antenatal maternal anaemia status and overall summed volumes for the body and total corpus callosum.
In comparing volumes based on maternal anaemia status, adjusted mean differences were calculated using pairwise comparisons of estimated marginal means based on the fully adjusted MANOVA and ANOVA models. Percentage differences were calculated using the adjusted mean difference relative to the unadjusted mean volume in the control group (no maternal anaemia) for each brain structure.
Maternal haemoglobin concentration. In regions where an association between maternal anaemia status and child brain volumes was observed (p < .05), we explored hierarchical multivariable linear regression models using standardised regression coefficients for continuous maternal haemoglobin concentrations. This allowed us to assess the relationship between maternal anaemia severity and regional child brain volumes.
Child anaemia sub-analysis. To explore the relative role of postnatal child anaemia on regional child brain volumes in a sub-analysis (children with both maternal and child haemoglobin data), child anaemia status was included as an additional covariate for consideration in the established fully adjusted models for maternal anaemia status described above.
Sensitivity analyses and statistical considerations. Sensitivity analyses were conducted to consider the potential role of a broader range of factors, including timing. We adjusted for trimester of pregnancy given the anticipated increase in maternal blood volume and haemoglobin with gestational time [37]. Secondly, other relevant clinical maternal exposures such as HIV and smoking, both of which are prevalent in this community, were adjusted for in the models to account for any unmeasured confounding [48].
All analyses were conducted using SPSS. A two-sided significance level of p<0.05 was used throughout. Collinearity and biological plausibility was considered in the establishment of all models and checks for assumptions including normality of residuals and homogeneity of variance were conducted throughout.