Intimate partner violence and growth outcomes through infancy: A longitudinal investigation of multiple mediators in a South African birth cohort

Intimate partner violence (IPV) has been linked to poor fetal and infant growth. However, factors underlying this relationship are not well understood, particularly in the postnatal time period. In a South African cohort, we investigated (1) associations between IPV in pregnancy and growth at birth as well as postnatal IPV and child growth at 12 months and (2) whether maternal depression, tobacco or alcohol use or infant hospitalizations mediated IPV-growth relationships. Mothers were enrolled in pregnancy. Maternal IPV was measured during pregnancy and 10 weeks postpartum; depression, alcohol and tobacco use were measured during pregnancy and at 6 months postpartum. Child weight and length were measured at birth and 12 months and converted to z-scores for analysis. Linear regression and structural equation models investigated interrelationships between IPV and potential mediators of IPV-growth relationships. At birth, among 1,111 mother–infant pairs, maternal emotional and physical IPV were associated with reduced weight-for-age z-scores (WFAZ). Only physical IPV was associated with length-for-age z-scores (LFAZ) at birth. Antenatal maternal alcohol and tobacco use mediated IPV-growth relationships at birth. Postnatally, among 783 mother–infant pairs, emotional and physical IPV were associated with reduced WFAZ at 12 months. Only emotional IPV was associated with LFAZ at 12 months. Maternal tobacco use was a mediator postnatally. Findings highlight the role of physical and emotional IPV as risk factors for compromised fetal and infant growth. Findings underscore the importance of programmes to address interrelated risk factors for compromised infant growth, specifically IPV and substance use, which are prevalent in high-risk settings. This study uses data from a birth cohort investigating the early‐life determinants of child health in a peri‐urban area in South Africa (Stein et al., 2015). The parent study collects comprehensive, longitudinal measures of key risk factors across a variety of disciplines (e.g., environmental, infectious, nutritional, genetic, psychosocial, maternal and immunological) that may impact child health. The DCHS is located 60 km outside Cape Town, South Africa with a population of approximately 200,000. The area is characterized by low socio‐economic status, low educational attainment and a high proportion of female‐headed households (Stats SA, 2021). Results from the cohort study have also shown high rates of interpersonal violence, substance use and child malnutrition (Barnett et al., 2018; Budree et al., 2017; Myers et al., 2018). This is despite a well‐established, free primary health care system, where more than 90% of the population access antenatal or child health services (Stein et al., 2015; Zar et al., 2015). Support for psychosocial issues and mental health disorders in LMIC settings such as South Africa is limited, with some studies estimating a 90% treatment gap (Demyttenaere et al., 2004). Thus, the psychosocial risk profile of this community may be an important driver of child outcomes such as growth. Pregnant women were recruited from two public primary health care clinics, Mbekweni (serving a Black African community) and TC Newman (serving a mixed‐ancestry community). Mothers were enrolled in their second trimester while attending routine antenatal care and have completed assessments antenatally and following birth (ongoing). Women were eligible for the study if they were 18 years or older, between 20 and 28 weeks of gestation, planned attendance at the two recruitment clinics and intended to remain in the area. Data included in the current study were collected antenatally at 28–32 weeks of gestation, at birth, and postnatally at 10 weeks, 6 and 12 months, Figure S1. Between March 2012 and March 2015, 1,225 pregnant women were enrolled into the DCHS, as has been described (Zar et al., 2015). Details of study attendance and loss‐to‐follow‐up are provided in Figure S2. A total of 1,111 children had growth outcome data at birth and were included in the antenatal analysis. A total of 783 children who had growth data at 12 months were included in the postnatal analyses, Figure S2. Infant birth length and weight measurements were conducted by trained labour ward staff, with a subgroup of measurements checked by study staff to confirm reliability. Postnatal anthropometric measurements were done by trained study staff at 12 months. Comprehensive quality control measures were in place including regular training and assessment of staff, routine calibration of equipment and taking multiple measurements. Measurements were performed twice in each child to ensure accuracy. Infant’s weight was measured (to the nearest 10 g) in light or no clothing using a Tanita digital platform scale (TAN1584; IL, USA). Recumbent length was measured using a Seca length‐measuring mat (Seca, Hamburg, Germany), performed on a firm surface by two staff members. Equipment was checked and calibrated weekly (Budree et al., 2017). Birth weights and lengths were converted to z‐scores based on gender and gestational age using the INTERGROWTH‐21st standards (Villar et al., 2014). Postnatal weight‐for‐age z‐scores (WFAZ) and length‐for‐age z‐scores (LFAZ) were calculated using weight and length measurements at 12 months, based on age and gender using Anthro software (World Health Organization [WHO], 2006). The Intimate Partner Violence Questionnaire (IPVQ) is a 12‐item inventory adapted from the WHO multicountry study (Jewkes, 2002) and the Women’s Health Study in Zimbabwe (Shamu et al., 2011). The IPVQ has shown high internal consistency and reliability in similar settings and has been widely used in South Africa (Schraiber et al., 2010; Shamu et al., 2016). Further, unpublished data from our cohort have shown a Cronbach’s alpha of 0.91, indicating relatively high internal consistency. The IPVQ assessed recent (past‐year) exposure to emotional, physical and sexual abuse. Mothers reported frequency of exposure to partner behaviour (‘never’, ‘once’, ‘a few times’ or ‘many times’). Items were summed to create a total score for each IPV subtype, with higher scores indicating higher frequency and severity of IPV (subtype scores ranged from 3 to 20). Above threshold IPV was defined as more than an isolated event within each subtype, specifically where mothers reported multiple responses of ‘once’ or at least one response of ‘a few times’ or ‘many times’ for each IPV subtype. Mothers completed the IPVQ at the 28‐ to 32‐week antenatal visit and at 10‐week postpartum. Socio‐demographic variables were collected from a shortened questionnaire used in the South African Stress and Health (SASH) study, a large population‐based study (Zar et al., 2015) and specifically developed for use in a South African setting. Household income and maternal education (any secondary versus completed secondary) were self‐reported antenatally at 28–32 weeks of gestation. Maternal height was measured at enrolment, using a wall‐mounted stadiometer (CE stature meter). Maternal HIV diagnosis was established at enrolment through maternal self‐report and confirmed during routine HIV testing of pregnant women per the Western Cape PMTCT guidelines. Number of hospitalizations included all‐cause child admissions to Paarl Hospital, the only hospital serving the study catchment area. Active surveillance was conducted by study staff at Paarl Hospital. In addition, at routine study visits, mothers were asked whether children had been hospitalized. Where admissions were reported that were missed by surveillance efforts, study staff abstracted relevant details from hospital folders. A score for total number of hospitalizations from 10 weeks through 12 months of child age was calculated for the postnatal analyses. Validated questionnaires were administered to mothers antenatally and at 6‐month postpartum to assess maternal substance use and depression (Stein et al., 2015). The Alcohol, Smoking and Substance Involvement Screening test (ASSIST), a tool that was developed by the WHO to detect and manage substance use among people attending primary health care services, assessed maternal alcohol and tobacco use risk. It has shown good reliability and validity in international multi‐site studies (Humeniuk et al., 2008) as well as in South Africa (Cronbach’s alpha of 0.81 and 0.95 for alcohol and illicit drugs respectively, van der Westhuizen et al., 2016). Previously published results found that self‐reported tobacco use on the ASSIST correlated well with urine cotinine measures, a biomarker of tobacco smoke (Vanker et al., 2017). Individual item responses were summed to generate total scores, with a higher score indicative of greater risk for substance‐related health problems. Scores of 0–10 for alcohol and 0–3 for tobacco have been used to indicate that a participant is at low risk for substance‐related health problems from their current pattern of use; scores of >10 for alcohol and >3 for tobacco indicate moderate or high risk (WHO, 2010). The Edinburgh Postnatal Depression Rating Scale (EPDS) is a 10‐item self‐report measure of recent depressive symptoms (Cox et al., 1987). It has been validated for use in South Africa (de Bruin et al., 2004) and shown high internal consistency (Cronbach’s alpha = 0.89). Each item is scored on a frequency scale ranging from 0 to 3, with higher total scores indicative of more severe depressive symptoms. Within a total range possible range of 0 to 30, a cut‐off score of ≥13 was used to indicate probable depression (Cox et al., 1987). Categorical variables were summarized using frequencies and percentages, while continuous variables were summarized using median (interquartile range [IQR]), where not normally distributed. Normality of data was assessed using Shapiro–Wilkes. Mann–Whitney rank sum and Kruskal–Wallis tests were used to test for associations between categorical and continuous variables. Pearson chi‐square test was used to determine if significant associations existed between categorical variables. To investigate potential mediators of IPV‐growth relationships in this study, we used an iterative approach. Specifically, we used linear regression to identify significant relationships within the proposed models (i.e., Path a, Path b and Path c) prior to formally testing mediation using structural equation models (SEMs). To do this, the following steps were taken. First, we explored bivariate relationships between IPV subtypes and growth outcomes (i.e., total effect; Path c), specifically between antenatal IPV and WFAZ and LFAZ at birth as well as between postnatal IPV and WFAZ and LFAZ at 12 months, and further analyses were only done where these relationships were significant (p < 0.05). Second, we investigated bivariate associations between IPV subtypes and each hypothesized mediator (i.e., Path a). Third, we ran adjusted linear regression models, adjusting for covariates and potential mediators to identify mediators that were significantly (p < 0.05) associated with growth outcomes considered (i.e., Path b). These relationships were then used to inform criteria, as laid out below, for running formal mediation analyses using SEM. IPV subtypes were included separately in adjusted models due to collinearity. Adjusted linear regression models considered key factors known to affect growth, namely, maternal education, household income, maternal height, maternal HIV status and child sex. Additionally, WFAZ at birth was considered in models investigating growth at 12 months. Covariates were included based on strength of association (p value < 0.05) with the growth outcome explored. Final models were estimated using the maximum likelihood method. Variance inflation factor (VIF) was used to check for multicollinearity. These analyses were run using STATA 15.0. As a final step, multiple mediation analyses were conducted using a SEM approach, which allowed for the estimation of direct and indirect effects, Figure 1. SEMs were used to investigate simultaneously maternal depression, alcohol and/or tobacco use as mediators in the IPV‐growth relationship at birth and at 12 months. Additionally, number of hospitalizations was considered as a potential mediator in IPV‐growth relationships at 12 months. Mediators were tested concurrently (VanderWeele & Vansteelandt, 2014) in final mediation models (1) where IPV sub‐type and the hypothesized mediator were associated (Path a) and (2) where the hypothesized mediator was associated with growth outcomes (p < 0.05) in adjusted models (Path b). Hypothesized mediators that did not meet these criteria but were associated with growth outcomes were included as covariates in final models. SEMs were used to estimate the indirect effect of IPV subtypes on growth outcomes via proposed mediators as well as the direct effect of IPV on growth outcomes (Path c′), Figure 1. SEMs were estimated using the maximum likelihood method to impute missing values. All mother–child pairs with anthropometry data at birth or 12 months were included in models of growth at birth and 12 months, respectively. These were conducted using R version 3.6.1 (R Core Team, 2019) and the library ‘lavaan’ version 3.5.3 (Rosseel, 2012). Confidence intervals (95%) as well as direct effects and casual mediation effects were calculated. Model fit was evaluated using root mean square error of approximation (RMSEA; acceptable fit < 0.08) and comparative fit index (CFI > 0.90). Each analysis was based on 5,000 bootstrapped samples (generated from the ‘lavaan’ package version 3.5.3 in R). Proportion mediated was calculated as the indirect effect/total effect and was done using estimations from the SEM models. Diagram of hypothesized pathways in the association between intimate partner violence and growth at birth or 12 months The DCHS was approved by the Human Research Ethics Committee at the University of Cape Town (401/2009) and by the Western Cape Provincial Health Research committee. Mothers completed informed consent in their preferred language: isiXhosa, Afrikaans or English. Study staff were trained on the ethical conduct of violence research, including confidentiality, mandatory reporting and safety issues. Where substance abuse or mental health issues were identified, staff referred participants to social services or appropriate care in the Paarl area.