Background If effective interventions are to be used to address child mortality and malnutrition, then it is important that we understand the different pathways operating within the framework of child health. More attention needs to be given to understanding the contribution of social influences such as intimate partner violence (IPV).
Aim To investigate the relationship between maternal exposure to IPV and child mortality and malnutrition using data from five developing countries.
Methods Population data from Egypt, Honduras, Kenya, Malawi and Rwanda were analysed. Logistic regression analysis was used to generate odds ratios of the associations between several categories of maternal exposure to IPV since the age of 15 and three child outcomes: under-2-year-old (U2) mortality and moderate and severe stunting (<–2 Z-score height-for-age and <–3 Z-score height-for-age) in 6–59-month-old children. Analyses were adjusted for potential confounders, and the role of mediating factors was explored.
Results The prevalence of physical and/or sexual IPV since the age of 15 years ranged from 15.5% (Honduras) to 46.2% (Kenya). For child stunting, prevalence ranged from 25.4% (Egypt) to 58.0% (Malawi) and for U2 mortality from 3.6% (Honduras) to 15.2% (Rwanda). In Kenya, maternal exposure to IPV was associated with higher U2 mortality (adjusted odds ratio (OR)=1.42, 95% CI 1.18 to 1.71) and child stunting (adjusted OR=1.36, 95% CI 1.16 to 1.61). In Malawi and Honduras, marginal associations were observed between IPV and severe stunting and U2 mortality, respectively, with strength of associations varying by type of violence.
Conclusion The relationship between IPV and U2 mortality and stunting in Kenya, Honduras and Malawi suggests that, in these countries, IPV plays a role in child malnutrition and mortality. This contributes to a growing body of evidence that broader public health benefits may be incurred if efforts to address IPV are incorporated into a wider range of maternal and child health programmes; however, the authors highlight the need for more research that can establish temporality, use data collected on the basis of the study's objectives, and further explore the causal framework of this relationship using more advanced statistical analysis.
- Millenium Development Goals
- developing countries
- child nutrition
- child mortality
- child health
- child survival
Statistics from Altmetric.com
- Millenium Development Goals
- developing countries
- child nutrition
- child mortality
- child health
- child survival
Child mortality and malnutrition are key issues highlighted in the Millennium Development Goals, with specific objectives set to halve malnutrition rates and to reduce child mortality by two-thirds by 2015. However, data used to monitor progress suggest that these goals may not be reached, especially in Sub-Saharan Africa and Southern Asia.1 2 More effort is needed if these targets are to be met, including increasing our understanding of factors contributing to the complex conceptual framework of child health. Such factors include diverse public health and economic and social factors, one of which is violence by an intimate partner (IPV). Despite growing recognition of the high prevalence of IPV globally, there is little evidence available explaining the relationship between IPV and child malnutrition and mortality, especially in developing countries, and further investigation is needed.
IPV is a global occurrence and includes physical aggression, sexual coercion, verbal abuse and controlling behaviour.3 A large 10-country cross-sectional study conducted by the World Health Organization (WHO) estimated that site-specific prevalence of women who had ‘ever’ experienced physical or sexual violence from a partner ranged between 15% (Japan) and 71% (Ethiopia).4 The study also found that there was an overlap between sexual and physical IPV in some countries and that usually an event was indicative of a pattern of abuse and not just an isolated incident.
IPV-related health outcomes in women include injury, chronic disease, increased clinic visits, depression, post-traumatic stress disorder, strained interpersonal relationships, increased risk of homicide, suicidal thoughts and infectious diseases including HIV/AIDS.5–11 These, in turn, can impact on household economics, child care practices and child health, both directly through maternal physical impairment or death and indirectly through impairment of care for a child, and contribute to the household.12–15
Research has linked IPV with poor child health outcomes—for example, low birth weight and increased miscarriage during pregnancy—although the exact mechanisms are unclear and may be related to a combination of trauma, stress hormones, lower maternal weight gain, unintentional pregnancy, or a result of confounding by substance abuse.16–20 A review by Bair-Merritt et al21 in 2006 looked at 22 studies mainly from developed countries, and found evidence of under-immunisation and an increase in risk-taking behaviours in children of mothers exposed to IPV. However, they also stated that sparse literature, weak study design, overuse of shelters and small sample sizes left other relationships with, for example, healthcare use, breast feeding and weight gain uncertain.21 A more recent study in Uganda found that maternal lifetime exposure to IPV was associated with acute infant illness.22 A case–control study in Brazil looking at acute wasting in children aged 1–24 months found that the odds of child malnutrition was significantly higher if the mother had been exposed to severe physical violence in the past 12 months.23
Other studies have linked IPV and higher rates of mortality in children. In 2003, Asling-Monemi et al24 found that mothers of children under 5 in Nicaragua who were exposed to physical and/or sexual IPV before, during and/or after their pregnancy had increased odds (odds ratio (OR)=6.3, 95% CI 2.3 to 7.1) of child mortality compared with those who were not. Secondary analysis from longitudinal data collected in Bangladesh only found this relationship in the female offspring of more educated women, and did not find similar patterns of association for poorer women.25
This literature highlights that the link between IPV and child heath outcomes follows a complex pathway, and needs further quantification. This study investigates further the relationship between IPV and child stunting and mortality, using data from five developing countries.
The Demographic and Health Surveys (DHS) use a standardised multistage sampling procedure (mostly two-stage, although Egypt has three-stage) with weighting to reproduce nationally representative data on maternal and child health, in developing countries. Details are available eleswhere.26 In this analysis, the primary sampling unit was used. Data were weighted according to household weight variables (and stratification units specified within the datasets). Some countries presented occasional strata with single sampling units because of observations lost due to missing data or cleaning, and, in such cases, the average of variance from other strata with multiple units was used to calculate the standard errors.
The domestic violence module is a relatively new addition to DHS and is administered to one, randomly selected, woman per household. The DHS datasets are available for public access upon application.26 We selected five low-income countries: Egypt (2005), Honduras (2005), Kenya (2003), Malawi (2004) and Rwanda (2005). The selection was based on the availability of cleaned datasets comprising the variables of interest, adequate sample sizes without large amounts of missing data, use of modified conflict tactics scales (CTS) for measurement of IPV, and adherence to WHO 2001 guidelines for ethical and safe data collection in IPV research.27–34 Information was obtained from the household, domestic violence and child level questionnaires. Weighting was used to reproduce nationally representative samples.
Intimate partner violence
Modified CTS were used to collect data on physical and sexual violence. Sexual violence was not included in the original CTS27 and was incorporated by DHS into the domestic violence module using a similar format. Respondents were asked questions about their experiences of specific acts of physical and sexual violence by a current or former partner. Questions varied slightly by country, but covered acts of physical violence ranging in severity from arm twisting and slapping to strangling and attacking with a knife or gun. Sexual violence was identified if the woman was forced or coerced by her partner into sex or sexual acts. Two exposure variables were defined for this analysis: any physical/sexual violence versus none; and exposure to different categories of violence versus none (defined as exposure to physical violence only, sexual violence only, or physical and sexual violence). The recall period was defined as ‘since the age of 15 years’.
For child mortality, we looked at under-2-year-old (U2) mortality (all deaths up to 2 years). This was measured by observing the survival status of children born using an 8-year observation period before the survey, starting 2 years before the interview date. This period was used to ensure sufficient statistical power while not being too long so as to minimise recall bias. For each respondent, all births within the time frame were included. In order to avoid censoring effects, births in the 2 years preceding the interview were not included. We used under 2 rather than under 5, because survival to age 5 involves a longer censoring period, resulting in older estimates of IPV. Also there is little difference between the two rates as most under-5 mortality occurs in the first 2 years of life.35
This study was specifically concerned with the information collected on height-for-age. Measured in Z-scores, which equate to standard deviations from the median of a reference population (WHO Child Growth Standards),36 those children with a height-for-age Z-score less than –2 are termed ‘stunted’ and those with a Z-score <–3 are termed ‘severely stunted’. Stunting was chosen because it reflects long-term nutritional deficiency or a period of severe faltering with failed catch-up growth, and because it was the most prevalent marker of chronic malnutrition in these datasets. The aetiology of wasting or acute malnutrition is usually dependent on a recent set of events such as illness or food shortage. Therefore the use of stunting or chronic malnutrition, which is often related to a nutritional deficiency over time, was determined to be more appropriate when defining exposure to IPV as one or multiple events since the age of 15 years. Wasting also would be correlated with child mortality, which was already included in the analysis. Anthropometric observations outside the range +6 to −6 Z-scores were excluded.
All DHS studies stated that enumerators, including those collecting anthropometric data, received observed training and testing that included lecture and practical exercise components. Teams also often included personnel with advanced medical training.
A conceptual framework for the proposed link between IPV and the child outcomes was developed, outlining potential confounders and hypothesised pathway variables (figure 1).
Maternal age, maternal education, number of living children, urban/rural residence and household socioeconomic status (as measured by a country-specific wealth index) were considered as a priori confounding variables. Mediating variables were considered at the level of both the individual child and the respondent. Child-level variables were chosen to be proxy for maternal care ability and were defined as whether or not the pregnancy was intended (at the time versus later/never), duration of breast feeding (never–11 months, 12–23 months, ≥24 months/still breast feeding) and BCG vaccination (based on vaccination record or maternal recall). Maternal-level variables were only available for last birth, and were used as a proxy for maternal care-seeking behaviour rather than as direct pathways through which a child-specific event may have occurred. Maternal care-seeking behaviours could be limited when women who are exposed to physical IPV experience depression, loss of control, social isolation and/or injury and are less likely to seek services and more likely to have a child with a negative health outcome. These variables could also provide a protective effect if women who are exposed to IPV but have access to services are less likely to have a child with a negative health outcome.
These included antenatal care (four or more visits), tetanus toxoid coverage (received two or more tetanus toxoid vaccinations before the birth) and skilled delivery (delivered by a doctor, nurse/midwife or auxiliary nurse/midwife). Data on mediating variables were only available for those children born in the 5 years before the survey.
Logistic regression models were fitted separately to data from each country to estimate the crude associations between the two indicators of IPV and the child stunting and mortality. Adjusted models controlled for all of the potential confounders identified in the conceptual framework.
The role of mediating variables was explored using data from the subset of children born 2–5 years before the survey (who therefore had data on pathway variables). Logistic regression models were fitted to the data to model the crude associations between ‘any’ IPV and the child stunting and mortality. Blocks of variables from the conceptual framework were then added into the models in the following combinations: potential confounders, confounders plus maternal pathway variables, confounders plus child-specific pathway variables, confounders plus all pathway variables. This was done to see how the addition of other variables affected the relationship between maternal IPV and child outcomes, and therefore how much of the association could be explained by these postulated pathways. The same models were applied to data from each country in order to provide a means of comparison between the observed relationships in each place.
All analyses were conducted using Stata software V10.0, taking into account the sample frequency weights provided by the DHS to reproduce the national population. The WHO ANTHRO software37 was used to calculate anthropometric indices.
Final sample sizes for which respondent-level and child-level data were available are presented in table 1. All countries except for Rwanda (11.8%) had a non-response rate of lower than 10% for the IPV data. Non-response included non-participation and inability to obtain the necessary privacy for the interview. Missing or flagged nutrition indices accounted for a loss of anthropometric outcome data of between 4.5% (Egypt) and 16.6% (Malawi).
Descriptive data for the respondents included in the mortality analysis are presented in table 2. Variable distributions were similar in the sample used for the stunting analysis, except for respondent's age, which on average was lower for the stunting sample (mean age of 30.4 years for mortality and 28 years for stunting). The prevalence of IPV since the age of 15 years ranged from 15.5% (Honduras) to 46.2% (Kenya). The prevalence of child stunting in the samples ranged from 25.4% (Egypt) to 58.0% (Malawi) and the prevalence of U2 mortality from 3.6% (Honduras) to 15.2% (Rwanda). Although the prevalence of abuse did not show a consistent trend, prevalence of stunting and mortality was higher in the countries with lower Human Development Index scores (Kenya, Malawi and Rwanda).
After adjustment for confounding variables, the strongest relationship between ‘any’ IPV and U2 mortality was observed in Kenya (adjusted odds ratio (OR) 1.42, 95% CI 1.18 to 1.71). Weak associations were seen in Honduras (OR 1.24, 95% CI 0.98 to 1.57) and Malawi (OR 1.12, CI 0.98 to 1.28).
Table 3 shows that, by IPV category, significant relationships with mortality were identified for physical violence in Kenya, for physical and sexual violence in Honduras, and for sexual violence in Malawi.
For moderate child stunting, significant associations with IPV were only observed in Kenya, with adjusted ORs for stunting of 1.40 (95% CI 1.07 to 1.85) if the respondent was exposed to physical and sexual IPV and 1.36 (95% CI 1.16 to 1.61) if exposed to any IPV (table 4). Severe stunting was similar to mortality when looked at by category of IPV (table 5). In Kenya severe stunting showed a stronger association with physical violence, whereas in Honduras and Malawi it was associated more with physical and sexual IPV and only sexual IPV, respectively.
Although the addition of mediating variables did not notably change the estimates of effect of IPV on any of the outcomes in any consistent direction across the country for samples in our analysis, there could be issues with the accuracy of the proxy variables and the timing of events. Although the mediators we explored still might partly explain the observed associations between IPV and child stunting and malnutrition outcomes in some settings, there might be other pathways that we have not captured in this analysis.
This analysis indicates that there may be a relationship between IPV and child mortality and stunting in Honduras, Kenya and Malawi; however, issues such as reverse causality, measurement error and advanced modelling methods would need to be addressed or incorporated to better explain this relationship. When IPV was considered by type, associations remained with both stunting and malnutrition, although there was no consistent trend whereby specific types of violence were more strongly associated than others with the outcomes.
The mediators explored in this analysis may have a role in the relationship between IPV and child malnutrition and mortality in some settings, although they were not evident in these results. There are other mechanisms of association (causal, reverse causal or confounding) that we have not been able to account for. It is plausible that IPV may negatively impact on a mother's mental health and care-seeking behaviour when her child becomes ill/malnourished, and on her more general ability to provide care for her child. This could occur as a result of impaired physical and mental health, perceived lack of empowerment, and physical/psychological barriers to accessing services.5–15 Social isolation may correlate with IPV, which may theoretically affect support networks for a woman. Also, not only may physical IPV directly limit a woman's capabilities, but it can also be a proxy indicator for greater gender inequality in the household, including, for example, a lower control of how household assets are used and lower educational attainment, both of which have been linked to lowered maternal and child health and nutritional status.38 39
The results of this study are in keeping with those from other studies.21–25 A strength of this analysis is the use of comparative data from five diverse countries. The fact that the size and strength of associations differed by country suggests that pathways of effect vary depending on contextual factors.
This study does have several limitations. Data quality in large-scale surveys can vary markedly between sites. This is especially true for data pertaining to domestic violence. DHS has worked to improve the validity of its domestic violence module using widely accepted tools and standardised methodology; however, cultural factors and data collection techniques have been shown to influence rates of disclosure.3 28 40 If under-reporting of violence was non-differential with respect to the outcomes in this study, this would weaken our observed effect estimates.
Missing data were mostly due to anthropometric variables being outside the specified ranges for quality or with an age, weight or height variable missing. This accounted for considerable loss in some datasets (Malawi). As missing anthropometric data were probably related to household refusal or enumerator error, it is not know how this will bias the results of the analysis. Missing IPV also accounted for data loss. Women who experience IPV or those with a sick child may be more or less willing to allow these data to be collected depending on their circumstances. Enumerator error was not controlled for in this study, but should not considerably bias the relationship between the exposure and outcome. Missing IPV data related to exposure to abuse could cause the relationship to be underestimated, as they would have lowered the number of exposed individuals within each dataset.
The data used in this analysis are cross-sectional, and we are therefore unable to establish the temporality of observed associations. In addition to the plausible mechanisms of association identified above, causal pathways in the reverse direction are possible whereby child malnutrition and mortality, and the stresses engendered by them, may affect a woman's risk of experiencing IPV. Because there was no timeline for the occurrence of IPV, abuse during pregnancy was not evaluated. As discussed above, there is some evidence of a relationship between maternal exposure to IPV during pregnancy and low birth weight and infant death. Therefore, if a woman experienced IPV during pregnancy, this may explain some of the associations in this study.
In addition to inhibiting the establishment of temporality, using any exposure since the age of 15 years for a timeframe to determine abuse allowed for issues such as respondent recall bias and did not allow the authors to explore the effect of frequency or duration of abuse.
Although we have controlled for a range of potential confounders, we have not taken into account co-occurring influences, such as other forms of family violence (including child abuse/neglect) or household substance abuse, which may influence child outcomes. There may thus be residual confounding in our observed associations and more complex pathways which we have not explored.
Finally, there is the potential for intra-household clustering of outcomes, as all of a woman's eligible children were included in the analysis. However, a study by Le Thanh and Verma41 found that this did not largely influence the design effect for these child outcomes in DHS, and for this reason we did not adjust for household-level clustering in this analysis.
Despite the above limitations, this analysis supports the evidence that IPV has an influential role in child stunting and mortality in developing countries.21–25 From both a human rights and evidence-based approach to women's and children's health, it is imperative to develop effective responses to the problem of domestic violence. Programming initiatives that address domestic and other forms of gender-based violence in the context of maternal and child health programmes are being introduced.42 However, studies that explore the framework of this relationship in a way that provides a means for temporality would be useful in determining cause and effect of the relationship. Also, research that uses data collected specifically for its objective will allow more advanced statistical analysis, quality control and exploration of the causal pathway. This is necessary if we are to develop policy and programming interventions that more effectively address these inter-related public-health issues.
What is already known on this subject
Links between maternal exposure to intimate partner violence (IPV) and child health outcomes have been noted, although the exact context and mechanisms that support this relationship in developing countries is yet unclear.
What this study adds
The datasets and methodology used in this study highlight some of the relationships between IPV and child stunting and mortality. However, variations between countries, with type of violence and mediating variables, indicate that the strength of these relationships may be closely linked to the social, demographic and cultural context in which IPV occurs.
The relationship between intimate partner violence (IPV) and under-2-year-old (U2) mortality and stunting in Kenya, Honduras and Malawi suggests that, in these countries, IPV plays a role in poor child health outcomes. This study also contributes to a growing body of evidence that broader public health benefits may be incurred if efforts to address IPV are incorporated into a wider range of maternal and child health programmes.
We acknowledge the people who contributed to the collection, analysis and presentation of DHS data.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.