Intimate partner violence and exclusive breastfeeding of infants: analysis of the 2013 Nigeria demographic and health survey

This study analyzed data collected from the 2013 Nigeria Demographic and Health Survey (DHS). The information was collected from February to July 2013. Nigeria is Africa’s most populous country with an estimated population of over 200 million people of diverse ethnic and cultural backgrounds [28]. The National Population Commission (NPC) in collaboration with ICF Macro, Calverton, MD, USA conducted the survey. The sampling involved a three-stage cluster design. This consisted of a selection of 904 primary sampling units (PSUs), 372 in the urban, and 532 in the rural. A nationally representative sample of 40,680 households was then selected across the PSUs. Both married men and women in the households were eligible to be interviewed with the corresponding version of the questionnaire designed for males and females separately. About 39,902 women aged 15 to 49 years were identified as eligible of which 98% were successfully interviewed with the women’s version of the questionnaire. Questions were asked relating to household sociodemography, maternal health, as well as child wellbeing. The IPV module was a subsample survey within this general survey and it was based on a shortened and modified version of the Conflict Tactics Scale (CTS) [29]. This modification to the original scale was done between 1998 and 99 by ICF Macro, the organizer of the DHS programs, after consultation with experts on domestic violence measurement, gender, and survey research [30]. It was subsequently tested and validated through pilot studies in Cambodia, Colombia, and Haiti in 2000, and then The Dominican Republic in 2002 [30]. Concurrent validity has also been established for this modified scale, as it has been used for the DHS programs implemented in over 90 countries afterward across Africa, Asia, and Latin America. Additionally, studies utilizing data from those surveys have consistently reported a high Cronbach alpha indicating an internal reliability of construct [31‐34]. The advantage of the DHS program’s modified CTS includes the fact that it incorporates questions on sexual violence alongside physical violence, and also does not assume that violence takes place only in situations characterized by conflict.

During the survey, only one woman per household was selected for the IPV module. Specially constructed weights were used to adjust for this selection pattern to ensure that the IPV subsample was nationally representative. Three specific protections were built into the survey questionnaire under the World Health Organization’s (WHO) ethical and safety recommendations [35]. These include the informed consent of the respondent, privacy during the inquiry, and confidentiality of the information shared. The team of interviewers comprised four females and two males who had been equipped with the necessary training to conduct the IPV module. The questionnaire was originally designed in English, but before its implementation for the survey in Nigeria, it was also translated into the three major Nigerian languages—Hausa, Igbo, and Yoruba, by the NPC, through a stakeholder meeting in March 2012. It was pretested, refined, and finalized for the survey. More information about the survey setting, and data collection is provided in a final report [14].

For our study, sample selection was limited to women who were interviewed in the IPV module, currently resides with her partner, had a child under 6 months of age who also resides with her, and the woman indicated to be currently breastfeeding. A total of 2668 mother-infant dyads met these criteria.

The outcome variable of interest was binary, indicating if a breastfeeding infant under 6 months of age was undergoing EBF (=1) or mixed-feeding (=0). This was determined by questions on whether the child was given certain types of solid or semi-solid food in the prior 24 h to the survey. These were typically foods that were not recommended for infants under 6 months. Only those who responded not to have fed the child with any of the listed food items other than breast milk were regarded to be practicing exclusive breastfeeding.

The exposure variables were forms of IPV measured through 13 questions contained in the DHS program’s modified CTS known as the DHS domestic violence module [30]. It bordered on possible violent events a woman experienced from a current male partner in the last 12 months (Table 1). It was assumed that the time period captured events that may have occurred during pregnancy or postpartum.

These questions were grouped into three forms of IPV. Psychological IPV involved three questions (items 1–3) with a Cronbach’s alpha of 0.73. Physical IPV involved seven questions (items 4–10) with a Cronbach’s alpha of 0.86. While sexual IPV involved three questions (items 11–13) with a Cronbach’s alpha of 0.86. A combined variable was also created that included the experience of any of the forms of intimate partner violence. All four variables were binary coded. Additionally, the frequency of IPV was generated from the 13 possible events. This was used to measure the dose-effect of an IPV experience. The questions in the survey had asked about the frequency of occurrence of each violence. The responses include; Never (=0), Sometimes (=1), and Often (=2). This, therefore, yielded a scale from 0 (no violent events in the past year) to 26 (experiencing every violent event often in the past year).

Some of the covariates adopted in this study have been discussed in the literature as possible determinants of exclusive breastfeeding [26, 36‐38]. These include; child’s age measured in months; mother’s age measured in years; mother’s education measured as the number of years of formal education acquired; parity measured as the count of children a woman has; husband’s education measured as the number of years of formal education acquired by the partner; the number of other children under 5 years old in the household measured in count form; family wealth index constructed using household asset data via a principal component analysis. The family wealth index variable was already computed and available as part of the DHS dataset. Other covariates include mother’s employment status within the previous 12 months to the survey and we categorized this as binary (employed vs. not employed); the number of times antenatal care was attended based on WHO’s recommendation stipulating a minimum of four times, we grouped this as binary (less than 4 times vs. 4 times or more); metropolitan status grouped as binary (rural vs. urban) residency; and child’s size at birth as perceived by the mother grouped into 3 categories (small, medium and large). Since the child’s actual birthweight was not adequately captured in the survey, the mother’s perception of this was used as a proxy.

We used non-weighted (sampling weights) cross-tabulation to present the distribution of the independent variables and covariates by the groups of EBF practice (EBF vs. mixed-feeding). We conducted both chi-square and t-test bivariate analyses to examine the association between the dependent variable and the independent variables. Chi-square to test for the relationship between variables and t-test for difference between means. To avoid the issue of multicollinearity, we conducted a diagnostic check between the independent variables and covariates, all variance inflation factors (VIF) were below 10, with an average VIF of 1.82.

Furthermore, we performed two logistic regression analyses. One was a complete case analysis that only included observation with no missing values (N = 2465). The second regression was a multiple imputation analysis where the missing observation of 7.6% was imputed (N = 2668). This was to help determine if there had been any serious attenuations as a result of the case-wise deletion of observations and to also make use of all available information in the selected sample. We conducted this imputation using the Markov Chain Monte Carlo (MCMC) method [39, 40], under the assumption of missing at random (MAR) [40]. We specified 10 imputations, and this was deemed sufficient to re-create the variance-covariance estimation as indicated by the Monte Carlo error check [39, 41]. Each of the regression methods contained five models, each model had one independent variable with covariates. Stata version 15.1 was used to implement the data imputation, as well as to conduct all analyses. All regression analyses were done using the survey design (sampling weights). The adjusted odds ratio was reported at the 95% significance level threshold.

As shown in Table 2, the mean age of respondents was 28 years (SD = 6.5), the average years of education attained was 5 years (SD = 5.5), and majority (62.4%) were not employed. In regards to their fertility, a greater proportion (83%) were multiparous, and a little more than half (55%) reported to have attended antenatal care four times or more during the pregnancy of their last child which was included in the dyad selection. The mean age of the children was about 2 months (SD = 1.1) and the majority of them (the children) 84.7% were perceived to be of an average or large size at birth. A greater proportion of the respondents (64.8%) were residents in rural areas.

The prevalence of IPV among the study sample was 21.1%, with psychological IPV being the most reported (16.6%) and sexual IPV the least (4.6%). Physical IPV on the other hand had a prevalence of 11.7%. The average frequency score of IPV among the respondents was 0.69 (SD = 2.0). Among the exposure variables, the chi-square bivariate only indicated a significant relationship between physical IPV and EBF at a 95% significance level (Table 3). Additionally, the t-test statistics also indicated a significant difference in the means of frequency score of IPV by the groups of EBF (EBF vs. mixed-feeding) at 95% significant level: t-test = 2.03; p < 0.05 (Table 3).

As shown in Fig. 1, the age of infants is negatively associated with the practice of exclusive breastfeeding. While over 85.3% of infants in the first month of life had EBF, it was only 55.2% among children in their third month of life and a further decline to 29.5% among children in their sixth month of life.

Furthermore, while the proportion of EBF practice among women who experienced any form of IPV does not clearly differ for children aged 2 months or less, and children 3–4 months old, the proportion of non-EBF practice was clearly higher among children 5–6 months old (Fig. 2).

The regression analyses examined the association between forms of IPV and EBF practice (EBF vs. mixed-feeding), while adjusting for covariates. In the complete case analysis, those who reported experiencing physical IPV had a 31% (AOR 0.73; 95% CI 0.53, 1.01; p < 0.05) reduced likelihood of EBF practice when compared against those who did not (Table 4, Model 2). Similarly, a unit increase in the frequency of IPV was associated with 5% (AOR 0.69; 95% CI 0.49, 0.98; p < 0.05) reduced likelihood of EBF practice (Table 4, Model 5). Although, the experience of psychological, sexual, and the combined variable (any form of IPV), showed the tendency of a reduced likelihood of EBF practice, but the effects were not statistically significant at the 95% threshold (Table 4, Models 1, 3, 4).

The results of the imputed regression were similar in direction, but with a difference in magnitude. Additionally, statistical significance was retained for both psychological IPV and the combined variable (any form of IPV). Those who reported experiencing psychological IPV had 34% (AOR 0.66; 95% CI 0.47, 0.92; p < 0.05) reduced likelihood of EBF practice when compared against those who did not (Table 5, Model 1). While those who reported experiencing any form of IPV had 26% (AOR 0.74; 95% CI 0.55, 1.00; p < 0.05) reduced likelihood of EBF practice when compared against those who did not (Table 5, Model 4). Furthermore, the effect of physical IPV increased from 31 to 37% (AOR 0.63; 95% CI 0.44, 0.90; p < 0.05) (Table 5, Model 1), while the effect of a unit increase in the frequency of IPV remained unchanged. (Table 5, Model 5).