Study settings and period
The study was conducted in communities of Eastern Ethiopia living in town of Harar and Kersa district under Health and Demographic Surveillance System (HDSS), namely Kersa and Harar HDSS sites. The Kersa HDSS has 24 randomly selected Kebeles (smallest administrative unit in Ethiopia) with an estimated 127,000 inhabitants. The Harar HDSS has 12 Kebeles with an estimated 56,000 inhabitants. Trade and farming are the main sources of living in these communities and Khat (Catha edulis), stimulant green leaf, chewing is prevalent in the area [39]. According to the 2016 Ethiopian Demographic Health Survey, the rates of neonatal mortality for the regional states of Harari (where Harar town is located) and Oromia (where Kersa district is located) was 34/1000 and 37/1000 live births, respectively[7]. The study was conducted between January 2018 and October 2018.
Study design and sample
A community-based matched case–control study was conducted from January to October 2018. The study population for the study was currently partnered mothers in their sixth week of giving live births permanently residing in the study area. A mother was considered currently partnered if she was either living together with husband (formally married) or living together with the intimate partner in an informal marriage at least during the index pregnancy. Cases constituted biological mothers of the live born baby who later died before the age of 28 days of life. Controls were biological mothers of the live neonates at the end of the first 28 days of birth.
The sample size was determined using PS (power and sample size calculation) software version 3.1.2 for a matched case control study with the following assumptions. A survey of violence against women in Ethiopia documented prevalence of violence during pregnancy of 4 % [7], and thus we have assumed experience of violence during pregnancy in mothers with live born children to be 4%. As no prior published study relating IPVP and neonatal mortality in Ethiopia, and based on earlier study in Egypt[40], we assumed a 3 times higher odds (OR) for neonatal mortality in those exposed to violence during pregnancy than not exposed to it. Thus, assuming IPVP among the controls to be 4%, OR of 3, power of 80%, a 5% significance level, a 4:1 control to case ratio; a correlation coefficient (ɸ) for exposure between matched case and control being unknown and hence considering a 0.2 phi(ɸ) coefficient, as suggested by Dupont (p.1166-67)[41], a minimum of 86 cases and 344 controls were required for the study. Adding 20% for possible non-response, we needed a total of 103 cases and 412 controls.
In the study, we employed the following procedures to identify the cases and controls. Initial census and subsequent three months interval house-to-house surveys were carried out to identify pregnant women using a standard pregnancy procedure as described by Assefa et al [42]. Then, for those women who were found pregnant in the screening, a follow up of the pregnancy status were carried out throughout the study period. Interviews were then administered for those women who had live births, but only when they reached in their sixth week of postpartum. In such a way, the required number of cases (103 dead neonates) was reported among sample of 3015 eligible women in their sixth week of postpartum during the study period. The details of sample recruitment and procedures have been described previously in [43]. Each case was matched with four controls by date of birth, sex of the newborn and area of residences.
Data collection
Face-to-face interviews were carried out using a structured questionnaire that consists of socio-demographic, obstetric history and experiences of violence related questions. The WHO multi-country study domestic violence tool which was translated to local languages of the study area was employed to elicit the necessary information on the IPV and related questions [44]. The instrument was pretested in a district outside the study area and necessary adjustments were made accordingly. In administrating data collection, secondary level education and above female data collectors were involved with training given to them using the WHO training guide [45]. The WHO field work manual [46] was adapted to be used as a quick reference in the field work. The Institutional Health Research Ethics Review Committee of Haramaya University, College of Health and Medical Sciences ethically cleared the study (IHRERC/153/2017, dated October 26/2017). Written informed consent was obtained from each participant before interview. Following the WHO ethical and safety recommendation for studies on VAW[47], the study was initially framed as women’s health, family relationship and life experiences study. However, while in the consent acquisition, each woman was personally introduced to the specific nature of the study before the interview was conducted. Interviews were conducted in a private setting and voluntarily. To maintain confidentiality, no identifying information was recorded on questionnaires and a unique code was used to distinguish the questionnaires. Each study participant was also informed the purpose of the study and the risks and benefits of participating in the study.
Variables and measurements
Outcome and main exposure variables
Neonatal mortality is an outcome variable determined based on the woman's report of the death of her baby within 28 days of life. The women experience of different forms of IPVP (Psychological, Physical, Sexual violence and husband's controlling behaviors) were considered as main independent variables, and were assessed using the WHO multi-country study tool[44]. Our study population is comprised of mothers currently partnered with father of index child, so for the purpose of this study, "intimate partner" refers to the father of the respondent's index child.
The Psychological IPVP was measured by at least one positive response to the following four questions that asked whether the respondent's husband/partner had (1) insulted her or made her feel bad about herself; (2) humiliated or belittled her in front of others; (3) scared/intimidated her on purpose (4) threatened to hurt her or someone close to her. Sexual IPVP was measured by at least one positive response to the following three acts: (1) If the woman reported having been physically forced to have sexual intercourse when she did not want to; (2) when the woman had intercourse out of fear of the husband/partner; or (3) when she was forced to perform other sexual acts that she found degrading or humiliating. An affirmative answer to any one of the following six behaviors constitutes physical IPVP: (1) pushing or throwing something at the woman; (2) slapping; (3) punching or hitting with a fist or something harmful; (4) kicking or dragging her; (5) choking or burning intentionally; or (6) threatening or attacking with weapon (a knife, gun). Overall, women who responded “yes” to at least one of the above 13 questions were considered as having “experienced overall IPVP”, and women who answered “no” to all of the questions were considered as “no violence experienced.” Moreover, based on the WHO violence severity scale, we labeled the violence act as moderate if husband/partner slapped the woman or throws something at her during the pregnancy (i.e. we labeled the physical violence as 'moderate' if only one or both of the first two items of the mentioned six physical violence acts happened to the woman). Whereas, experiencing any one of the other/the rest four acts (items 3 to 6 of the above mentioned physical violence) was labeled as 'severe' physical violence[44] .
Controlling behavior of the husband/partner were assessed using the below seven questions asking whether the women's husband/partner: commonly attempted to restrict women’s contact with her family, kept her from seeing her friends, insisted knowing where she was all the time, ignored her or treated her indifferently, expected her to seek permission for seeking health care for herself, constantly was suspicious that she was unfaithful, and got angry if she spoke with another man. If the women reported one or more of these acts happened during pregnancy, then they are considered experiencing controlling behavior from their husband/partner[44].
Control variables
Women's household empowerment was derived from women's household decision making autonomy and their belief/attitude towards wife beating [7]. Household decision making was determined based on the number of household decisions a woman made alone or jointly with her husband/partner about: (1) spending income; (2) obtaining health care for herself; (3) major household purchases; (4) purchases for daily household needs; (5) visit to family or relatives; The response options were: (a) respondent alone, (b) respondent and husband/partner, (c) husband/partner alone, (d) others/someone else. Each question was assigned a value of 1 if the response was (a) or (b), and 0 for (c) or (d). Attitude towards wife beating was measured as a composite index consisting of the following items: a husband/partner is justified in hitting or beating his wife if she (a) is unable to carry household chores (b) disobeys/argues with him (c) asks him about his extramarital affairs (d) is barren and (e) refuses to have sexual intercourse with him. Each item had a response of ‘Yes’ or ‘No’, and a value of ‘0’ was given if the woman accepted/justified beating and ‘1’ otherwise.
Social support was measured by the Oslo Social Support Scale (OSSS-3) which has a 3-item questions commonly used to assess social support level[48, 49]. Women were asked to respond on (i) the number of people so close to them that they count on if they encounter great personal problems ('none', '1-2', '3-5' and 5+); (ii) the interest and concern people show them in what they do ('none', 'little', 'uncertain', 'some', and 'a lot') and (iii) how easily they get practical help from neighbors when they need it ('very difficult', 'difficult', 'possible', 'easy' and 'very easy'). The sum score of the scale ranges from 3 to 14, and has been grouped in to three broad categories[50]: “poor social support” 3–8, “moderate social support” 9–11, and “strong social support” 12–14.
In line with the Mosley and Chen conceptual framework[27], different confounders that have been conceptually linked to neonatal survival, such as women's age at birth[51, 52], education level[30, 31], working status[30], and alcohol consumption during index pregnancy[53] were also considered in this study. Moreover, obstetric and neonatal related variables such as utilization of antenatal and postnatal cares, number of previous pregnancies, parity, perceived weight of neonate at birth, place of birth and type of birth (singleton or multiple births) were considered as confounders in the association of IPVP and neonatal mortality [28, 30, 32, 34]. Mothers' perceived reports on the birth size of the baby were considered as the proxy indicator of birth weight, and were dichotomized as small and normal size babies [7, 54].
Data management and analysis
The data were double entered and cleaned using Epi-data Version 3.1 and analyzed using STATA version-14. A principal component analysis (PCA) was carried out in order to derive the women's empowerment index using each of the variables representing women’s household decision-making autonomy and their belief in/attitude towards wife beating. The PCA yielded three categories of women's household empowerment: low, medium and high status.
Conditional logistic regression model was used to assess association between the different IPVP types (physical, sexual and psychological violence and controlling behaviors) and neonatal mortality, by considering socio-demographic and obstetrical/ pregnancy related confounders. To examine the influence of IPVP on the odds of neonatal mortality, we ran series of adjusted multivariable models; one for each type of IPVP to assess the separate effects of the different forms of IPVP (Models I-IV), and one full model (Final model) for all the types of IPVP together, in order to examine the effects of each type of IPVP controlling for the other. Each model also included the confounding variables described above.
Variables with p-value ≤0.25 in the bivariate analysis were first considered for multivariable analysis; however, known confounders such as education of women and their parity that deemed to be considered were also included in the multivariable model though they did not reach p-value ≤0.25. A p value <0.05 was used to declare the level of statistical significance in the multivariable analysis and adjusted odds ratios (AOR) along with 95% CIs were estimated. The Log-likelihood ratio test was used to select the final model. Pearson Chi-square and Hosmer–Lemeshow goodness-of-fit tests were used to test for model fitness. The explanatory variables were tested for multi-collinearity before entering to multivariable models using Variance Inflation Factor (VIF) of less than 10 values.