**2. Methods**

#### **2.1 Data sources**

The present study used data from the 2007 Bangladesh Demographic Health Survey (BDHS), conducted by the National Institute for Population Research and Training of the Ministry of Health and Family Welfare of Bangladesh from March 24 to August 11, 2007. The BDHS sample was drawn from Bangladeshi adults residing in private dwellings. A stratified, multistage cluster sample of 361 primary sampling units was constructed (134 in urban areas and 227 in rural areas). The primary sampling units were sourced from a sampling frame created for the 2001 census of Bangladesh, in which they were termed "enumeration areas".

The 2007 BDHS used five questionnaires. Of the 11,178 women deemed eligible to participate in the women's questionnaire on maternal and child health behaviors and outcomes, 10,996 participated (98.4% response rate). One woman from each household was selected at random for the domestic violence module to answer an additional set of questions regarding IPV perpetrated by her husband. Out of 4,489 women eligible to respond to the domestic violence module, only seven had to be excluded due to lack of privacy. An additional 15 women were not interviewed for other reasons. The present analyses included only currently married women aged 15-49 years with at least one singleton child below five years of age living with the respondent (n=1851) **(Figure 1).** 

#### **2.2 Outcome measures**

To provide an assessment of child morbidity outcomes we analyzed three common childhood illness: diarrhea, ARI and fever, assessed via responses to the BDHS questionnaire given to women. For each child under five years of age, women indicated

Within and outside of South Asia, increasing evidence has shown a linkage between high rates of IPV among women (IPV; 18%-66%) (Bates et al., 2004; Bhuiya, Sharmin & Hanifi, 2003; Jain et al., 2004) and poor child health outcomes, such as miscarriage (Silverman et al., 2007; Bair-Merritt, Blackstone, & Feudtner, 2006), child under-nutrition (Ackerson & Subramanian, 2008; Hasselmann & Reichenheim, 2006), and infant and child mortality Jejeebhoy, 1998; Ahmed, Koenig, & Stephenson**,** 2006; Leland KA &Subramanian, 2009). However, the literature on consequences of IPV on young children's morbidity pattern is limited, and weaknesses in methodology. Within South Asia a recent investigation in India indicates an association between IPV and childhood asthma (Subramanian, Ackerson, & Subramanyam, 2007). Another study found that young children of Bangladeshi women abused by their husbands were more likely to be at risk of ARI and diarrhea diseases (Silverman et al., 2009). Outside the region of South Asia, a recent study in Uganda supports that the history of women subjected to IPV predicts the risk of diarrhea and overall illness of the infant (Karamagi et al., 2007). However, most of these studies have some methodological weaknesses such as based on community specific small samples or based on husband's report of IPV or measured only the physical type of IPV by using single global question. This lack has limited our understanding of the extent to which childhood morbidity may be affected by the physical and sexual IPV, using the multiple, behaviorally specific questions based on women's report of IPV. Thus, this study, aimed to examine the association of physical and sexual forms of IPV with childhood fever, ARI, and diarrheal morbidity in a

The present study used data from the 2007 Bangladesh Demographic Health Survey (BDHS), conducted by the National Institute for Population Research and Training of the Ministry of Health and Family Welfare of Bangladesh from March 24 to August 11, 2007. The BDHS sample was drawn from Bangladeshi adults residing in private dwellings. A stratified, multistage cluster sample of 361 primary sampling units was constructed (134 in urban areas and 227 in rural areas). The primary sampling units were sourced from a sampling frame created for the 2001 census of Bangladesh, in which they were termed

The 2007 BDHS used five questionnaires. Of the 11,178 women deemed eligible to participate in the women's questionnaire on maternal and child health behaviors and outcomes, 10,996 participated (98.4% response rate). One woman from each household was selected at random for the domestic violence module to answer an additional set of questions regarding IPV perpetrated by her husband. Out of 4,489 women eligible to respond to the domestic violence module, only seven had to be excluded due to lack of privacy. An additional 15 women were not interviewed for other reasons. The present analyses included only currently married women aged 15-49 years with at least one singleton child below five years of age living with the respondent (n=1851) **(Figure 1).** 

To provide an assessment of child morbidity outcomes we analyzed three common childhood illness: diarrhea, ARI and fever, assessed via responses to the BDHS questionnaire given to women. For each child under five years of age, women indicated

nationally representative sample in Bangladesh.

**2. Methods** 

**2.1 Data sources** 

"enumeration areas".

**2.2 Outcome measures** 

whether the child had been ill with fever, experienced an episode of diarrhea, and ill with a cough accompanied by short, rapid breathing in the 2 weeks prior to the survey. A symptom of ARI was defined as report of cough accompanied by short, rapid breathing. Binary variables were created to define diarrhea, ARI and fever, which indicated the presence of each of these outcomes among the children in the past 2 weeks. A binary variable was also created to assess the overall level of illness in the child, which was dichotomized into "0" no illness and "1" as illness (combined fever, ARI and diarrhea).

Fig. 1. Selection of sample

#### **2.3 Exposures**

Women's experience of IPV was the main exposure of interest in this study. The BDHS measured IPV using a shortened and modified Conflict Tactics Scale (CTS) (Straus, 1979;

while 95% confidence intervals (95% CI) were estimated for significance testing. Multicolinearity in the logistic regression analyses was checked by examining the standard errors for the regression coefficients. A standard error larger than 2.0 indicates numerical problems, such as multicollinearity among the independent variables (Chan, 2004). In this study, all of the independent variables in the two models for each nutritional outcome had a standard error <0.90 indicating an absence of multicolinearity. Stata, version 9.0 (Stata Corp., College Station, TX, USA) with survey commands was used to account for stratification, clustered sampling, and weighing provided by the BDHS to reproduce the

Data collection procedures for the BDHS were approved by the ORC Macro Institutional Review Board. Several specific protections based on WHO"s ethical and safety recommendations for research on domestic violence were built into the 2007 BDHS (Straus, 1979; WHO, 2001). For the domestic violence section, respondents were read an additional statement informing them that the questions to follow could be sensitive and reassuring them of the confidentiality of their responses (NIPORT, 2009). Interviews were conducted under the most private conditions afforded by the environments encountered. If privacy

Nearly half of the women (49.1%) were 15-24 years old, 29.2% were uneducated, and 78.5% lived in rural areas **(Table 1)**. About 11% of the respondents had no decision-making autonomy. Regarding nutritional status, 60.3% women were considered to have normal BMI; 32.0% were undernourished or thin (BMI less than 18.5); and 7.5% were overweight or obese (BMI 25 or higher). From the total sample population, 69.9% of children were below three years of age, nearly half were female and 42.8% of the children were breastfed for 24

The prevalence of underweight, stunting, and wasting was 40.8%, 42.0%, and 19.0% respectively, while the prevalence of diarrhea, fever, and ARI was 10.1%, 38.7%, and 13.3% respectively. Overall, 45.6% children were suffering from any type of illness (diarrhea or ARI or fever) two weeks before the survey. Substantial numbers of mothers (29.0%) reported that they had suffered any IPV in the year prior to the survey; 15.5% of mothers indicated that they had experienced only physical IPV, 6.2% indicated that they had experienced only sexual IPV, and 7.3% indicated that they had experienced both types of IPV **(Table 1)**. In bivariate analysis, several significant differences were observed in the prevalence of IPV perpetration across various socio-demographic groups **(Table 2)**. Specifically, significantly a higher prevalence of perpetration of any form of IPV, physical IPV only and both physical and sexual IPV was identified among younger women (aged 15-24 years) and women who used LPG/natural gas/biogas as cooking fuel compared with older women and women used Biomass/charcoal as cooking fuel. Regarding educational status, significantly a higher prevalence of any form of IPV and past-year perpetration of both physical and sexual IPV was identified among women having no education. Significantly a higher prevalence of any form of IPV and past-year perpetration of both physical and sexual IPV was also identified

could not be ensured, the interviewer was instructed to skip the module.

months or more, and only 8.8% used LPG/natural gas/biogas as cooking fuel.

among women having children suffering from any types of recent illness.

national population.

**3. Results** 

**2.6 Ethical considerations** 

**3.1 Descriptive statistics** 

Straus & Gelles, 1990). Perpetration of IPV by the husband in the year prior to the survey was assessed via 8 items included in the survey given to the women. A positive response to any one of the following behaviors indicated the perpetration of physical IPV: (1) pushing, shaking, or throwing an object; (2) slapping; (3) twisting her arm or pulling her hair; (4) punching or hitting with a fist or something harmful; (5) kicking or dragging; (6) choking or burning; or (7) threatening or attacking with a knife or gun. Perpetration of sexual IPV was indicated by a positive response to 'physically forcing her to have sexual intercourse even when she did not want to'. These assessments were recorded to create a four-level categorical variable reflecting the experiences of three categories of IPV: physical IPV only, sexual IPV only, and both physical and sexual IPV. The fourth category was a referent group of no IPV perpetration of either form. We also created a binary variable measuring whether a mother reported any form of IPV (physical, sexual or both); this was termed "any IPV". Though psychological violence is one of the important indicators of all IPV incidents (Leland & Subramanian, 2009) this information was not available in the current study, as it was not collected in the BDHS.

#### **2.4 Covariates**

We included several socio-demographic, environmental and nutritional variables theoretically and empirically linked to IPV (Uthman, Lawoko,& Moradi T, 2009; Bates et al., 2004) and common childhood illness (Rayhan, Khan, & Shahidullah, 2007; Daniel et al., 2008; Tomkins, Dunn, & Hayes, 1989; Gasana et al., 2002; Cairncross e al., 2010; Barros et al., 2010).These variables included: maternal age (15-24 years, 25-34 years or 35-49 years), maternal education (no education, primary or secondary and higher), maternal decision making autonomy, mother's occupation (unemployed or agriculture/non-manual or manual), mother's BMI (thin, normal or overweight), residence (rural or urban), household members (2-4, 5-6 or 7+), parity (1, 2 or 3+ ), wealth index, type of cooking fuel (biomass/charcoal or LPG/natural gas/biogas), child sex (male or female), child's age (0-11 months, 12-23 months, 24-35 months or 36-59 months), initiation of breastfeeding (early or late), and duration of breastfeeding (0-11 months, 12-23 months or ≥24 months). We used BDHS wealth index as a proxy indicator for socioeconomic position. The BDHS wealth index was constructed from data on household assets, including ownership of durable goods (such as televisions and bicycles) and dwelling characteristics (such as source of drinking water, sanitation facilities, and construction materials). Principal components analysis was used to assign individual household wealth scores. These weighted values were then summed and rescaled to range from 0 to 1, and each household was assigned to either the poorest, middle, or richest tertials.

#### **2.5 Statistical analyses**

We calculated descriptive statistics for socio-demographic, environmental, IPV, nutritional, and morbidity characteristics for our sample. Demographic and socio-economic differences of any physical or sexual IPV perpetration were assessed by χ2 analyses. The 2-tailed significance level for all analyses was *p* < .05. We created 2 fully adjusted models to analyze the appropriate binary for each morbidity outcome of diarrhea, ARI, fever, and any illness (any vs. no IPV; and the separate effects of physical only, sexual only and both physical and sexual IPV). We entered all covariates simultaneously in the multiple regression models. Adjusted odds rations (AOR) were estimated to understand the strength of the associations while 95% confidence intervals (95% CI) were estimated for significance testing. Multicolinearity in the logistic regression analyses was checked by examining the standard errors for the regression coefficients. A standard error larger than 2.0 indicates numerical problems, such as multicollinearity among the independent variables (Chan, 2004). In this study, all of the independent variables in the two models for each nutritional outcome had a standard error <0.90 indicating an absence of multicolinearity. Stata, version 9.0 (Stata Corp., College Station, TX, USA) with survey commands was used to account for stratification, clustered sampling, and weighing provided by the BDHS to reproduce the national population.

#### **2.6 Ethical considerations**

Data collection procedures for the BDHS were approved by the ORC Macro Institutional Review Board. Several specific protections based on WHO"s ethical and safety recommendations for research on domestic violence were built into the 2007 BDHS (Straus, 1979; WHO, 2001). For the domestic violence section, respondents were read an additional statement informing them that the questions to follow could be sensitive and reassuring them of the confidentiality of their responses (NIPORT, 2009). Interviews were conducted under the most private conditions afforded by the environments encountered. If privacy could not be ensured, the interviewer was instructed to skip the module.
