Sample population
Our cross-sectional study is derived from a nationally representative cohort study conducted jointly by the Economic Growth Center (EGC) at Yale University and the Institute of Statistical, Social, and Economic Research (ISSER) at the University of Ghana, Legon. This study was approved by the Yale Human Research Protection Program (protocol number 901004605). The survey used a two-stage stratified sample design in which enumeration areas (EA) throughout the 10 regions in Ghana were randomly selected, in proportion to 2009 regional population estimates, and then 15 households were randomly selected from each EA. EAs were oversampled in the Upper East and Upper West regions to allow for a sufficient number of households to be interviewed. A total of 5,009 households from 334 EAs were interviewed. Less than one percent of households (32 households) refused to be interviewed. All participants first provided informed consent and then completed face-to-face interviews. There were seventeen interview teams each consisting of a supervisor, a senior interviewer, four interviewers and a driver. Our analytic sample was limited to non-pregnant women who had been pregnant and involved in a relationship within the last 12 months (N = 418).
Measures
Outcome
We used two items to measure inadequate antenatal care, including whether or not women used antenatal care in their most recent pregnancy and if so, the number of antenatal care visits they attended. Women who reported attending fewer than 4 antenatal care visits were classified as having inadequate antenatal care, according to current recommendations by the World Health Organization [19].
Independent variables
Empowerment was measured using several conceptualizations based upon prior literature [8],[9] and data availability. First, we assessed partner control with four items about their relationship in the last 12 months: 1) “partner frequently accused [respondent] of being unfaithful”, 2) “partner frequently tried to limit [respondent’s] contact with her family”, 3) “partner insisted on knowing where [respondent] was at all times”, (freedom of movement) and 4) “partner did not trust [respondent] with money”. Women agreed or disagreed with each statement. Additionally, we examined each indicator individually and also tallied the number of items with which the women agreed in order to derive an index of partner control, ranging from 0 to 4. Higher scores indicated greater partner control among women. Last, participants indicated whether or not they had been emotionally abused (including being insulted or threatened) and physically abused (including being pushed, hit, slapped, having had something thrown at them, being kicked, dragged, or beaten up) in the last 12 months by their partners.
Socio-demographic characteristics
We included socio-demographic characteristics as possible controls as has been done in previous research [8]-[10],[13] and based on data availability. These characteristics included maternal age in years, marital status (never married, married, and formerly married), formal education (none, primary or less, middle, and secondary or above), religion (Christian, Muslim, traditional, and no religion), residential location (urban and rural), geographical region (10 regions in Ghana), overall health (very healthy, somewhat healthy, and unhealthy), and total number of children born. We also included overall wealth which was estimated with a 5-level household asset index constructed with principal component analysis of groupings of durable assets and living conditions (e.g., ownership of a refrigerator or computer and if the household uses safe roofing material or electricity for cooking or lighting) [20],[21].
Statistical analysis
We generated descriptive statistics to describe the sample and to explore our measures of empowerment. We then examined unadjusted associations between empowerment variables and the use of inadequate antenatal care and between socio-demographic variables and the use of inadequate antenatal care with Rao-Scott chi-square tests and t-tests for categorical and continuous variables, respectively. We then constructed a multivariable logistic regression model to determine the independent and unique associations between the empowerment variables and inadequate use of antenatal care. We included all socio-demographic variables and then used backwards elimination to derive the most parsimonious model relative to the empowerment variables. We removed non-significant (P-values >0.05) empowerment variables one by one beginning with the variable with the greatest P-value until the remaining empowerment variables were significantly (P-values <0.05) associated with inadequate use of antenatal care. Last, we explored the possibility of education moderating effects between empowerment and inadequate use of antenatal care. We created interaction terms by multiplying the empowerment variable by education dummy variables and then testing the interaction terms in the final multivariate logistic regression model. All analyses were weighted and adjusted for the complex sample design. SAS V.9.3 was used to conduct all analyses. We present results as unweighted counts and weighted percentages and odds ratios.