Factors affecting birth preparedness and complication readiness in Jimma Zone, Southwest Ethiopia: A multilevel analysis

Introduction: birth preparedness and complication readiness have been considered as comprehensive strategy aimed at promoting the timely utilization of skilled maternal health care. However, its status and affecting factors have not been well studied at different levels in the study area. Thus, this study was aimed to fill this gap by conducting community based study. This is a community based cross-sectional study conducted in Jimma Zone from June-September 2012. Jimma Zone is one of the 17 Zones of the Oromia Regional State of Ethiopia. The capital of the Zone, Jimma Town, is located at 346kms Southwest of Addis Ababa. The Zone has a total of 17 rural districts (“Woredas”) and two town administrations. Based on the 2007 national population census conducted by the Central Statistics Agency (CSA) of Ethiopia, the Zone has a total population of 2.6 million, of whom 88.7% are rural inhabitants [14, 15]. The study population for this study was pregnant women. The sample size was determined by using Epi-Info V.3.5.1 by considering two sample comparison of proportions based on the following assumptions. Among all the factors considered, antenatal care (ANC) was found to give the largest sample size. The prevalence of birth preparedness and complication readiness among mothers who attended ANC is estimated to be 22% (p1=0.22) and among those who didn’t attend ANC is to be 13% (p2 = 0.13) [13]; 95% level of confidence and 90% power were considered. The prevalence of ANC in the general population in same study was 45%. As a result, a ratio of 1:1 was used (r = 1). As multistage clustered sampling method was used, a design effect of 2 was considered. Finally, 10% was added for non-responses and the final sample size became 1650. However, this study was part (baseline) of a big longitudinal study to determine the effect of maternal and neonatal health care on neonatal health status, in which 3612 pregnant women have been followed up. As a result, all the 3612 pregnant women were included in the analysis for this study. Multi-stage clustered sampling technique was used to identify pregnant women for the study. At first stage, the Zone was stratified as rural districts (17 in number) and town administrations (2 in number, Jimma and Agaro). Then, 5 districts were selected by simple random sampling from the 17 districts. At second stage, all the selected 5 districts were clustered by “Kebeles” (A “kebele” is the smallest administrative unit having 5000 population on average) and stratified in to urban and rural “Kebeles”. Then, by simple random sampling method, 9 rural “Kebeles” and 2 urban “Kebeles” were selected from each selected district. This number of “kebeles” was determined based on expected number of pregnant women per “Kebele. Jimma town administration and Agaro town administration have 13 and 5 “Kebeles”, respectively and all were included. With this, a total of 73 “Kebeles” (clusters) were included in the study. Then, for all selected “kebeles” pregnant women were enumerated by using house-to-house visit and all obtained were included in the study. Pre-tested interviewer administered structured questionnaire was adapted from the safe motherhood questionnaire developed by maternal and neonatal health program of JHPIEGO [11]. The indicators for the wealth index were adapted from EDHS [6]. The questionnaire was prepared in English, then translated to local languages “Afan Oromoo” and Amharic and back translated to English by different experts to check its consistency. Females, who had completed 10th grade or above, were recruited, trained and collected the data. The data collection process was supervised strictly by trained supervisors and principal investigators. To control the quality of data, in addition to training, pretest, supervision and use of local languages, the inter-item consistency of the indicators to measure the composite score of BP and CR was checked by using Chronbach-alpha at 0.7 cut-off point. The collected data were coded and entered into Epidata V.3.1. and exported to SPSS for windows version 20.0 for cleaning, editing and analysis. Descriptive analysis was done by computing proportions and summary statistics. Wealth quintiles were determined by using Principal Component Analysis (PCA). As Jimma town and Agaro town administrations were both purposefully included, the status of BP and CR was estimated by calculating weighted percentage based on the complex sample survey procedure by considering probability of exclusion at each stage and non responses in order to avoid over estimation. Bivariate analysis was done by using cross tabulation to see associations between the independent and dependent variables. Then, all variables having P < 0.25 were considered as candidates for the mixed-effects multilevel logistic regression model. Because of the multi-stage clustered sampling procedure, mixed effects multilevel logistic regression model was used by using STATA 13. This model was preferred in order to avoid the clustering effects as well as ecological fallacy. “Kebeles” was considered as clusters and "kebele" level variables were taken as higher level (level-2). Mothers were nested within their households and maternal individual variables and household characteristics were taken as lower level (level-1) variables (Table 1, Table 2). Goodness-of-fit of the multilevel model was tested by the log likelihood ratio (LR) test. To evaluate the existence of sufficient variance at the cluster level to use multilevel analysis, intercept-only model was fitted as Logit (pij) = y00+u0j and Interclass Correlation Coefficient (ICC) was determined by dividing between group variance (δ2µ0) by the total variance variance (between group variance (δ2µ0)+within group variance (δ2e)). However, within group variance (δ2e) is not directly obtained for dichotomous outcome variable; instead, it was estimated by Π2/3. Description of the dependent, level-2 and level-1 predictors, Jimma Zone, Southwest Ethiopia, June-September, 2012 (Part I) Description of obstetric related level-2 predictors, Jimma Zone, Southwest Ethiopia, June-September, 2012 (part II) Then, to identify the determinate factors, the full model was fitted as: Logit (pij)= p00+y01Zj+y10Xij+u0j+u1jXij. Where: Logit(p ij) = dependent variable at unit i in cluster j, Xij= individual explanatory variable in cluster j, Zj= group level explanatory variable, y 00= fixed intercept, y01 and y10 = fixed slopes and u 0j and u1j= random effects at level-2. Multicollinearity between the independent variables was assessed by using variance inflation factors (VIF > 10 was considered as suggestive of multicollinearity) before interpreting the final output. But, no significant multicollinearity was detected as VIF for all variables were < 5. In addition cross-level two-way interactions were checked, particularly between place of residence, access to health facilities and ANC visits. However, no significant interaction was detected (P > 0.05 for each). Ethical approval was obtained from the Institutional Review Board (IRB) of College of Health Sciences of Addis Ababa University. Formal permission letters were secured from all respective local administrators. Written informed consent was obtained from each respondent before actual data collection and confidentiality of the data were strictly maintained.