Background Birth preparedness and complication readiness (BPCR) is a package of interventions recommended by the World Health Organization to improve maternal and newborn health and it is provided and implemented through a focused antenatal care program. This study aimed at assessing the uptake of birth preparedness and complication readiness messages, and compliance with each key message, among Ethiopian women during their recent pregnancies using the 2016 demographic health survey report. Methods The data for this study was taken from the Ethiopian Demographic and Health Survey, which was conducted from January to June 2016 and covered all administrative regions. STATA version 16 was used to analyze a total of 4,712 (with a weighted frequency of 4,771.49) women. A multilevel mixed-effects logistic, and multilevel mixed-effect negative binomial regressions were fitted, respectively. Adjusted odds ratio (AOR) and Incidence rate ratio (IRR) with their corresponding 95% confidence interval (CI) were used to report significant determinants. Results More than half, 56.02% [95% CI: 54.58, 57.41] of women received at least one birth preparedness and complication readiness message. Being in the richest wealth quintiles (AOR = 2.33; 95% CI: 1.43, 3.73), having two birth/s in the last five years (AOR = 1.54; 95% CI: 1.13, 2.10), receiving four or more antenatal visits(AOR = 3.33; 95% CI: 2.49, 4.45), and reading a newspaper at least once a week (AOR = 1.27; 95% CI: 1.07, 1.65) were the individual-level factors, whereas regions and residence(AOR = 1.54; 95% CI: 1.11, 1.96) were the community-level factors associated with the uptake of at least one BPCR message. On the other hand, receiving four or more antenatal visits (IRR = 2.78; 95% CI: 2.09, 3.71), getting permission to go to a health facility (IRR = 1.29; 95% CI: 1.028, 1.38), and not covered by health insurance schemes (IRR = 0.76; 95% CI: 0.68, 0.95) were identified as significant predictors of receiving key birth preparedness and complication readiness messages. Conclusion The overall uptake of the WHO-recommended birth readiness and complication readiness message and compliance with each message in Ethiopia was found to be low. Managers and healthcare providers in the health sector must work to increase the number of antenatal visits. Policymakers should prioritize the implementation of activities and interventions that increase women’s autonomy in decision-making, job opportunity, and economic capability to enhance their health-seeking behavior. The local administrative bodies should also work to enhance household enrollment in health insurance schemes.
The study relied on population-based, nationally representative data from the 2016 Ethiopian Demographic and Health Survey (DHS), the fourth in a series of standardized national-level population and health surveys carried out as part of the global Demographic and Health Survey (DHS) program [4]. Ethiopia is in North-eastern (horn of) Africa, between 3° and 15° North latitude and 33° 48° and East longitudes. Ethiopia’s healthcare system is divided into three levels: primary, secondary, and tertiary care. Primary hospitals, health centers, and health posts provide primary care, general hospitals provide secondary care, and specialized hospitals provide tertiary care. The survey was conducted from January 18, 2016, to June 27, 2016, by the Central Statistical Agency (CSA) in collaboration with the Federal Ministry of Health (FMOH) and the Ethiopian Public Health Institute, with technical assistance from ICF International and financial support from USAID, the government of the Netherlands, the World Bank, Irish Aid, and UNFPA. Data of the study participants were accessed on October 23, 2022 from DHS website, their URL: www.dhsprogram.com by contacting them via personal email communication with a possible justification for the data request. Permission was granted via email after reviewing the account. A cross-sectional study design using secondary data from 2016 EDHS was conducted. The source population consisted of 15,683 women who had given birth within five years preceding the survey. The study population consisted of 4,712 women who had complete information on the uptake of BPCR messages during their ANC visit, as well as the contents of those messages, and the entire analyses were conducted on them. Due to a lack of information on service uptake, a total of 10,971 respondents were excluded from the analysis (Missing values). The 2007 Ethiopia Population and Housing Census sampling frame was used, which included 84,915 enumeration areas (EAs), with each EA covering 181 households. A stratified two-stage cluster design was used to select respondents, as each region was stratified into urban and rural areas. The first step was to select 645 clusters (202 urban and 443 rural areas) with a probability proportional to the size of the enumeration area and independent selection within each stratum. The household listing was completed in all of the selected EAs between September and December 2015. The second stage involved the selection of 28 households per cluster using an equal probability systematic selection of eligible women aged 15–49 years. With a response rate of 94.6%, a sample of 16,650 households and 15,683 women aged 15–49 years was identified. Furthermore, the survey design and methodology were detailed in the 2016 EDHS [4]. This study had two outcome variables. The first outcome variable. Was the receipt of BPCR messages during ANC visits which was assessed by the question “During any of your antenatal visits were you told about BPCR?” If a mother said “Yes,” the response was labeled as 1, otherwise it was labeled as 0. The second outcome variable. Was the number of WHO-recommended BPCR messages received by a mother during pregnancy which was assessed using six items. Those key messages were about determining the place of birth, obtaining necessary supplies for childbirth, preparing emergency transportation, saving money for emergency expenses, identifying companions during labor and childbirth, and securing potential blood donors. Information on these six key BPCR messages was derived from the response to the question: Were you told about your Place of birth? Were you told about supplies needed for birth? Were you told about emergency transportation?… The answers were recorded as Yes (= 1) or No (= 0). During the same pregnancy, a single mother may be informed about the place of birth or the supplies required for childbirth several times. However, because the mother was asked to report any messages she received at least once, any response was recorded as a single message. Based on the responses, a composite index of BPCR was created, which is simply a count of the number of key messages received. The variable had a minimum value of zero indicating that no BPCR messages were delivered to the women and a maximum value of six indicating that the women received all six key messages. A similar type of content index was used by other recent studies [22, 23]. Religion, ethnicity, age, place of residence (urban and rural), educational level (no education, primary, secondary, and higher), husband’s education (no education, primary, secondary, higher), and wealth status of women’s household were considered from socioeconomic and demographic characteristics of women. The wealth index was divided into five categories: poorest, poorer, middle, richer, and richest. The wealth quintile of women’s households in the EDHS is a composite indicator based on housing characteristics and ownership of household durable goods that was calculated using principal component analysis. Obstetric characteristics like parity (nulliparous, primiparous, multiparous, and grand multiparous), gravidity, the total number of birth in the last five years, pregnancy status when she became pregnant (wanted, mistimed, unwanted), total children ever born, ever had a termination of pregnancy. Maternal health service-related characteristics like the frequency of Antenatal care visits, place of receiving ANC (home, public, private, and NGO), contraceptive use (Yes or No), the decision-making power on own health care (self-decision/joint decision with husband, husband alone, and other), covered by health insurance(Yes or No), and exposure to the newspaper, radio, and television (not at all, less than once a week or at least once a week) were considered. Furthermore, problems encountered by women in accessing medical help for themselves, such as distance to a health facility, obtaining permission to visit a health facility, and obtaining the money required for treatment, were assessed and rated as a big problem or not a big problem. Some of the behavioral characteristics of the respondents like alcohol consumption and cigarette smoking were assessed with a “Yes” or “No” response. The necessary data from EDHS 2016 report were checked for consistency and missing values. STATA/SE version 16.0 was used for cleaning, recoding, variable generation, labeling, and analysis. The sample allocation to different regions, as well as urban and rural settings, was not proportional in the EDHS. As a result, sample weights were used to estimate proportions and frequencies to account for disproportionate sampling and non-response. The weighting procedure was thoroughly explained in the 2016 EDHS report [4]. Descriptive statistics were computed to describe the characteristics of the overall sample respondents (mothers) across a set of covariates. The use of a multilevel modeling approach accounts for the EDHS data’s hierarchical nature, as households were selected within EA clusters. There may be unobserved cluster characteristics influencing BPCR message uptake among women, such as the availability and accessibility of health services, cultural norms, and predominant health beliefs [4]. Thus multilevel mixed-effects models (cluster/region-specific random effects) were applied to identify the predictors. Accordingly, two different modes of analysis were implemented to estimate both the independent (fixed) and community-level (random) effect of the explanatory variables on our dependent variables: i. First, to examine the relationship between each predictor and the first outcome variable(a receipt of at least one BPCR message), a multilevel bivariable logistic regression analysis was performed. In this analysis, variables with p-values less than 0.25 were candidates for a multilevel multivariable mixed-effect logistic regression analysis. Then a multilevel mixed‑effects logistic regression analysis was run. In a multivariable multilevel mixed-effect logistic analysis, four models with the variables of interest were fitted, and the best-fitting model was chosen. Model-I is a null model, Model II is a model with only individual-level factors, Model III is a model with only community-level factors, and Model IV is a full model. The full model (Model IV) was fitted to examine the effect of individual and community-level predictors on the outcome variable at the same time. The adjusted odds ratio with the corresponding 95% confidence interval was computed and reported to demonstrate the strength of the association and its significance. Variables having a p-value <0.05 were considered as having a significant association with the outcome variable. The model comparison was done using deviance and the fourth model with the lowest deviance was selected as the best-fitted model ii. To identify factors associated with the secondary outcome variable (a receipt of the recommended number of BPCR messages), a generalized linear model (GLM) with a multilevel mixed-effect negative binomial regression was run. Since the number of key BPCR messages received is a non-negative integer (count), most of the recent thinking in the field has used the Poisson regression model as a starting point [24, 25]. The most serious limitation of Poisson regression is that it assumes that the variance of the count response variable’s distribution is equal to its mean, which is known as the assumption of equidispersion. If this assumption is breached, the Poisson regression model’s estimates remain consistent but produce incorrect inferences about the parameters [26]. In the current case, the mean and the variance of the count outcome variable were 1.25 and 2.51, respectively. As a result of the assumption being violated, the data were over-dispersed, and a multilevel mixed-effect negative binomial regression model was fitted [25, 27]. Independent t-tests and analysis of variance (ANOVA) were used to determine whether there were statistically significant differences in the mean number of BPCR messages across each categorical variable. Those variables with p-values less than 0.05 were eligible for a multilevel mixed-effect negative binomial regression using a generalized linear model (GLM) to identify the determinants of the number of BPCR messages uptake. During multivariable multilevel mixed-effect negative binomial regression, four models with the variables of interest were fitted, and the best-fitting model was chosen. Model-I is a null model, Model II is a model with only individual-level factors, Model III is a model with only community-level factors, and Model IV is a full model. The full model (Model IV) was fitted to examine the effect of individual and community-level predictors on the outcome variable at the same time. Finally, the incident rate ratio (IRR) with a 95% confidence interval was reported, and statistical significance was determined at a p-value less than 0.05. Measures of random effect like Intra-class correlation coefficient (ICC), a proportional change in variance (PCV), and median odds ratio (MOR) were estimated. ICC explains the cluster variability, while MOR can quantify unexplained cluster variability (heterogeneity). In both cases (first and second outcome variables), the results of the random effects model showed the presence of variations of the random factor in the null model, indicating the existence of variation in the receipt of BPCR messages. Thus, to account for this variation, a multilevel mixed-effect logistic regression (for the first outcome variable) and a multi-level mixed-effect binomial regression (for the second outcome variable) model were considered for further analysis (Tables (Tables55 and and6).6). Model IV had the lowest AIC value in both cases (primary and secondary outcome variables) and was selected as the best model fit for the data. Furthermore, as fitted models progressed from the empty model (Model-I) to Model-II, Model-III, and Model-IV, the value of the deviance (-2*log-likelihood) results consistently decreased, indicating that the fitted models were a better fit to the data. Key: 1: Reference category; AOR = Adjusted odds ratio ** Statistically significant at p-value <0.05 Key: 1: Reference category; IRR = Incidence rate ratio * Statistically significant at p-value <0.05 Following registration with possible justification, ICF International granted permission to access the dataset used for this study. The retrieved data were only used for the registered research, and data were not shared with anyone other than the coresearchers. The information was kept private, and no attempt was made to identify any household or individual respondent.