Data source
This study was based on secondary data analysis from the 2016 Ethiopia Demographic and Health Survey, which was collected cross-sectional from January to June 2016. The Federal Democratic Republic of Ethiopia is the second-largest populous country in Africa, with 102.4 million people with an annual growth rate of 2.46%. The country has nine regional states and two city administrations and has a three-tier health care system with primary health care units situated to the nearby communities. The FDRE ministry of health declared that essential maternal health services like antenatal and postnatal follow up, delivery, and provision of family planning drugs are available and accessible to women free of charge in all health facilities. The institutional delivery rate was 66% in 2016, which indicates one-third of deliveries continued to be at home.
Population and sample
All reproductive-aged women in Ethiopia were the source population, and those women in the enumeration areas (EA) were the study population. We used individual record (IR) file to extract the study participants of reproductive age women.
The Ethiopia demographic and health survey used a stratified two-stage sampling technique to select the final study participants. Initially, the enumeration areas were stratified into urban and rural, of which 202 and 443 EAs were selected from urban and rural, respectively, and the probability sampling technique used to select the study participants from each EA stratum independently. In the second stage, a fixed number of 28 households per cluster were selected with an equal probability systematic method from the newly created household lists. In the interviewed households, 16,583 eligible women identified for interviews of which, 15,683 women had completed the interview and included in the final analysis. The detail of the methodology is available in the full report of 2016 EDHS [3].
Measurement of variables
The 2016 EDHS used five questionnaires, of which women's poll was used to collect data about women and child health characteristics. From women's Individual Record (IR) file, socio-demographic and reproductive traits were extracted from the most substantial dataset. The Perceived barriers of health care access were the response variable. In contrast, age, residence, wealth (economic) status, level of education, marital status, working status, health insurance coverage, and reproductive characteristics such as contraceptive use and intention, place of delivery, ANC follow up, and pregnancy during data collection were independent variables.
Each woman was interviewed to rate the difficulties of accessing health care based on obtaining money, health facility's distance, permission to consult the doctor, and not wanting to go alone. Women reported at least one challenge of health care access (money, distance, companionship, and permission) considered as having perceived barriers of health care access, which coded as "1". On the other hand, if a woman didn't report challenges of the obstacles mentioned above, like obtaining money, distance, companionship, and permission, it was considered no perceived barrier to health care access, coded as “0” [3, 12].
Data processing and analysis
The analysis of this study done by using STATA version 14.1 software. Summary measures such as median with IQR, frequencies with percentages computed, tables, figures, and text used to present the results. We checked for the presence of correlation among observations within clusters (enumeration areas), and the result showed that there was a within-cluster correlation (ICC=0.40) which indicated that there is a correlation among observations among cluster level. The generalized estimating equation (GEE) model was fitted to identify factors associated with the perceived barriers of health care access among reproductive-age women [25]. The generalized estimating equation fitted with a logit link function and binomial family and working correlation structures (independent, exchangeable, unstructured, and autoregressive), was compared for the smallest standard error difference between robust and model-based standard errors. Then, the exchangeable correlation structure showed the lowest standard error difference and selected for this study to handle within correlation. Crude and Adjusted odds ratio with a 95% confidence interval (CI) computed to assess the strength of association between independent and outcome variables. All figures presented in the result section used weighted value unless specified otherwise.
Model comparison
In this study, we fitted two models GEE which is a marginal model that considers correlation among clusters and a conventional logistic regression model with a robust standard error that also controlled for within-cluster correlations. To select the best-fitted model, we used QIC for model comparison and the result as follows.
Table 1: Model comparison
Types of the model fitted
|
QIC
|
GEE
|
18301
|
Conventional logistic
|
18441
|
Since GEE had the smallest QIC it better fits the data than the conventional logistic regression model.