Determinants of institutional delivery service utilization in Ethiopia: a population based cross sectional study

Utilization of institutional delivery was the outcome variable of the current study. At the time of survey, women were asked whether they were delivered at health facility or not. We created a binary response variable and coded as 1 for institutional delivery, 0 otherwise. Independent variables considered in this study were selected based on existing literature. These include socio-demographic factors (age of the women at birth, place of residence, distance to health facility, women’s education levels, sex of household head, wealth index, religion, employment status), and pregnancy related factors (antenatal care visits, parity, birth type, child size, desire of pregnancy).

Descriptive statistics were performed to describe and summarize the data. In the bivariate analysis, we conducted cross-tabulation between each independent variable and place of delivery. Chi-square test and p-values were used to test for the significance of each factor. Further analyses were carried out to study the relationship between the independent variables and place of delivery in a multivariate setup. The impacts of collinearity among the variables were detected using VIF and tolerance. The VIF values of all independent variables in the model are below 5, which shows there is no multicollinearity problem symptoms. To properly account the hierarchical nature of the data and to avoid possible under-estimation of parameters from a single-level model, we fitted multilevel model [17]. In the current study, we consider region of residence and wealth index as the level-2 variable under which the respondents are nested. This approach extends from single level logistic regression model by including random effects from the model. Three models namely model 1 a random intercept model where the model is fitted with only the intercept, model 2 with only individual-level factors and model 3 with the effects of individual-level and region level factors were estimated. Hence, the log of the probability of utilization of institutional delivery was modeled using two-level multilevel model as follows: where π_ij is probability of utilization of institutional delivery and 1- π_ij is probability of home delivery. The first part of the equation, \( {\beta}_0+{\sum}_{k=1}^{10}{\beta}_k{x}_{kij} \), is called the fixed part of the model and the second part \( {u}_{0j}+{\sum}_{p=1}^2{u}_{pj}{x}_{pij} \) is called the random part. The distribution of u_0j is normal with mean 0 and variance \( {\sigma}_{\mathrm{u}0}^2 \) and also the distribution of regional effect variables u_pj is normal with mean 0 and \( \mathrm{variance}{\sigma}_{up}^2 \). The intra-class correlation coefficient (ICC) measures the proportion of variance in the outcome explained by the grouping structure. It can be calculated as: \( \mathrm{ICC}=\frac{\sigma_{u0}^2}{\sigma_{u0}^2+{\sigma}_e^2} \) where, \( {\sigma}_e^2 \) is variance of individual level units which is constant as \( \raisebox{1ex}{${\uppi}^2$}\!\left/ \!\raisebox{-1ex}{$3$}\right.\approx 3.29 \) in logistic regression. Statistical significance was considered at p < 0.05 and the strength of statistical association was assessed by adjusted odds ratios (AOR) with 95% confidence intervals.

Of the 7174 women included in this study, 36.3% of the respondents had visited health facilities for ANC at least four times during their last pregnancy. About 66.9% of women who had at least four ANC visits were delivered at health institution. Higher proportion (78.8%) of women are resides in rural areas, and only one-fourth (26.9%) of them delivered their most recent birth at a health facility. Similarly, large proportion of women had no formal education (60.6%). Only 4.4% of the women had attended higher level of education. More than half (53.2%) of women resides in households with poor wealth index. About 93.6% of women reported that their most recent pregnancy was intended. About 43.3% of the women were unemployed. Distance from their home to health facility was a big challenge for more than 56% of the respondents, and of which, only 37% of women were discharge their recent birth in health facility. Detail descriptions of the study participants is presented in (Table 1).

In this study, about 38.9% women delivered their most recent birth in a health institution. Among those who delivered at health institution, large proportions of women (17.9%) were in Tigray region followed by Addis Ababa city (12.9%) (Fig. 1). Women from the Afar region had the lowest institutional delivery services (3.2%) (Table 1).

In the multivariable analysis, the fitted models along with the estimated effects and their standard errors are presented in (Table 2). As shown in the empty model, 34% of the total variance in the odds of women delivered at health institution was accounted by grouping the women with respect to their geographical regions (ICC = 0.34). The total variance accounted by the between region in the odds of giving birth at health institution in intercept only model (model 2) and random coefficient model (model 3) were 9.2% (ICC = 0.092) and 11% (ICC = 0.11) respectively (Table 2).

In addition, the between-region variance \( {\hat{\sigma}}_{0j}^2 \) was 1.68. To examine the hypothesis that whether or not this between- regional variance would have statistically significant effect on utilization of institutional delivery, a 50:50 mixture chi-square test was employed. Because the constrained variance component test lies on the boundary of the parameter space (variance is not expected to go below zero), the likelihood ratio test statistic converges to a 50:50 mixture of chi-square distributions with 0 and 1 degree of freedom given as 0.5\( {\chi}_0^2 \) + 0.5\( {\chi}_1^2 \) [18, 19].

Therefore, to see whether geographical region in Ethiopia an effect in the utilization of institutional delivery care service had, the deviance-based chi-square (Table 2) is 1373.6 with average p-value < 0.0001. This indicates that there exists a statistically significant variation across regions for the likelihood of institutional delivery. Accordingly, the random coefficient model (model 3) was considered for predicting women’s decision about place of delivery. The random coefficient model showed that the random effects of residence and wealth index to deliver in health facility vary across regions of Ethiopia. The variance of intercept in the random slope model is 0.38, which is still large and statistically significant. Thus, there remains some regional level variance unaccounted for in the model. The between-regions variance in the effect of place of residence on institutional delivery estimated as 0.24. This suggests that the effect of place of residence may be justified in constructing the effect to be random.

Details of the effect sizes of factors on the odds of institutional delivery are shown in Table 2. The frequency of ANC visits was significantly associated with place of delivery. After adjusting other variables in the model, the odds of giving birth at health facility was 3 times (AOR = 3.12, 95% CI: 2.73–3.56) higher among women who had recommended four or more visits compared with those who had less than four ANC visits. The odds of delivering at health facility was higher for women residing in urban (AOR = 7.18, 95% CI: 5.10–10.12) compared rural areas. The odds of delivering at health facility among women who attend primary school was 1.77 times (AOR = 1.77, 95% CI: 1.49–2.10), secondary school was 3.79 times (AOR = 3.79, 95% CI: 2.72–5.30) and higher level was 5.86 times (AOR = 5.86, 95% CI: 3.25–10.58) higher compared to women who had no formal education, respectively.

Increasing in parity was negatively associated with the odds of delivering at a health facility. Compared to prim parous women, the odds of institutional delivery were 48 and 40% lower for women who had five or more and two to four previous births, respectively. Compared to women residing in poor households, the odds of institutional delivery were higher for women residing in with rich (AOR = 2.39, 95% CI: 1.86–3.06), and middle (AOR = 1.66, 95% CI: 1.36–2.03) wealth index households. Distance from the health facility was positively associated to the odds delivering at health facility. The women who live near to health facility were 1.13 (AOR = 1.13, 95% CI: 1.06, 1.22) times higher chance of deliver at heath facility than those living far from health facility.

This study was population-based study, which involved large sample data. Multiple confounding variables were assessed using advance methodology. The cross-sectional nature of the study precludes drawing conclusions about the temporal nature of the association between the independent variables and the outcome of interest. Other limitations include recall bias and unmeasured confounders such as distance to health facility that was not captured in the 2016 EDHS.