Multilevel analysis of the predictors of completion of the continuum of maternity care in Ethiopia; using the recent 2019 Ethiopia mini demographic and health survey

Secondary data analysis was employed based on the 2019 EmDHS data, which were collected from 21^st March/2019 to 28^th June/2019. The EDHSs are a nationally representative cross-sectional survey done in nine regions and two city administrations every five years. EmDHS is also a survey conducted in between the standard EDHS (two to three years). Two EmDHSs in 2014 and 2019 have been conducted. For each of the survey, stratified two-stage sampling has been used. In the first stage, stratification was done by region, and then each region was stratified as urban and rural. For 2019 EmDHS data, a total of 305 (94 urban, and 211 rural) enumeration areas (EAs) were selected using probability proportional to EA size in the first stage. In the second stage, households were selected proportionally from each EA by using a systematic sampling method. The detailed method of data collection was accessed at the DHS database [40].

The data were used from the Measure Demographic and Health Surveys (DHS) website (http://​www.​dhsprogram.​com/​) after permission was taken through an online request by explaining the objective of the study.

All women reproductive age women (15 - 49 years) who gave birth in the five years preceding the survey and who had at least one ANC visits for their last child all over Ethiopia were the source population, whereas women who gave birth in the five years preceding the survey and who had at least one ANC visit for their last child and lived in the selected EAs were the study populations.

All reproductive-age women who gave birth in the five years preceding the survey and found in the selected clusters at least one night before the data collection period were included, whereas, women who gave birth and had no ANC visit were excluded. Finally, a total of 2,905 weighted samples of reproductive-age women were included in this study.

The data were extracted from the birth record (BR) file data set using SPSS version 24 software and further analyzed using STATA version 15. By using sample weight the data were weighted for probability sampling and non-response to restore the representativeness of the survey and get reliable statistical estimates. Data editing, cleaning, and coding were done. Descriptive statistics were done and presented by tables. Socio-demographic and other profiles of the study participants were compared using the chi-square test. In this study, two levels of data hierarchy were considered due to the sampling technique applied in EDHS (Multistage stratified cluster sampling). Level one unit was the individual pregnant woman in the households and level two units were enumeration areas. Level one (reproductive-age women in the households) was nested within units at the next higher level (enumeration areas). The outcome variable was represented by \({Y}_{ij}=\left\{\begin{array}{c}Completion of the continuum of care\\ Drop out from the continuum of care\end{array}\right.\), the category is dichotomous. Therefore, the multilevel mixed-effect logistic regression model was fitted to identify the predictors influencing the completion of maternity services along the continuum of care pathway at each level (individual level and community level). For the multilevel logistic regression four models were fitted. The first model (a model without covariate) was fitted to determine the extent of cluster variation in the completion of maternity services along the continuum of the care pathway. The second model was fitted with individual-level factors alone. The third model was fitted with community-level variables. Finally, the fourth model was fitted with both individual and community-level factors. Both bivariable and multivariable analyses were employed. Variables that have a P-value of ≤ 0.25 in the bivariable two-level binary logistic regression were candidates for multivariable multilevel logistic regression analysis. Then Variables in multilevel multivariable logistic regression were declared to be statistically significant at a P-value of < 0.05. The presence of Multicollinearity among covariates were checked using the variance inflation factor (VIF), and not present. The fitted models were compared based on Akaike’s Information Criteria (AIC) and a model with a small AIC value was nominated and all interpretations and inferences were made based on this model. The random-effects measure the variation of the completion of the continuum of a maternity care pathway across clusters (EAs) and are determined using the Intraclass correlation coefficient (ICC), median odds ratio (MOR), and proportional change in variance (PCV) statistics. The ICC determines the variation within-cluster and between-cluster differences. The PCV determines the total variation of early initiation of ANC at the individual- and community-level factors in each model. The MOR measures the MOR of completion of the continuum of care at the high-risk cluster (clusters having high dropout rate from the continuum of care) and low-risk cluster (clusters having a high prevalence of completion of the continuum of care) when we select randomly two reproductive-age women during data collection from two clusters. The formulas used to tabulate these three measurements are as follows;

\(\mathrm{ICC}={\mathrm{V}}_{\mathrm{i}}/\left({\mathrm{V}}_{\mathrm{i}}+{\uppi }^{2} /3\right) \sim \frac{Vi}{Vi+3.29}\), where V_i = between cluster (community) variances and π2 /3 = within-cluster (community) variance [43].

\(\mathrm{PCV}=\frac{Vi-Vy}{Vi}\), where Vi = variances of the null model, where Vy=variance of the model with more terms [43].

\(MOR=exp \left[\sqrt{2\times Vz\times 0.6745}\right]\sim exp.\left[0.95\sqrt{Vz}\right]\) where Vz = variance at the community level [43].

In this study, a total of 2,905 weighted samples of participants have participated from all over the country. More than half (53.14%) of the study participants were in the age range of 25-34 years old. Of the total study participants, 2,036 (70.08%) were rural residents, 2,717 (93.51%) were married, 1,208 (41.59%) were Orthodox Christian by religion, 1,269 (43.67%) didn’t attend formal education, and 2,519 (86.71%) of the households were headed by males. The majority of the participants were from the Oromia regional state (37%) followed by the Amhara regional state (24.05%) (Table 2).

Of the total study participants, more than half (54.62%) of the households have had ≤ 5 family members and nearly 47% were multiparous. The majority (60.15%) of the women had been advised for obstetric danger signs during their last pregnancy. More than sixty percent of the woman have had a birth spacing of 24 months and above, and 62% of the women started their first antenatal care visits lately for the most recent pregnancy (Table 3).

In this study, the overall prevalence of completion of the continuum of maternity care was 12.9% (95%CI: 11.1 – 14.9%). Of the total respondents, more than half (58.1%) have had four and above ANC follow-up, and more than two-thirds (65.8%) were delivered at the health institutions. But, only 23.1% of the respondent had postnatal care after discharge/deliver at home within the recommended time period (Fig. 1).

Of the total women who have completed the continuum of care, 47% were in the age group of 25-34 years old, 93% were married, 56.5% were urban residents, 52.2% were multiparous, and 63% of women belonged to rich households. Based on birth interval, the completion of the continuum of care was higher among women who spaced for ≥ 24 months (87.2%). Among women who advised on danger signs during pregnancy, the completion of the continuum of care was 76.4%. Concerning the family size, households having less than six family sizes were experienced a higher prevalence of completion of the continuum of care (57.3%). Regarding the age of first birth, 64.9% of women had completed the continuum of maternity care (Table 4).

The Intraclass correlation coefficient (ICC) in the null model was (0.197), which means that 19.7% of the variability of completion of the continuum of maternity care was due to the differences between clusters or unobserved factors at the community level. This noted that the multilevel logistic regression model is better to estimate the completion of a continuum of care among reproductive-age women who gave birth in the last five years and had booked ANC visits than single-level logistic regression. The Akaike’s Information Criteria (AIC) is smallest at model 4 (AIC = 1532.4) as compared to random intercept only model or null model (AIC = 2217.2), a model with only individual-level factors (AIC = 1536.2), and model with only community-level factors (AIC = 2174.6). Therefore this model is the best-fitted model for the data because it has the smallest AIC as compared to the rest models. As a result, all interpretations and reports were made based on this model. In addition, the median odds ratio (MOR) in all models was greater than one noted that there is a variation in completion of the continuum of care between community levels (clusters). The value of MOR (2.31) in the null model depicts that there was a variation of completion of the continuum of care between clusters when we randomly select women from two clusters, women from clusters that have a high prevalence of completion of the continuum of care were 2.31 times more likely to complete continuum of care as compared to women from low prevalence of continuum of care cluster. The higher proportional change in variance (PCV) value in the fourth model (47.5%) depicted that about 47.5% of the variability of completion of the continuum of care was explained by both the individual-level and community-level factors (Table 5).

As noted in the final model (model 4) both individual and community-level factors were added for multilevel analysis, of which, maternal educational status, household wealth status, the timing of the first initiation of antenatal care visit, danger sign advised during pregnancy, residence, and region were significantly associated with completion of maternity care uptake in Ethiopia at P-value < 0.05 (Table 4). The odds of completion of the continuum of care for the maternity services were two times (AOR = 2.03: 95%CI; 1.14 - 3.61) more likely among women who attended higher educations as compared to those women who didn’t attend formal education. Women who belonged to the household wealth status of medium, and rich were 1.7 (AOR = 1.69: 95%CI; 1.07 - 2.66), and 2 (AOR = 2.05: 95%CI; 1.32, 3.17) times more likely to complete the continuum of maternity care as compared to women who belonged to poor wealth status, respectively. Women who had advised on danger signs of pregnancy during their ANC visits were 2 (AOR = 2.23: 95%CI; 1.61, 3.10) times more likely to complete the continuum of maternity care than their counterparts. Whereas, the odds of the completion of the continuum of maternity care was 34% (AOR = 0.66: 95%CI; 0.49, 0.89) less likely among women who attended their first ANC visits late as compared to women who attended their first ANC visits early. Likewise, the odds of completion of the continuum of maternity care was 33% (AOR = 0.67: 95%CI; 0.42 - 0.93) less likely among women who resided in rural areas as compared to their counterparts. Women who lived in the Afar and Gambella regional states were 64% (AOR = 0.36: 95%CI; 0.12 – 0.83), and 48% (AOR = 0.52: 95%CI; 0.19 – 0.95) less likely to complete the continuum of maternity care than women who lived in Addis Ababa, the capital city of Ethiopia (Table 6).

The study had many strengths, for instance, it used nationalily representative data of the Ethiopia demographic health survey with a large sample size, a high response rate, and high-quality data which could minimize sampling and measurement-related bias. Moreover, we have used an appropriate statistical approach, multilevel mixed-effect analysis, to estimate the cluster effect on completion of the continuum of maternity care. As a limitation, owing to the cross-sectional nature of the study, the exact cause-effect relationship between completion of the continuum of maternity care and its predictors doesn’t exist and recall bias might be introduced. The other limitation was the study failed to assess some important variables like distance to the health facility, pregnancy intention, and media exposure which may affect the completion of the continuum of care. The study also failed to assess the effect of utilization of one basic maternal health service on the next MCH service.