Background: Postnatal care (PNC) services are an essential intervention for improving maternal and child health. In Ethiopia, PNC service has been poorly implemented, despite the governments and partners’ attempt to improve maternal and child health service utilization. Moreover, many literatures identified that women with no education are significantly underutilized the PNC services. Thus, this study aimed to assess the PNC service uptake among women at high risk for underutilization of PNC services and to identify the individual and community level determinants of PNC services uptake in Ethiopia using the positive deviance approach. Methods: Data from the Ethiopia Demographic and Health Survey 2016 were used. A total of 2417 deviant women (women with no education) were identified through a two-stage stratified sampling technique and included in this analysis. A multilevel mixed-effect binary logistic regression analysis was computed to identify the individual and community-level determinants of PNC services uptake among deviant women. In the final model, a p-value of less than 0.05 and adjusted odds ratio (AOR) with 95% confidence interval (CI) were used to declare statistically significant determinants of PNC services uptake. Results: In this analysis, the uptake of PNC service among deviant women was 5.8% [95% CI: 4.9–6.8]. Working in the agriculture (AOR = 2.15, 95% CI: 1.13–3.52), being Orthodox religion follower (AOR = 2.56, 95% CI: 1.42–4.57), living in the highest wealth quantile (AOR = 2.22, 95% CI: 1.25–3.91) were the individual level determinants, whereas residing in the city administration (AOR: 3.17, 95% CI: 1.15–8.71), and living closer to health facility (AOR: 1.57, 95% CI: 1.03–2.39) were the community level determinants. Conclusion: The study highlighted a better PNC service uptake among deviant women who are working in the agriculture, follows orthodox religion, lives in highest household wealth status, resides in city administration, and living closer to the health facility. The positive deviance approach provides evidences for health policy makers and program implementers to improve health behavior in specific target population, and ultimately to bring better maternal and child health outcomes, despite acknowledged adverse risk profile. Such strategy and knowledge could facilitate targeted efforts aimed at achieving national goals of maternal and newborn mortality reduction in the country.
The study used the EDHS 2016 data, a nationally representative household survey that has been implemented by the Central Statistical Agency (CSA) of Ethiopia every 5 years [28]. Ethiopia is a home country of an estimated 114 million population (CSA 2015). Administratively, the country is divided into nine regions (Tigray, Afar, Amhara, Oromia, Benishangul, Gambela, South Nation Nationalities and Peoples’ Region (SNNPR), Harari and Somali) and two City Administrations (Addis Ababa and Dire-Dawa). Those nine regions can be divided in to developed regions (Tigray, Amhara, Oromia, South Nation Nationalities and Peoples’ Region (SNNPR), and Harari), and emerging regions (Afar, Benishangul, Gambela, and Somali). The 2016 EDHS used the 2007 Ethiopian population and housing census as a sampling frame, which was conducted by the CSA of Ethiopia and a complete list of 84,915 enumeration areas (EAs) were used in the census. The 2016 EDHS sample was stratified in two stages. A sample of EAs were selected from each stratum, independently. Then a total of 645 EAs were selected with probability proportional to the EA size, and each sampling stratum was selected from the given samples. The total residential households in the EA were the EA size, and a household listing operation was implemented. Then, the resulting lists of households were used as the sampling frame for selecting households in the second stage. Accordingly, all women aged 15–49 years who are regular members of the selected households were eligible for the survey. Finally, from a total of 4081 women identified from the EDHS 2016, 2417 deviant women were included in this analysis, and data were extracted from the datasets using STATA version 14 software. Variables at the individual and community-level were also extracted and further analyzed. We used the Anderson’s behavioral model of health service use [29] and other related studies [30, 31] to identify the positive deviant for PNC services uptake. Accordingly, education is the major determinant of health services utilization. We selected women who had no education as a sub-group with a very low likelihood of PNC services utilization, as education was the strongest predictor of PNC after adjusting for the other risk factors associated with PNC in this population. Positive deviant women were those who had no education but had an adequate uptake of PNC services. Finally, in the analysis, we compared the characteristics of the PD women to those of their counterparts. Due to significant variations by clusters in the overall use of PNC and also the individual and household level data were nested under the community level data, the analysis was stratified by individual and community level. Uptake of PNC services among deviant women was the dependent variable. The uptake was assessed when a woman received PNC services within 2 months after delivery, irrespective of their place of delivery. The information on the uptake of PNC services for their recent birth was assessed based on the women’s verbal responses during the survey. Accordingly, it was categorized as “yes” if a woman received at least one PNC visit, otherwise “no”. Individual-level variables; socio-demographic and economic variables (age, occupational status, religion, marital status, age at first birth, desire for child, household wealth status) were included in this analysis. On the other hand, place of residence, region, living closer to health facility, and media exposure were the community-level variables. Household wealth status was assessed using the asset index based on data from the entire sample on separate scores prepared for rural and urban households, and combined to produce a single asset index for all households and ranked into three (lowest, middle, and highest). The difficulty of getting health services was assessed by the question “living closer to health facility” and the responses were categorized as “yes” or “no”. Media exposure was assessed based on whether people had access to read newsletters, listen to the radio, and watch TV. Accordingly, if they have access to all three media (newsletter, radio, and TV) at least once a week, we categorized them as “yes”, otherwise “no”. The data were extracted, cleaned, re-coded, and analyzed using STATA version 17. The data were weighted using sampling weight during the statistical analysis to adjust for unequal probability of selection due to the sampling design used in DHS data. Tables and narrations were used to present the descriptive statistics. Since the DHS data are hierarchical (individual were nested within communities), a two-level binary logistic regression model was fitted to estimate the effect of both individual and community-level variables on PNC services uptake [32]. In this multilevel analysis, we fitted four models; i) Model 0: an empty (null) model without any explanatory variables, ii) Model 1: a model with individual-level variables, iii) Model 2: a model with community-level variables, and iv) Model 3: a model with both individual and community -level variables. In the survey, the individual and household level data were nested under the community level data, for model comparison, Intra-class Correlation Coefficient and deviance (− 2* log likelihood ratio) were used. Accordingly, a model with lowest deviance was chosen. The variation between clusters was assessed by computing Intra-lass Correlation Coefficient (ICC) [33]. The ICC is the proportion of variance explained by the grouping structure in the population which: ICC=; Where, the standard logit distribution has a variance of,, indicates the cluster variance. The ICC greater than 5% is eligible for multilevel analysis and in our analysis, the ICC was 22.5%. A mixed effect multilevel binary logistic regression analysis was done. A low deviance value was used to estimate the model goodness of fit by comparing the full model with the preceding three models. Finally, a p-value of less than 0.05 and an adjusted odds ratio (AOR) with 95% confidence interval (CI) were used to declare statistically significant factors associated with PNC services uptake among deviant women.
N/A