Determinants of postnatal care utilization in Ethiopia: A multilevel analysis

Background: The expansion of primary health care services in Ethiopia made basic health services available and accessible. The Last Ten Kilometers (L10K) project has strengthened the primary health care system through implementing innovative strategies to engage local communities to improve maternal and newborn health care behavior and practices in Amhara, Oromia, Southern Nations, Nationalities and Peoples [SNNP], and Tigray regions over a decade. Despite the efforts of the government and its partners to improve the use of maternal health services, the coverage of postnatal care is persistently low in the country. This study examined the individual and community level determinants for the persistently low uptake of postnatal care in the project areas. Methods: The study used a cross-sectional population-based survey that measured maternal and newborn health care practices among women who had live births in the last 12 months preceding the survey in Amhara, Oromia, SNNP, and Tigray regions. Multilevel random effects binary logistic regression analysis was used to assess the independent effects of community-and individual-level factors and moderating effects on the uptake of postnatal care. Results: This study identified region of residence, obstetric factors, and health service-related factors to be significant determinants for use of postnatal care. Obstetric factors include knowledge of obstetric danger signs (AOR: 1.30; 95% CI: 1.05-1.60), cesarean section mode of delivery (AOR: 1.96; 95% CI: 1.28-3.00), and institutional delivery (AOR: 10.29; 95% CI: 7.57-13.98). While the health service-related factors include attended family conversation during pregnancy (AOR: 1.48; 95% CI: 1.04-2.12), birth notification (AOR: 2.66; 95% CI: 2.15-3.29), home visits by community health workers (AOR: 1.98; 95% CI: 1.58-2.50), and being recognized as a model family (AOR: 1.27; 95% CI: 1.03-1.57). Conclusion: This study demonstrated that community-level interactions and promotive health services including antepartum home visits by community health workers, family conversation, birth notification, and model family, are important determinants to seek postnatal care. The findings also highlight the need for expansion of health facilities or design appropriate strategies to reach the disadvantaged communities. Program managers are recommended to strengthen community-based interventions to improve postnatal care utilization. Ethiopia has expanded primary health care services through expansion of the Health Extension Program (HEP) and expansion of health centers and promotion of early healthcare-seeking through community mobilization to reach most communities and households and provide preventive, promotive, and basic curative services. The promise of the HEP has been the graduation of model families where health extension workers (HEWs) train households to acquire the necessary knowledge, skills, and behavior change in health practices. When these households demonstrate practical changes in the use of health service programs: environmental health, personal hygiene, and serve as models in their community, then they graduate. To reach more communities, this strategy has been accelerated through the participatory engagement of model families that are early adopters of desirable health practices and have acceptance and credibility by their community and the women development army (WDA) group. The WDA network builds on the critical mass of model families and creates volunteer leaders that scale-up the dissemination of knowledge, innovation, and service utilization through social networks. They are actively engaged in promotion and prevention activities as well as social mobilization efforts to expand HEP deeper into communities and families and ensure community ownership [20]. The Last Ten Kilometers (L10K) project has been supporting in strengthening the HEP through the implementation of innovative strategies to engage local communities to improve high-impact reproductive, maternal, newborn and child health (RMNCH) care behavior and practices in 115 woredas (i.e., districts) in four of the most populous regions of Ethiopia (i.e., Amhara, Oromia, Southern Nations, Nationalities and Peoples [SNNP], and Tigray) since 2008. During the first funding cycle, i.e., between 2008 and 2015, the Project implemented innovative strategies to engage local communities to reach its objectives. Three community-based strategies—Community-Based Data for Decision-Making (CBDDM), family conversation, and birth notification—were implemented in the 115 woredas covering about 3070 kebeles, the lowest administrative unit. Community-based data for decision-making, introduced in July 2013, was used to identify pregnant women and to ensure they received antenatal, intrapartum, and postpartum care [21]. CBDDM fostered the kebeles to generate and use data to improve maternal and newborn health practices. The strategy identified underserved households and linked them with HEWs and kebele managers helping to address barriers in accessing maternal and newborn health services. Accordingly, HEWs were trained to support WDA team leaders to map 30 households in their catchment areas, to keep each household under surveillance, and to ensure the provision of maternal and newborn health services along the continuum of care. The surveillance system used images so that they could be maintained and updated by individuals with little or no education. HEWs collect data from WDA team leaders’ surveillance to help kebele leaders to identify and address barriers that make access to maternal and newborn health services difficult. The CBDDM strategy has resulted in improvements in institutional deliveries and newborn health care behaviors and practices. However, there was no evidence of any effect of the intervention on postnatal care within 2 days of childbirth [21]. Accordingly, alternative strategies, complementary to CBDDM, such as family conversation and birth notification, were designed to promote maternal and newborn health behaviors. Family conversation, on the other hand, is a forum conducted at the house of a pregnant woman with her family members and relatives who are expected to support her during her pregnancy, labor, delivery, and postpartum period. Family conversation session was introduced early in 2014 to promote birth preparedness and the use of early postpartum care and essential newborn care. Birth notification strategy was introduced by mid-2014 to promote early postnatal care. Since October 2015, L10K focused on institutionalizing the CBDDM strategy and strengthen the implementation of family conversations and birth notification strategies in the 115 woredas [21]. During the intervention period, L10K Project’s contribution has shown an increase in skilled delivery from 10 to 67% between 2011 and 2017. Nevertheless, despite the efforts and huge investments made to improve the use of maternal health services, only one-third of mothers received PNC within 48 h [22]. This study used a cross-sectional population-based design. The study population consists of women who had live births in the last 12 months preceding the survey. We used the data that was collected from the representative sample of 2724 women aged 15 to 49 years, from 298 Kebeles. The data were drawn from a cross-sectional household survey datasets representing the 115 woredas (where L10K operates) and were collected from October–November 2017. The survey employed a two-stage stratified cluster sampling method. The details of the design are described somewhere else [23]. The adequacy of the sample size, required to address this study’s objective, was checked using StatCalc for a double population formula for comparative cross-sectional study design considering different exposure variables including birth order, place of delivery, maternal education, and wealth quintile from the 2016 EDHS report [5]. The assumption made to calculate sample size was considering a 95% confidence level (Zα/2 = 1.96), and design effect (D = 2), power of 80%. Based on Anderson’s health-seeking behavior model [24], the researchers assumed first birth order, home delivery, no maternal education, and poorest wealth quintile as exposure and highest birth order/6+, facility delivery, higher education, and richest wealth quintile as non-exposure. Adding a 10% non-response rate, the maximum sample size obtained by birth order was 750. Accordingly, we ensured that the available sample was adequate to address this objective. Data were collected through interviews of eligible mothers aged 15–49 years. During the interview, information about household and socio-demographic characteristics of mothers, awareness, and experiences related to the women’s use of maternal health services, was collected from women with children in their first year of life. The data were collected using a structured interview questionnaire (Additional file 1) and archived using a web-based mHealth platform (i.e., SurveyCTO) using smartphones. The outcome variable of interest was, the proportion of women and their newborns who received postpartum care at the health facility or their home within 6 weeks of delivery. It is measured interview of women weather pre-discharge care provided for them and their newborns after 24-h of stay for whom delivered at the health facility and any postnatal check-up for mother’s and newborn’s health by a health care provider within 6 weeks of giving birth. The independent variables considered include individual and community level factors. The individual socio-demographic variables include maternal education, husband’s education, religion, marital status, household wealth index, and distance to health facility. Obstetric characteristics include mode of delivery, place of delivery, utilization of ANC services, knowledge of obstetric danger signs, and encountered obstetric complications (during pregnancy, childbirth, or postpartum), as well as program characteristics including community health workers (CHWs) home visit, model family, and birth notification, were considered. Administrative region and area of residence (clusters/kebeles) were considered as community-level random components. Model families are defined as those households who received training from HEWs, acquire the necessary knowledge, skills, and behavior in health practices and demonstrate practical changes in the use of health service programs and serve as models in their community. Family conversation is a forum conducted at the house of a pregnant woman with her family members and relatives who are expected to support her during pregnancy, labor, delivery, and the postpartum period. It creates an opportunity to discuss issues such as birth preparedness and essential newborn practices with all these family members together. It is measured by asking women regarding their attendance of at least one family conversations at home during their last pregnancy. While birth notification is a strategy introduced to promote early postnatal care. It is measured through interviews of women weather they took measures to inform the HEW about their childbirth immediately after delivery or not. A wealth index score was constructed for each household with the principal component analysis of the household’s possessions (electricity, watch, radio, television, mobile phone, telephone, refrigerator, table, chair, bed, electric stove, and kerosene lamp), and household characteristics (type of latrine, water source, floor, and wall material). Subsequently, households were ranked according to wealth score and then divided into five quintiles using Principal Component Analysis (PCA) method. Data were analyzed for both descriptive and inferential statistics using Stata version 15.1. Socio-demographic data were summarized by frequency tables and summary statistics. The researcher conducted an individual-level analysis to assess the association between the individual and contextual characteristics of the respondents with PNC use using Pearson’s Chi-squared statistics taking the complex survey nature into account. Post-stratification sampling weights were used to adjust the non-proportional allocation of the sample to the different regions and weighted analysis was used to ensure the representativeness of the survey estimates. Multilevel bivariate and multivariable analysis using cluster-level random effects logistic regression models were used to assess the independent effects of community factors and moderating effects on the association between individual variables and PNC. The models were adjusted for survey design and individual and contextual characteristics of the respondents. To assess the contribution of the respondents’ characteristics to the model, a two-level logistic regression model was applied (i.e., individuals (level 1) were nested within communities (level 2). In fitting models, iterative based backward elimination and forward selection methods were used to select explanatory variables. For the purpose, the independent variables were retained in each of the models if the p-value was less than 0.2. Accordingly, maternal education, household wealth index, distance to health facility, family currently recognized as model, family conversations during pregnancy, knew at least 3 obstetric danger signs, encountered obstetric complications, CHW (either by HEWs or WDAs) home visit during pregnancy, ANC visit, place of delivery, mode of delivery, and birth notification, were retained in the model. Both random-intercept and random coefficient logistic models were fitted to estimate associations between the individual and community variables and the likelihood of receiving PNC using xtlogit and xtmelogit Stata commands. The null model is fitted without the explanatory variable. The random-intercept logistic models were fitted to assess the influence of unobserved community-level characteristics on the overall variation in PNC utilization allowing the probability of receiving PNC to vary randomly across communities assuming the effects of individual characteristics are the same in each community, i.e. the coefficients of all explanatory variables are fixed across communities. While the random coefficient model was fitted for household wealth (mean-centered), allowing to vary across communities. Finally, we adjusted both individual and community variables and a cross-level interaction between region and distance to the health facility to see any evidence of effect modification of the association between distance to health facility and PNC utilization by region. The likelihood ratio test showed that the effect of wealth index variables significantly varied across communities indicating the random coefficient model provides a better fit than the random intercept model (p = .005). Moreover, the model fit statistics, AIC, reduced as compared to intercept models indicating that allowing wealth index to vary across communities increased the predictive ability of the multilevel models for PNC. Moreover, we noticed that the intercept models do not provide additional information. As such, we reported random coefficient models. Measures of associations were expressed as odds ratio (OR) and 95% confidence interval (CI). Important assumptions and fitness of such statistical models were checked using the standard procedures. The goodness-of-fit of the models was assessed using the global Wald’s statistics, the likelihood ratio test of the cluster-level random effects, and sensitivity of the quadrature approximation that was used to estimate the models.