Background: Most maternal and infant deaths occurred within the first month after birth. Nearly half of the maternal deaths and more than a million newborn deaths occurred within the first day of life but these were preventable through early initiation of postnatal care (PNC) services. However, the available evidence on the level of early initiation of PNC service utilization was not adequate to inform policy decisions. Therefore, this study aimed to assess time to early initiation of postnatal care and its predictors using the 2016 Ethiopian Demography and Health Survey (EDHS) datasets. Methods: Two-stage stratified cluster sampling technique by separating each region into urban and rural areas. A total weighted sample of 6364 women of the 2016 EDHS datasets who gave birth within 2 years preceding the survey was used. Time to early initiation of the PNC visit was estimated using the Kaplan-Meier (K-M) method. Shared frailty model with baseline distributions (Weibull, Gompertz, exponential, log-logistic, and lognormal) and frailty distributions (gamma and inverse Gaussian) were used by taking enumeration areas/clusters as a random effect for predictors of time to early initiation of PNC visit. The adjusted hazard ratio (AHR) with a 95% confidence interval (CI) and p-value less than 0.05 were used to declare the significant predictor variables for time to early initiation of the PNC service utilization. Results: The prevalence of women who utilized PNC services within 42 days was 13.27% (95% CI, 12.46, 14.13). Among these women, only 1.73% of them had got within the first 24 h of birth; 4.66% of them received within 48–72 h and 1.74% of them also had got within 7–14 days. Variables, such as parity (AHR = 1.61, 95% CI: 1.21, 2.15), media exposure (AHR = 1.42, 95% CI: 1.21, 1.68), place of delivery (AHR = 14.36, 95% CI: 11.76, 17.53), caesarean delivery (AHR = 2.17, 95% CI: 1.60, 2.95) and antenatal care visit (AHR = 2.07, 95% CI: 1.63, 2.63) had the higher hazard for PNC services utilization. On the other hand, women who faced with healthcare access problems (AHR = 0.74, 95% CI: 0.60, 0.87) had a lower hazard of PNC service utilization. Conclusion: The overall postnatal care service utilization among women in the survey was low, particularly within the first 24 h of delivery. Policy-makers and implementers should promote the utilization of antenatal care and institutional delivery using mass media to increase the continuum of maternity care. The government should also design a new approach to enhance the uptake of postnatal care services for poor households and to scale up the PNC services, including the different possibilities for women who give births at the health facilities and homes. Future researchers had better assess the capacity and accessibility of the local health systems, the level of decentralized decision making, common cultural practices, knowledge, attitude, and perception of mothers towards PNC service utilization.
Secondary data analysis was conducted based on the EDHS 2016. The EDHS used a stratified two-stage cluster sampling technique selected using the 2007 Population and Housing Census as a sampling frame. A total of 84,915 Enumeration Areas (EAs) were created in Ethiopia and stratification was done by dividing each of the nine regions into urban and rural areas and a total of 21 sampling strata were formed. In the first stage, 645 enumeration areas (202 in the urban area) were selected with proportional allocation to the size of the enumeration areas (EAs) with an independent selection of each sampling stratum. A total of 243 EAs that have less than 10 observations per cluster (a total of 1225 observations) were dropped. A total of 402 EAs were, therefore, included for analysis to get a reliable estimate. A minimum of 10 and maximum of 21 women or on average 15 women per EAs were selected using systematic sampling technique. A total weighted sample of 6364 women who gave birth within 2 years preceding the survey were included for this study (Fig. 1). The detailed sampling procedure was presented in the full EDHS 2016 report [14]. The sampling procedures for selecting the study participants in EDHS 2016 Women who gave birth within 2 years preceding the survey were considered in this study and those who had PNC visits within 42 days of birth were considered as a success while those who didn’t have a visit were treated as a failure. It is defined as the time of first PNC checkup within the first 42 days of birth. The time to first PNC visit was recorded in days if the women have a PNC visit within 42 days of birth. The event is binary form, coded as “1” if a woman had a PNC visit within 42 days of birth and “0” if the women didn’t have a PNC visit within the 42 days. The independent variables considered for this study were categorized as socio-demographic and economic variables (residence, region, religion, maternal education, husband education, maternal occupation, sex of household head, distance to the health facility, and wealth status), and obstetric related factors (the type of gestation, preceding birth interval, ANC visit, place of delivery, mode of delivery, parity, birth order). The data were weighted using sampling weight, primary sampling unit, and strata before any statistical analysis to restore the representativeness of the survey and to take into account the sampling design to get reliable statistical estimates. STATA version 14 software was used for the descriptive as well as for the frailty analysis. Because of the hierarchical structure of EDHS data, women are nested within a cluster and we expect that women within the same cluster may be more similar to each other than women in the rest of the country. This violates the assumption of the traditional regression model which is the independence of observations and equal variance across clusters. A total of 243 EAs that have less than 10 observations per cluster (a total of 1225 observations) were dropped from the analysis to balance the size of clusters and detect the random effect efficiently. The standard survival analysis models are applicable when the time to event data is independent but the EDHS was a cluster survey that has hierarchical nature and assumed to be correlated at the cluster level. The correlation could be due to unobserved cluster or EAs specific covariates and assumes that time to early initiation of PNC service of a mother is a function of measured variables and a random (frailty) on the baseline hazard to the unobserved cluster effect. Schoenfeld residual global test was applied to check the Proportional Hazard (PH) assumption, and it was violated with a p-value < 0.05 (Supplementary file 1). Parametric survival models were fitted since the PH assumption was violated. The EDHS data has a hierarchical structure and a frailty model (random effect survival model) was used to check whether there is clustering or not. The theta was significant at the null model (θ = 1.47, 95% CI: 1.26, 1.72). It indicates that there was unobserved heterogeneity or shared frailty in which women in one cluster were more likely to be correlated with women in the same cluster. Shared frailty model with baseline distributions (Weibull, Gompertz, Exponential, log-logistic, and lognormal) and frailty distributions (gamma and inverse Gaussian) were used by taking EAs of v001 as a random effect for predictors of time to early initiation of PNC visit among women who gave births. A Weibull gamma shared frailty model was the best-fitted model for this data since it has the smallest deviance and AIC values. Variables with a p-value less 0.20 in the uni-variable of the Weibull gamma shared frailty analysis were included in the multivariable analysis. The Hazard Ratio (HR) with 95% Confidence Interval (CI) and p-value less than 0.05 were used to declare the significant predictor variables for time to early initiation of the PNC service utilization. The model was formulated as: For time to event data, where i (1 ………., n) denotes the cluster, while j (1, ……, n) denotes the subjects (woman) within the cluster. The frailty, u i is a random positive quantity shared within groups, whereas hij (t│xij,ui) is the probability of women initiating PNC service at time t; h0(t) is the baseline hazard and Xij is a vector of covariates with the associated vector of fixed parameters β. Time to early initiation of the PNC visit was estimated using the Kaplan-Meier (K-M) method. The log-rank test was used to compare survival time between groups of categorical variables.