The Ethiopian DHS was conducted in all regions of Ethiopia, Ethiopia has nine geographical regions and two administrative cities [12]. The data for this study was extracted from the 2016 EDHS which were collected from January 18, 2016, to June 27, 2016. It is the fourth nationally large scale dataset of DHS that is conducted by the central statistical agency (CSA)[13].The study population was all mothers of the reproductive age group who had given birth in the last 5- years before the EDHS 2016.
Measurements
Protection of last live birth against Neonatal Tetanus was the dependent variable and was dichotomized in to “protected” or “not protected”. Protected at birth was measured when mothers during pregnancy had received >2 TT dose (protected at birth) or had received <2 TT doses (not protected at birth). The independent variables includes individual factors (socio-demography variables of mother and husband, parity, ANC visit, place of birth, mode of birth), community factors (place of residence, region, community media exposure, community poverty). All region in Ethiopia includes Tigray, Afar, Amhara, Oromiya, Somali, Benishangul-Gumuz, Southern Nations, Nationalities and People’s Region (SNNPR), Gambela, Harari, Addis Ababa, and Dire Dawa. Community media exposure was categorized as exposed if the proportion of women in the community exposed to media was >=19.35% and categorized as not exposed if the proportion was 0-19.35%. Community poverty status was defined as the proportion of poor or poorest mothers within the cluster and aggregated to show overall poverty status within the cluster. This was classified as high if the proportion of women protecting last live birth against neonatal tetanus in a community was >=25% and as low if the proportion was 0-25%. Community-women education was defined as the proportion of mothers who attended primary/secondary/higher education within the cluster. It was categorized as higher or lower according to the national median value. It was classified as high if the proportion of women in a community attending at least primary-school and above was >=33.3% and as low if the proportion was 0-33.3%.
Media exposure: Based on their exposure status to radio or television, two categories were created: no exposure to either media and exposed to either media. Community poverty status: It is defined as the proportion of poor or poorest mothers within the cluster. Within the cluster proportion of poor or poorest were aggregated and show over all poverty status within the cluster. There were two categories for this variable with reference to the national median value; higher proportion of poor/poorest mothers and lower proportion of poor or poorest mothers within the cluster. This was classified as high if the proportion of women protecting last live birth against neonatal tetanus in a community was >=25% and as low if the proportion was 0-25%.
Statistical Analysis
The analysis was done using STATA version 14. Frequencies and percentages were used to describe the categorical variables. Cross-tabulation was also performed between all explanatory variables and the outcome variable. To assess the factors associated with the dependent and independent variables a multilevel binary logistic modeling was used by account the hierarchical nature of the EDHS data. The multilevel model involves two levels (individuals nested in communities). The explanatory variables with a p-value of <0.25 in this binary model were entered into the multivariable regression for adjustment [14]. Those variables with a p-value of < 0.05 in the multivariable multilevel regression model were declared statistically significant. The measures of association of the individual and community level factors with the protection of last live birth against neonatal tetanus were reported using the odds ratio with their respective 95% confidence interval (CI).
Four models (model I-IV) were fitted. Model I (null model) was run to test the inter-group (community) variability on neonatal tetanus and to decide whether the data is fit for multilevel modeling or not. Intra-class Correlation Coefficient (ICC) was calculated– the percentage of variability explained by the upper level (community). Model II includes individual-level factors only. Model III includes community-level factors only. Model IV includes the mixed model with both individual and community level factors. The model has a fixed or deterministic part and the random part. Proportional Change in Variance (PCV) and ICC were calculated [15] and compared between each models.
Parameter estimation
The Maximum Likelihood estimation method was used to estimate the parameters. Whereas the random-effects (measures of variation) were reported as an ICC which is the proportion of community-level variance as compared to the total variance and PCV express the change in variance between the null model (Model I) and the consecutive models. Those Akaike Information Criteria (AIC) was used to compare and select the model that best fits the data [16].