Study design, area and period
A community based cross sectional study was conducted from October 22 to November 30, 2018 in Gondar city administration, northwest, Ethiopia. The city administration had an estimated total population of 390, 644 of which 12,149 were under one year of age. Gondar city administration has a total of 24 Kebeles (13 urban and 11 rural). In addition, the city administration has a total of 23 public health facilities (one comprehensive specialized hospital, eight health centers and 14 health posts (39).
Source and study Population
The source population were all children aged 12 to 23 months with a history of routine vaccination in Gondar city administration. The study population were those children aged 12 to 23 months with their caregivers living in the eligible households of the selected kebeles and included for this particular study.
Inclusion and exclusion criteria
Households with at least one live child aged 12 to 23 months and who resided in the study area for at least 6 months prior to the study period were eligible for inclusion in this study. Those children included in this study had a history of routine vaccination. Children having history of vaccination from vaccination campaigns only were excluded.
Sample size calculation
The required sample size was calculated for both completeness and timeliness of vaccination using single population proportion formula by considering the following assumptions:
- Using proportion of 58.7 % for full vaccination [Pilot study], 95% confidence level, 5% margin of error, design effect of 2 and non-response rate of 10%, the sample was 821.
- Using proportion of 37.1% for on time-full vaccination [Pilot study], 95% confidence level, 5% margin of error, design effect of 2 and non-response rate of 10%, the sample was 395.
Hence, we used a sample size of 821 for this particular study.
Sampling procedure
Two stages sampling technique was used. During the first stage, from the 24 kebeles (13 urban and 11 rural) 40% of the kebeles were considered to be included in this study. From the total kebeles in the city administration, five urban and five rural kebeles were selected proportionally from each stratum using simple random sampling technique.
In the second stage of sampling, at each selected kebele, individual households were selected using systematic random sampling technique. Children in the selected households were further selected. If there were two or more children in the same household, lottery method was used to select one child per household. When there was no eligible child in the selected household, the next household was considered in the study.
Study variables measurement
Dependent variable: On-time full vaccination
Independent variables: Socio-demographic characteristics of the caregiver, health service related characteristics of the mother and contextual factors at community level were considered as independent variables for this study.
Full vaccination was defined as the child vaccination status once an infant has received all recommended vaccines included in the national schedule: a dose of Bacille Calmette Guérin (BCG); three doses of Oral Polio Vaccine (OPV); three doses each of Penta-valent and Pneumococcal Conjugate Vaccine (PCV); one dose of Inactivated Polio Vaccine (IPV); two doses of rotavirus and one dose of measles vaccines by the age of 12 months (5,22,26,40,41). On-time vaccination for specific vaccines was defined as vaccine dose administered within 4 days prior (31,42–45) and within 4 weeks after the recommended age specified in the national immunization schedule (12,30,31,37,43–49). On-time full vaccination was also defined as all vaccine doses administered within 4 days prior (31,42–45) and within 4 weeks after the recommended age specified in the national immunization schedule. Otherwise, it was not considered as on-time full vaccination if at least one vaccine dose was given early, late or missed at all (30,31,43,44,48,50–53).
The household socio-economic status was created by principal components analysis (PCA), including variables on asset ownership, housing characteristics and ownership of animals and farming. This was done for rural and urban households separately (54). Having this, rural and urban households PCA loading scores were merged for household wealth index classification using quintiles. Finally, the merged scores for urban and rural were divided into three quintiles as poor, middle and rich households.
Data collection tools and procedures
Data collection instrument was adapted from EDHS (18). Face and content validity has been ensured by a group of six experts. Accordingly, the applicability of the data collection tools and procedures were checked and revised as necessary.
Pilot study was also done out of the study area (in four kebele’s of Bahirdar city administration) before the actual data collection with a sample size of 100. The results of the pilot study were used to determine the minimum sample size for the actual study. The reliability of the data collection instrument was assessed using Cronbach’s alpha (α). From the pilot study the internal consistency estimate for the full vaccination scale was found to be 0.87.
Interviewer-administered data collection instrument was used to collect socio-demographic characteristics, health service related factors and vaccination status of the children. Eight data collectors and two supervisors were recruited for the data collection. Vaccination status and age at vaccination were confirmed by checking the vaccination card kept by caregivers or from the health facility expanded program on immunization (EPI) registers. For children with a vaccination card, the interviewer copied dates of any recorded vaccination on to the data collection instrument. For children without a vaccination card, their vaccination status was verified from the health facility EPI registers.
Data processing, analysis and parameter estimation methods
Descriptive statistics
The data were entered into EPI-data version 3.1 software and transferred to STATA version 14 software for analysis. Prior to the commencement of the analysis data cleaning, labeling, coding and recoding were done for all variables. Frequency and percentages were used to report categorical variables.
Bivariable and multivariable multilevel regression analysis
At the bivariable multilevel regression analysis, the effect of each individual and community level predictor variables on the outcome variable were checked at significance level of 0.2 (55). Variables which were significant at the bivariable multilevel logistic regression analysis were considered as candidates for the individual and community level model adjustments. Finally, a significance level of 0.05 was considered for the multivariable multilevel regression models.
Model specification
This study applied binary logistic multilevel analysis techniques in order to account for the clustering nature of the data and the binary response of the outcome variable. For the bivariable and multivariable multilevel logistic regression analysis the STATA syntax xtmelogit was used. Accordingly, four models containing variables of interest were fitted.
Model-I: was the null model, used to check the variability among the communities without inserting any variable. It’s the first step used to provide evidence whether the data has a justifiable evidence to assess the random effects at the community level. Model-II: was a multivariable model used to adjust individual level variables which were significant at the bivariable multilevel regression analysis. Hence, independent variables which were significant in Model-II were considered as candidates of the final model. Model-III: was also a multivariable model which was used to adjust community level variables which were significant at the bivariable multilevel regression analysis. Community level independent variables which were statistically significant in Model-III were included in the final model. Model-IV: was a multivariable multilevel regression analysis model used to adjust the outcome variable against independent variables which were statistically significant either at Model-II or Model-III. Stepwise model building technique was used for all models.
Parameter estimation methods
The measures of association (fixed-effects) estimates the association between the likelihood of children to be fully vaccinated on-time and the predictor variables expressed as Adjusted Odds Ratio (AOR) with their 95 % Confidence Intervals (CIs). The measures of variation (random-effects) were reported as intra class correlation coefficient (ICC) which is the percentage of variance explained by the community level variables. Proportional Change in Variance (PCV), expresses the change in the community level variance between Model-I (empty model) and the consecutive models (Model-II, III and IV) (56).
Multicollinearity and interaction effect
The presence of multicollinearity was checked among independent variables using Variance Inflation Factor (VIF) at cut off point of 10 (58). Similarly, interaction terms between community and individual level variables were tested.
Comparison of models and model fit statistics
Akakie Information Criterion (AIC) and log likelihood were used to compare the models. The AIC and the log likelihood values for each subsequent models were compared and the model with the lowest value was considered to be the better model (57). Finally, Hosmer-Lemeshow goodness of fit test was used to estimate the goodness of fit of the adjusted final model (55).
Ethical considerations
Ethical approval was obtained from the University of Gondar Ethical Review Board (IRB) before the commencement of the study. In addition, study permission was sought at all levels of local governmental health administrations. Informed written consent was obtained from each of the caregivers of children for their participation in the study and to access their child vaccination data from health facilities. In the meantime, study participants were informed to withdraw and discontinue participation at any time if they felt discomfort. Moreover, confidentiality assurance was provided to study participants on the information provided by them. Information that was collected for this study was also secured and protected from unauthorized access. At last, official permission was requested from the health facilities to access the child vaccination records from EPI registers.