Low immunization coverage in Wonago district, southern Ethiopia: A community-based cross-sectional study

Introduction Immunization is a cost-effective intervention that prevented more than 5 million deaths worldwide from 2010 to 2015. Despite increased vaccination coverage over the past four decades in many African countries, including Ethiopia, universal coverage has not yet been reached. Only 39% of children aged 12–23 months received full vaccinations in Ethiopia, according to the 2016 Ethiopian Demographic Health Survey. This study aimed to evaluate immunization coverage and identify individual and community factors that explain incomplete vaccination coverage among children aged 6–36 months in the Wonago district of southern Ethiopia. Methods We conducted a community-based, cross-sectional study in three randomly selected kebeles in the Wonago district from June to July 2017. Our nested sample of 1,116 children aged 6–36 months included 923 child-mother pairs (level 1) within kebeles (level 2). We conducted multilevel regression analysis using STATA software. Results Among participants, 85.0% of children aged 12–36 months received at least one vaccine, and 52.4% had complete immunization coverage. After controlling for several individual and community variables, we identified six significant predictor variables for complete immunization: Older mothers’ age (AOR = 1.05, 95% CI: 1.00–1.09), higher utilization of antenatal care (AOR = 1.36, 95% CI: 1.14–1.62), one or more tetanus-toxoid vaccination during pregnancy (AOR = 2.64, 95% CI: 1.43–4.86), mothers knowing the age at which to complete child’s vaccinations (AOR = 2.00, 95% CI: 1.25–3.20), being a female (AOR = 0.64, 95% CI: 0.43–0.95), and child receiving vitamin A supplementation within the last 6 months (AOR = 2.79, 95% CI: 1.59–4.90). We observed a clustering effect at the individual and community levels with an intra-cluster correlation coefficient of 48.1%. Conclusions We found low immunization coverage among children in the Wonago district of southern Ethiopia, with significant differences across communities. Promoting maternal health care and community service could enhance immunization coverage. The institutional review board at the College of Medicine and Health Sciences of Hawassa University (reference number: HW/17/0668/15) and the regional ethical committee in Western Norway (reference number: 2016/1916/REK vest) provided ethical approval. Permission letters were obtained from Gedeo zone health department and Wonago district health office. Before the start of the study, community elders, health extension workers and kebele leaders were sensitized. Mothers or caretakers provided written and signed informed consent. Confidentiality was maintained and participants were informed that participation was voluntary and they had right to withdraw from the study at any time. We conducted community-based, cross-sectional study in Wonago district in southern Ethiopia from June 2017 to July 2017. Wonago is about 377 km south of Addis Ababa and 13 km south of Dilla, the capital of Gedeo. The district is 142 km2 and has 17 rural and 4 urban kebeles (the smallest administrative units) containing 29,227 households. Among the most densely populated areas in Ethiopia, it has 1,014 people per square kilometer. According to the 2007 census, the district’s population of 147,600 people included 22,140 (15%) children younger than five years [28, 29]. The majority of the population lives in the rural areas and agricultural is the dominant means of livelihood of Wonago district. Major cause of childhood illness is pneumonia and diarrhea. Wonago has 20 health posts, six governmental health center, two private clinics, and two drug stores. Expanded Program of Immunization (EPI) was all provided in health centers and health posts and supported by 34 outreach programs site with every week provision of immunization. Eligible participants included were all children aged 6 to 36 months and their mothers or guardians, who lived in Wonago for at least 6 months before data collection and who consented to participate in the study. We excluded those who had not lived in the study area for least 6 months prior to data collection. We calculated the sample size required for estimating immunization coverage using Open Epi software version 3.03 [30]. The calculation assumed a desired precision (sampling error) of 4% to get larger sample size with a 95% confidence interval (CI), a design effect of two to consider two stage sampling and to adjust the variance, and a 10% non-response rate. The anticipated proportion of full immunization coverage was 30.5% based on a study in Hosanna [20]. We thus calculated a sample size of 1,119 children aged 6 to 36 months old and their guardians. A two-stage sampling technique was employed. In the first stage, we used a random sampling lottery method to select 3 of 17 total kebeles using the Statistical Package for Social Science (SPSS) version 20 complex sample method [31], we then randomly selected 12 villages from the selected kebeles based on probability proportional to size (number of households). Before the survey, we conducted a census in selected kebeles to obtain socio-demographic, household status information and to identify eligible children. The sample is distributed to selected villages based on probability proportional to size (number of eligible children). The first household was identified by randomly from the center of the village. Once first household was identified the interviewer went to the next household with the mother of children age group of 6–36 months. Subsequent sampling of household was conducted from selected villages until the desired sample size was attained. When two children from the same household were identified, both children were selected. The outcome variable in this study was full (complete) immunization. Individual-level factors included were mother’s age, mother’s occupation, parity, religion, ethnicity, women’s education(Mothers education has 5 categories, no formal education, primary(1–8 grade), secondary(9–10 grade), preparatory (11–12 grade) and college or university), antenatal care (antenatal care defines as number of visit that mother get care during pregnancy), delivery place, post-natal care, sex of child, number of children younger than five years in the household, birth order, family planning use, household family size, attitude of mothers towards immunization(Mothers attitude towards vaccination was assessed by six attitude related questions and using a 2-point scale (agree and disagree), where 1 = positive perception and 0 = negative perception. The mean score was computed and dichotomized into positive and negative: if mothers reacted score below the mean, they were labeled as having a negative attitude; if mother reacted to at the mean and above the mean, they were labeled as having a positive attitude), presence of child vaccination card and wealth index. The wealth index was assessed to capture households’ socio-economic statuses and constructed using principal component analysis based on household asset and amenities. The generated wealth score was grouped into quartiles as a measure of socioeconomic status, with the first quartile representing the poorest group and the fourth quartile the richest. Community-level factors included were visits from healthcare workers, distance to healthcare facilities, access to health outreach (e.g. vitamin A supplementation) and families participating in food supporting programs(such as safety net program). Table 1 summarizes the infant immunization schedule of recommended vaccines, including Bacillus Calmette–Guérin; oral polio; diphtheria, pertussis, and tetanus; hepatitis B and Hemophilus influenzae type B; pneumococcal conjugate; rotavirus; and measles) in the study area [14]. The structured questionnaire was initially developed in English, translated into the local Gedeoffa language and then translated back to English to ensure consistency. Most questions were adopted from questionnaires in the Demographic and Health Survey of Ethiopia [5] and from literature reviews [15, 21, 23–25]. Before data collection the questionnaires reviewed by supervisors then we conducted pre-test on 56 children (6–36 months) in another kebele of Wonago district and that were not included in the study. Based on this pre-testing, we rephrased unclear questions, wording, and sequences. After data collection the data cleaned and checked for the consistency. Eight trained data collectors and two supervisors conducted the interviews. Immunization data were collected from vaccination cards (i.e., doses and types) and mothers’ or guardians’ verbal reports. We confirmed the information given by checking children for Bacillus Calmette–Guérin (BCG) scars on upper arm. The principal investigator checked the data for completeness, and errors were corrected accordingly. To control for recall bias, we used standardized questionnaires and trained data collectors in facilitating participant recall. We defined complete or full immunization among children aged 12–36 months as receiving one dose of BCG, three doses of polio, three doses of (Diphtheria, pertussis, tetanus, Hepatitis B and Hemophilus influenzae type B) and one dose of measles, as confirmed by immunization card or mother’s recall. We defined partial immunization as missing one or more of the recommended eight vaccines and children who were not vaccinated at recommended age. Children who never received any immunizations were classified as not vaccinated. We considered children younger than 11 months with complete vaccinations as completed immunization for age. We defined the age limit for immunization as nine months, by which each child should have had one dose of BCG, three doses of polio, three doses of pentavalent, and one dose of measles vaccinations. Immunization coverage by card was calculated based on card documentation only and excluded vaccinations confirmed by mothers’ recall. Immunization by card plus recall included card and verbal histories. An infant immunization card was yellow card that given to child when child starts vaccination and used for vaccination follow-up and monitoring. Data were double-entered using EpiData version 3.1. STATA 15 software (Stata Corp) was used for analyses. We compiled descriptive statistics, such as frequencies, percentages, means, and ranges. Cross-tabulation was used to show proportions of different categories of each characteristic, with respect to immunization status. A two-level logistic regression model was applied to analyze the hierarchical structure. Child-mother pairs (level 1) were nested within communities or villages (level 2). We used multilevel analysis to account for hierarchical and binary outcome variables. Four models were constructed. The first (null) model had no exposure or independent variables and was used to check there was variability in probability of children with fully immunized across the community. The second and third models comprised individual and community variables, respectively. The fourth multivariate, multilevel regression model adjusted for outcome variables and predictors that were significant at the individual or community level. The effects of individual and community level predictors on the dependent variable were assessed independently. Bivariate analysis was performed to test the effect of each independent variable on the immunization coverage. Only variables correlating with fully immunization (for our data set defined as all variables correlating with immunization with p-value of ≤ 0.25) were selected for the consecutive multivariate analysis[32]. Multicollinearity testing was performed using Variance Inflation Factors (VIF) and independent variables with VIF >5 were removed. Estimated associations (fixed-effects) between the likelihood of full vaccination and various explanatory variables were expressed as adjusted odds ratios (AOR) with 95% CI. Variations (random effects) were reported as intra-cluster correlation coefficients, or the percentage of variance explained by the community-level variables [33]. The Akakie information criterion was used to estimate goodness-of-fit of the adjusted final model in comparison with the preceding individual- and community-level models. The model with the lowest value was considered the best-fit model [33]. All variables with P-values less than 0.25 in the bivariate analysis were included in the multivariate logistic regression.