Menu
Menu
Menu
Menu
Background: Child vaccination coverage in low- and middle-income countries is still far from complete, mainly among marginalized people such as children with illiterate mothers. Objective: This study aims to examine the association between maternal literacy and immunization status of children in Ethiopia and southeastern India (Andhra Pradesh and Telangana) and test whether state-run health centers and community health workers moderate that association. Methods: This study is based on cross-sectional data from samples of children in Ethiopia and India, collected as part of round 2 within the Young Lives study (2006). Multilevel logistic regression was conducted to estimate the association between maternal literacy and the completion of four kinds of child vaccinations. We further tested for cross-level interactions between state-run health centers or community health workers and maternal literacy. Estimates were adjusted for several individual- and household-level demographic and socioeconomic factors. Results: Literate mothers were more likely to complete all four kinds of vaccinations for their children compared to illiterate mothers in Ethiopia (Odds Ratio (OR)=4.84, Confidence Interval (CI)=1.75-13.36). Presence of a health center was positively associated with completed vaccinations in India only (OR = 6.60, CI = 1.57–27.70). A cross-level interaction between community health workers and maternal literacy on the vaccination completion status of children was significant in Ethiopia only (OR = 0.29, CI = 0.09–0.96). Conclusions: Our findings suggest that increased availability of community health workers may reduce the child vaccination gap for illiterate mothers, depending on the country context.
Data were obtained from an international and longitudinal survey named the ‘Young Lives Study’. The Young Lives study is comprised of two cohort groups, the first of which is a ‘younger cohort’ including 1999 and 2011 children aged between 6 and 18 months in Ethiopia and India, respectively. The second group, labeled the ‘older cohort’, included 1000 and 1008 children aged 7 to 8 years in Ethiopia and India, respectively, at the time of recruitment in 2002. The first round of data from the younger cohort was obtained in 2002 (age of cohort: 6 to 18 months), the second round in 2006–2007 (age of cohort: 4 to 5 years), and the third round in 2009–2010 (age of cohort: 7 to 8 years). Our study mainly used the round 2 data set from the younger cohort because it included information on basic vaccinations which should have been completed when the child reached the age of 4 to 5 years. Only sibling status was extracted from the round 1 data set as that information was only collected during the first round. The Young Lives study employed a clustered sampling strategy with a semi-purposive sampling of 20 sentinel sites, with oversampling of sites covering poor areas rather than nationally representative sampling in each country, because the aim of the study was to assess the causes and consequences of childhood poverty [27]. In India, sentinel sites were chosen only within the states of Andhra Pradesh and Telangana, while sites were selected nationwide in Ethiopia. The sentinel sites include both urban and rural areas, representing a range of regions, policy contexts, and living conditions. Within each sentinel site, all households containing children in the target age groups were identified and listed, from which 150 (100 for the younger cohort and up to 50 for the older cohort) were randomly selected [27–29]. A comparison was made between the study samples and nationally representative samples using data from the Welfare Monitoring Survey 2000 for Ethiopia and data from the Demographic and Health Survey 1998/99 for India. Comparison of several living standard indicators showed that the samples in the Young Lives Study were slightly better off, which might be partly explained by a substantial decrease in national poverty rates over the gap between the survey year of the Young Lives Study and nationally representative samples. More information on the sampling strategies in each country can be found elsewhere [30,31]. Households that refused to participate (representing less than 2% of the selected households) were replaced with other households from the list. The response rate was above 90% in both countries. Attrition rates were notably low compared to other longitudinal studies in similar contexts, ranging from 0.50 to 3.52%, and were similar across countries [32]. Data were collected from the child’s main caregiver, which was either the child’s mother (mainly) or father (only minimally), using a standardized, interviewer-administered questionnaire. All interviewers received training based on common guidelines. After excluding cases with missing information or ‘don’t know’ responses for the outcome variable and any of the explanatory variables, 1157 children in 22 communities in Ethiopia and 1455 children in 75 communities in India comprised the final analytic samples for this study. A binary variable indicated whether the respondent’s child had completed four kinds of vaccinations including BCG, MCV, DTP3, and polio. We did not include Hib because Hib was introduced in Ethiopia in 2006 and in India in 2009, which is much later than our survey period. Answers to these questions were obtained from the child’s vaccination card if it was available. Otherwise, the answer was based on the respondent’s recall (yes/no) for each vaccination. Respondents were asked whether the mother and father of the index child could read and understand a letter or newspaper in their own language, which was Telugu in India, and the most commonly used language in that locality in Ethiopia. Each parent was assessed as illiterate if they could not read and understand it at all or as literate if they could read and understand it easily or with difficulty. Data on state-run health centers and community health workers were collected by asking key informants in the community, such as a community leader or village representative, whether community health workers were present and delivered their services in the locality and whether the state-run health centers were available 4 years prior when the index child received a series of vaccinations. The definition of community health workers did not include social workers or mental health workers. State-run health centers did not include health posts. Other explanatory variables used in this study were maternal age, child’s gender, child sibling status, household wealth status, type of residence, and region. Maternal education level was not included due to concerns about the high probability of collinearity with the literacy variable and the over-fitting statistical models problem, according to the recommendation of a previous study [33,34]. Maternal age was categorized into two groups: 30 years or younger and older than age 30. Sibling status was categorized as only child vs. child with sibling at the time of the first survey. (Note: as information on the child’s sibling status was assessed only during the first survey round, the status might have changed later if a child was born in the household after the first survey round.) Wealth index was measured with wealth status and living environment, including housing quality, water and sanitation quality, and access to energy sources, and then constructed by principal component analysis from three equally weighted components – a housing quality index, a services quality index, and a consumer durables index – with ranges from 0 to 1 [35]. Subsequently, these values were ranked into wealth quintiles from the poorest to the wealthiest. The selection of these variables as potential confounders was guided by a review of the previous literature. We performed multilevel multivariable logistic regression with a random intercept model to examine the association between maternal literacy and completion of four kinds of child vaccination while accounting for the clustering of observations at the community level for each country in 2006. To test for a moderating effect of state-run health centers and community health workers on that association, cross-level interaction terms were constructed by multiplying two dummy variables representing presence of a state-run health center or community health workers and maternal literacy. First, models 1–1 and 1–2 considered only individual-level and household-level variables with and without controlling for paternal literacy. Then, the two community-level variables were included as explanatory variables into separate regression models (model 2–1 for state-run health center and model 3–1 for community health worker). Finally, we extended models 2–1 and 3–1 by adding the cross-level interaction terms to test the moderating effect of two community-level variables on the association between maternal literacy and completion of four kinds of vaccinations (models 2–2 and 3–2). Analyses were carried out in 2016 using SAS version 9.3 [36].
