Trends and Determinants of Full Immunisation among Children Aged 12–23 Months: Analysis of Pooled Data from Mozambican Household Surveys between 1997 and 2015

The 1974 Expanded Program on Immunisation has saved millions of children worldwide by promoting full immunisation coverage (FIC). However, forty years later, many sub-Saharan African countries remain well below its target of 90% FIC. This study analysed the level, trends and determinants of FIC in 4322 Mozambican children aged 12–23 months from pooled data from four national surveys between 1997 and 2015. Descriptive statistics and multivariable logistic regression models were performed to analyse the factors associated with full immunisation coverage. Overall, the coverage of fully immunised children increased from 47.9% in 1997 to 66.5% in 2015, corresponding to a 1.8% yearly increase. The needed FIC growth rate post-2015 was 4.3 times higher. Increased maternal education and a higher household wealth index were associated with higher odds of FIS. Furthermore, attending antenatal care (ANC) visits, institutional delivery and living in southern provinces were also associated with increased odds of FIS. Between 1997 and 2015, FIC among 12–23-month-old children made modest annual gains but remained well below international targets. Factors related to access to healthcare, educational level, socioeconomic status and geographical location were associated with improved FIC. Targeted interventions to expand these factors will improve immunisation coverage among Mozambican children. In this study, we analysed a pooled dataset from multiple national representative cross-sectional household surveys (DHS 1997, DHS 2003 and DHS 2011, and IMASIDA 2015) [9,10,17,18]. The surveys were designed to collect characteristics of the household, including maternal and child health status, through a country-wide multistage and stratified random sample. Survey information was obtained through questionnaires administered to women who were between 15 and 49 years of age who provided information on the household’s sociodemographic, economic and health status, including maternal and child health, environment, behaviour, and pre- and postnatal care. The survey response rate of the interviewed women was high overall at 91.5%, 90.9%, 98.9% and 94.5% in 1997, 2003, 2011 and 2015, respectively. Data collection was conducted between March and June, August and December, April and November, and June and September for DHS 1997, DHS 2003, DHS 2011 and IMASIDA 2015 surveys, respectively. Detailed methodology for the survey’s participant selection has been described elsewhere [9,10,17,18]. Permission to use these datasets was granted by the Monitoring and Evaluation to Assess and Use Results Demographic and Health Survey (MEASURE-DHS). Mozambique is a southern East African country covering 801,590 km2 (Figure 1). Its estimated 30,066,648 inhabitants reside primarily (65.9%) in rural areas and about 15.3% of the population are children below the age 5 years of which 17.3% are children living the second year of life (Table S1 and Figure S1) [19]. Following independence from Portuguese colonial rule in 1975, Mozambique’s 16-year civil war destroyed its infrastructure and displaced more than half of its population [20]. Presently, the country is classified as a low-income country (LIC) by the World Bank, with a gross domestic product per capita (average annual income) of USD 503 in 2019 [21]. According to the United Nations 2020 Human Development Index, Mozambique was ranked 181 out of 189 countries, decreasing by one point from its ranking in 2019 and 2018 [22]. The 1979 EPI launch in Mozambique introduced BCG, polio, DTP and measles vaccines to the population [3]. Gradually, more vaccines were added to the national schedule, including an updated DTP (containing additional immunogens for hepatitis B and haemophilus influenzae type B) in 2008 and pneumococcal conjugate vaccine (PCV) in 2013. Rotavirus vaccine (RV), injected polio vaccine (IPV) and the second measles dose were introduced in 2015, data from which were not captured in the 2015 survey. The measles and rubella vaccine (MRV) was introduced in 2019 [3]. Despite increased availability of these vaccines, vaccine coverage, accessibility and uptake has remained below EPI targets. Map of Mozambique showing the province and province capitals. The present analysis included women surveyed who had a living child aged 12–23 months at the time of the survey. Interviewers communicated with study participants using participants’ preferred language. When available, children’s health cards, which contain information on vaccination doses and related dates, were used to determine children’s immunisation status. When health cards were not available, mothers’ verbal reports by recall of the vaccines their children had received were used to determine children’s immunization status. Table 1 details when health cards and when mothers’ verbal reports were used across surveys. Characteristics of the study population from 1997–2015 surveys. Complex survey methods were used for these estimates in each survey. For this analysis, full immunisation status is the outcome variable of interest. The WHO guideline considers a fully immunised child to have been administered eight doses in total: one dose of BCG; at least three doses of OPV (excluding that given immediately after birth); three doses of DTP; and one dose of measles vaccine between 12 and 23 months of age [23]. This outcome variable aligns with similar studies across many SSA countries [13,24,25,26,27]. Each vaccine dose variable had five response categories: (0) no vaccine; (1) vaccination dates on card; (2) reported by mother; (3) vaccination marked on card; and (8) don’t know. The categories (1), (2) and (3) were recoded as “1” to indicate those that received vaccines and (0) and (8) were recorded as “0” to indicate that no vaccination was received. The vaccination status of each of the 8 doses were then combined to compute a final variable, “immunisation status,” categorised as “1” for fully immunised and “0” for not fully immunised, capturing children who had missed one or more of the 8 doses. Drawn from the literature and data availability across all surveys, a total of 13 variables were selected for potential association with Mozambican children achieving full immunisation status. The variables included in this analysis were: (1) maternal age at the date of the survey, grouped as 15–24, 25–34 and ≥35 years old; (2) maternal educational level defined as illiterate, primary and secondary or above; (3) marital status, which originally had six categories (single, married, living with a partner, separated, divorced, widowed) and which was recoded into three categories (single/never in a union, married/living with a partner and divorced/separated/widowed); (4) mother’s occupation defined as unemployed (not working/household domestic) and employed (professional, clerical, sales, skilled and unskilled manual, services, agriculture, self-employed and all others); (5) geographic location whereby the provinces were included separately and aggregated to reflect the three regions (northern, central and southern) of the country; (6) area of residence, categorised as urban and rural; (7) wealth index, which was maintained as defined in each survey: poorest, poor, middle, richer and richest; (8) religion, the categorization of which varied among surveys but was regrouped to reflect four standardised responses to all datasets: Catholic, Islamic, Protestant and others, which also included no religion (although religion was analysed in all surveys, this variable was not included in the multivariable regression model due to its possible collinearity); (9) number of ANC visits (no visits, 1 to 3, ≥4); (10) place of delivery (at home/other, at health facility); (11) sex of the child (male and female); (12) birth order (1, 2 to 3, ≥4); and (13) health card possession (seen, not seen, no card). Descriptive statistics were used to summarise the selected factors. The FIC and its 95% confidence interval (95CI) were computed for each survey per each potential determinant. Then, per determinant we estimated the annual exponential growth rate (AGR) of FIC, which is a relative yearly growth rate between two subsequent surveys, s and s*, using the formula: where Covs and Covs* represent coverages of two subsequent surveys, respectively, and years and years* are the years of the two subsequent surveys, respectively. The standard error for AGR was computed through the delta method, from which 95CI were produced. In addition to the AGR, we computed the needed annual growth rate (NAGR) to reach 90% FIC by 2020 since 2015. To improve readability, we present the AGR and the NAGR in percentages by subtracting 1 and multiplying by 100, abbreviated AGRp and NAGRp, respectively. To analyse factors associated with full immunisation status, we used a mixed-effects logistic regression with the primary sampling unit (PSU) as random intercepts. This analysis assesses the pooled association across the period of time between 1997 and 2015. All covariates included in the model are based on the literature review and data availability. In addition, a linear term of the year of the survey was introduced in the model to capture potential changes over time. A linear term was chosen to improve the interpretability of the parameters. The province and the urban/rural variables were included as covariates in the models. The province is included as dummy indicators coded as sum contrasts. Therefore, its coefficients are to be interpreted as deviations from the overall country average. In addition, complex sampling logistic regression (reported in the Supplementary Materials) specific to each survey was performed as a sensitivity analysis for the overall analysis. We report the odds ratio (OR), 95CI and the p-values. Data management procedures and descriptive and complex survey analysis were conducted using Statistical Package for the Social Sciences (SPSS) version 26.0 [28]. The mixed-effects model was performed in R version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria) using maximum likelihood estimation with 30 Adaptive Gaussian Quadrature’s through the package GLMMadaptative [29].