This is a population-based study that utilized a cross-section secondary data from the Malawi Demographic and Health Survey (MDHS) 2015-16. Methods used in this study have been described in details elsewhere. (4) In brief, the MDHS was designed to produce a nationally representative sample using a two-stage sampling design. The sampling frame used for the MDHS is the Malawi Population and Housing Census (MPHC) conducted in 2008. In the first stage, 850 standard enumeration areas (SEAs), including 173 SEAs in urban areas and 677 in rural areas, were selected with probability proportional to the SEA size and with independent selection in each sampling stratum. In the second stage of selection, a fixed number of 30 households per urban cluster and 33 per rural cluster were selected with an equal probability systematic selection from the newly created household listing. A total of 27,516 households were selected with a response rate of 99%.
Data were collected using face-to-face interviews from women of reproductive age 15–49 years on the measures of population health, sociodemographic, environmental, anthropometry, and HIV/AIDS, immunization, and child health care indicators. Information on vaccination coverage was obtained in two ways. Mothers were asked to show whether they had a vaccination card for each child born 5 years prior to the data collection. If the mother could not show an immunization card, she was asked to report whether the child had received any vaccination.
The outcome variable of this study was complete vaccination. Complete immunization was defined as any child aged 12–23 months who received all age appropriate vaccination. All age appropriate vaccination include one BCG, three doses of DPT-HepB-HiB, four doses of oral polio vaccine, three doses of pneumococcal vaccine, two doses of rotavirus vaccine, and one dose of measles vaccine.
Individual–level factors included child, maternal and households factors; sex of the child (male and female), the birth order (1, 2–3, 4–5, and ?6), mother’s age (15–24, 25–34, and 35–49 years), the mother’s education (no formal education, primary school education, and secondary and higher education), antenatal (ANC) visits (?1, 2–3, and ?4), immunization card (no card/no longer had a card, had a card and seen, and had a card but not seen), the household wealth index (poorest, middle and richest), number of children under the age of 5 years (?1, 2, and ?3). The household wealth index is a composite measure and was calculated according to the ownership of selected assets, such as televisions and bicycles, materials used for constructing the house, access to water and sanitation facilities, and other characteristics of a household. The household asset scores were generated through a principal component analysis. (14) The resulting asset scores were standardized and categorized into quintiles. (14)
The community-level factors were constructed by aggregating individual-level data to the cluster level. We referred to the primary sampling unit (PSU) of the DHS data as a community. We included three continuous variables namely community wealth, distance to the nearest health facility, female education, Community wealth was defined as the percentage of households in the community categorized as the richest (upper 40% quintile), whereas community female education was defined as the percentage of women aged 15–49 in the community with primary education and above. Community distance to the nearest health facility was defined as the proportion of women aged 15–49 in the community who perceived the distance to the nearest health facility as a big problem. All continuous community-level factors were categorized as ”low”, ”medium” and ”high” depending upon each variable’s tertiles. In addition, we included two variables indicating the area of residence, i.e. the place of residence (urban or rural), and geographical region (northern, central, and southern region).
The characteristics of the study sample were expressed as frequencies and percentages. Bivariate analyses were performed using Pearson’s Chi-square to test the differences between groups. The multivariate analyses were conducted using two-level multilevel multivariable logistic regression, fitting four different models. Model 1 (null model) had no explanatory variable and was used to decompose the total variance of complete immunization between the contextual and individual levels. Model 2 contained the individual level factors and in Model 3, only community contextual factors. Model 4 controlled for both individual and community-level factors.
Measures of association between the individual-level and contextual risk factors and complete immunization were reported as adjusted odds ratios (aOR) with their p-values and 95% confidence interval 95% (CI) after considering potential confounders. Random effects were expressed in terms of Area variance (AV), Intra-Cluster Correlation (ICC) and Proportion Change in Variance (PCV). The fitness of the model was assessed using Deviance Information Criterion (DIC). Two-tailed Wald test at significance level of alpha equal to 5% was used to determine the statistical significance of the determinants and all the analyses were performed using SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA).
The 2015-16 MDHS was implemented by the National Statistics Office (NSO) and the Community Health Sciences Unit (CHSU). The protocol for the questionnaires was reviewed and approved by the Malawi Health Sciences Research Committee, the Institutional Review Board of ICF Macro, and the Centers for Disease Control (CDC) in Atlanta. Informed consent was obtained at the beginning of each interview and the authors sought permission from the DHS program for the use of the data.
Table 1 shows the descriptive characteristics of the 3,125 women aged 15–49 years, dwelling in 850 different communities. Complete immunization coverage was highest in children born to mothers with secondary education and above, mothers with four ANC visits and above, children who had immunization card and seen, children living in households with at most one under-?ve child, and richest households.
The findings of the multilevel logistic regression are shown in Table 2. The contextual-level variance indicated that total variance of complete immunization can be attributed to the community in which the mothers were living (Model 1). This could be justified by the statistical significance of variation in the communities area variance (AV) = 0.3252; standard error (SE) = 0.0890; p <0.05. The ICC was 0.0890, implying that approximately 9% of the total variation in complete immunization coverage can be attributed to the communities. In Models 2 and 3, the variation remained statistically significant, even after the adjustment for individual- AV = 0.4897; SE = 0.1335; p<0.05 and community-level factors AV = 0.3178; SE = 0.0889; p< 0.05 respectively. The estimated proportional change in variance (PCV) indicated that about 51% and 2% of the contextual-level variance was explained by the individual and community-level risk factors respectively. The ICC showed that about 13% and 9% of the total variance in the community could not be explained even after adding individual level and community level factors respectively. Furthermore, in Model 4 the variation remained statistically significant even after adjusting for the potential confounders (individual- and community-level factors) AV = 0.4687; SE = 0.1324; p<0.05. The PCV showed 44% of the contextual-level variance of complete immunization can be explained by the individual and community-level factors compositional characteristics. However, 12% (ICC) of the total variance remained unexplained. Fixed effects Table 3 (Model 4) shows adjusted effects of individual and community–level factors on complete vaccination coverage. Achieving complete vaccination (aOR: 0.56; 95% CI: 0.32–0.93) was significantly lower among children whose mothers had at most one ANC visits compared to those born from mothers with four and above ANC visits. Furthermore, achieving complete immunization was much lower among children who had no health card/no longer have a card (aOR: 0.06; 95% CI: 0.04–0.07) and who had a card but not seen (aOR: 0.08; 95% CI: 0.06–0.12). Achieving complete immunization was significantly high among children who were born in households with at most one under-5-year child (aOR: 1.60; 95% CI: 1.09–2.36). Children living in the poorest households had significantly lower odds ratios of achieving complete immunization (aOR: 0.60; 95% CI: 0.40–0.92). At the community level, children living in communities with the middle (aOR: 0.73; 95% CI: 0.53–0.98) and high (aOR: 0.73; 95% CI: 0.53–0.99) percentage of the household who perceived the distance to the nearest health facility as a big problem had reduced odds of achieving complete immunization.