Vaccination dropout rates among children aged 12-23 months in Democratic Republic of the Congo: a cross-sectional study

The Mont Ngafula II health district is located in the western part of Kinshasa. It shares borders with five districts: Binza Ozone, Binza Météo, Selembao, Mont Ngafula I, and Masa (in Kongo central province). This health district is subdivided into 14 health areas: Antenne, Don Bosco, Kimbwala, Kimbondo, Maman Mubutu, Matadi Kibala, Mambre, Matokame, Mazal, Mitendi, Ngombe, Mbudi and Sans Fil (Fig. 1). It has a general pediatric hospital, 107 health centers including 22 health posts. Only 14 health centers deliver EPI vaccines to the target population groups. The main local economic activities are based on agriculture, livestock, and fishery products.

A cross-sectional household survey was conducted. All mothers/caregivers with children aged 12–23 months, between the 1st of June and the 1st of July 2019, were eligible. Since the exhaustive number of households and children aged 12–23 months was not available, we referred to the WHO vaccination coverage cluster surveys to perform a two-stage sampling design [16]. As recommended to select an appropriate sample size by ensuring a maximum absolute confidence interval width of ± 10% at fairly low vaccination coverage, 30 clusters were first sampled using probability proportional to their size, based on estimated population data per village/neighborhood provided by the health district authority. The sample size was determined using the following formula: N= Z²×p(1-p)/d², where N is the sample size, Z (1.96) is the 2-sided level of significant, p is the immunization coverage of the country (45%) as proposed in the study of Baguune at al. [17], and d is the desired width of the confidence interval. Considering a non-response rate of 10%, we found a total sample size of 418 children aged 12–23 months. At the second stage, we used the EPI random walk method [18] to select 14 children from each of the 30 clusters. The starting point was considered the first household randomly selected at the geographic center of each cluster using the spinning bottle method, and then we continued to the nearest household in the same direction until 14 eligible children were obtained.

A structured questionnaire translated into the local language was used to collect data on sociodemographic characteristics (children, respondents and household) and respondents’ perceptions. Interviewers were previously recruited and trained on the study protocol and the questionnaire use. After the training, a pre-test was conducted in the field to check the reliability and validity of our study tools. The questionnaire included information on socio-demographic characteristics, child immunization status, and perception of respondents (mothers/caregivers) on the organization of vaccination in health facilities. Door-to-door visits were conducted and only consenting respondents of selected children participated in the survey. In case where two or more children aged 12–23 months were found in the same household, the youngest child was selected [18]. Information on children’s immunization, such as dates and doses of each vaccine received, was obtained from vaccination cards or respondents’ recall if the card was not available. The data collected was then recorded on Microsoft Excel® 2013 files.

Annual routine immunization coverage and dropout rate data were extracted from the 2017-2018 database collected at the health area scale provided by the local health authority. They were also recorded in Microsoft Excel® 2013 (Dataset V2). According to the Congolese routine immunization schedule (Table 1), the dropout rate was considered the percentage of children aged 12-23 months who received the 1st but not the 3rd dose of DTP: [(first dose – last dose)/first dose] ×100 [4].

We used data obtained from the Mont Ngafula II health authority to map the annual geographic distribution of routine immunization coverage and dropout rates at the health area scale using QGIS 2.14.4 (Open-Source Geospatial Foundation). The Mont Ngafula II health district shapefile was provided by the Directorate of Disease Control of the DRC Ministry of Public Health (validated by WHO DRC) and used to generate the map. The latter allowed to assess the dynamic system of vaccination services at the health area level.

Data collected through the cross-sectional household survey were used to identify factors predicting high vaccination dropout rates among children aged 12-23 months in the Mont Ngafula II health district. Descriptive statistics such as frequencies and percentages were calculated and presented into tables. Differences in children, respondents, household characteristics, and respondents’ perceptions were tested by the Chi-square test. Since our data were clustered at the health area scale, we performed logistic random effects regression models using the R package glmmML to measure the association between independent predictors and immunization dropout. The dependent variable “dropout” was coded as 0 for children who received the three-dose DTP series and 1 for children who received the DTP1 but not the DTP3. The explanatory variables were also summarized into two categories: sex of child (0 = male; 1 = female), birth order (0 = ≤ 2nd; 1 = ≥ 3rd), card possession (0 = yes; 1 = no), age of respondent (0 = ≤ 19 years; 1 = > 19 years), education level (0 = ≥ secondary; 1 = ≤ primary), occupation (0 = no; 1 = yes), place of residence (0 = urban; 1 = rural), type of family (0 = 2-parents; 1 = single parent), number of children (0 = ≤ 2 children; 1 = ≥ 3 children), attitude of health staff (0 = positive; 1 = negative), availability of seats (0 = yes; 1 = no), respect of the order of arrival (0 = yes; 1 = no), waiting time (0 = < 1 h; 1 = ≥1 h), reminder system (0 = yes; 1 = no). The final multivariate logistic random effects model included variables that were selected by a subsequent backward selection based on the results of the univariate analysis and then the multivariate model for each section of the household survey. At each step, variables with p ≥ 0.20 were excluded until only variables with a p < 0.20 remained. All statistical analyses were performed with R® version 3.6.1.