The study utilized data from the 2016-17 Burundi Demographic and Health Survey (DHS). Specifically, data from the birth recode file was used. The DHS is a nationally representative survey that is conducted in over 85 low-and middle-income countries globally. The survey focuses on essential maternal and child health markers including “health seeking behaviour”, “contraceptive use”, “skilled birth attendance”, “immunization among under-fives” and “intimate partner violence” . The survey employs a two-stage stratified sampling technique, which makes the data nationally representative. The study by Aliaga and Ruilin  provides details of the sampling process. The surveys employ a two-stage stratified sampling technique which makes the survey data nationally representative . Specifically, the initial stage had to do with the generation of a sampling frame which contained a catalogue of enumeration areas (EAs) that covered the given country. The EAs are mostly generated based on the most recent national census data in the country. Each EA is subsequently segmented into standard size segments of about 100–500 households per segment. Thereafter, a sample of predetermined segments was selected randomly with probability proportional to the EA’s. The next stage–second stage also involved the systematic selection of households from a list of previously enumerated households in each selected EA. This stage then involved the conduction of in-person interviews in selected households with the various target populations: women (15–49) and men (15–64). The number of selected households per EA ranged from 30 to 40 households/women per rural cluster and from 20 to 25 households/women per urban cluster. A total of 11,828 women who had complete information on all the variables of interest were included in our study.
Definition of variables
The outcome variable for the study was health seeking behaviour for childhood illnesses. It was derived as a composite variable from two questions, “Did [NAME] receive treatment for diarrhea?”, and “Did [NAME] receive treatment from fever/cough?” The responses were “Yes” and “No”. For the purpose of this study, “No” and “Yes” were coded 0 and 1 respectively.
The study looked at problems in accessing health care as the independent variable. This variable was generated by asking women if they had serious problems in accessing health care for themselves when they are sick and the type of problem. The problems were difficulty with distance to the facility, difficulty in getting money for treatment, difficulty with getting permission to visit health facility, and difficulty in not wanting to go for medical help alone. These variables were recorded as “Big problem” and “Not big problem”. For the purpose of this study, “Big problem” and “Not big problem” were coded 0 and 1.
Sixteen control variables were considered in the study. These variables are community literacy level, community socio-economic status, age, marital status, health care decision making capacity, employment, religion, place of residence, parity, exposure to mass media (radio, television and newspaper), sex of head of household, size of child at birth, birth order, twin status, and sex of child. The variables were not determined a priori; instead, based on parsimony, theoretical relevance and practical significance with health seeking behaviour for childhood illnesses [12,21,22,23]. Marriage was recoded into “never married (0)”, “married (1)”, “cohabiting (2)”, “widowed (3)”, and “divorced (4)”. We recoded parity (birth order) as “one birth (1)”, “two births (2)”, “three births (3)”, and “four or more births (4)”. We recoded religion as “Christianity (1)”, “Islam (2)”, “Traditionalist (3)”, and “no religion (4)”. We recoded size of child at birth as “larger than average”, “average”,and “smaller than average”. And we recoded twin status as “single birth” and “multiple birth”.
The data was analysed with STATA version 14.2. The analyses were done in three steps. The first step was the computation of the prevalence of women’s health seeking behaviour for childhood illnesses in Burundi (see Figure 1). The second step was a bivariate analysis that calculated the prevalence and proportions of health seeking behaviour for childhood illnesses across the independent variables with their significance levels (see Table 1). Statistical significance was considered at a p-value less than 0.05. Variables that showed statistical significance in the bivariate analysis were further analyzed using multivariate hierarchical logistic regression. Before conducting the multivariate hierarchical logistic regression analysis, a multi-collinearity test was carried out among all the statistically significant variables to determine if there was evidence of multicollinearity between them. Using the variance inflation factor (VIF), the Multicollinearity test showed that there was no evidence of collinearity among the explanatory variables (Mean VIF=1.20, Max VIF=1.53, Minimum=1.03).
The multivariate hierarchical logistic regression analysis was carried out in three stages. At the initial stage, the key independent variables, maternal and child level variables that showed statistical significance with women’s health seeking behaviour for childhood illness at the bivariate analysis were entered in the first model to assess their association with women’s health seeking behaviour for childhood illness (see Model I). In the second stage, the community level variables that were significant at the bivariate analysis were added to the first model to assess their association with the individual level distal socioeconomic determinants variables and women’s health seeking behaviour for childhood illness (see Model II). For the third stage, all the variables were added to the full model (III). A sample weight (v005/1,000,000) to correct for over and under sampling was applied and the SVY command to account for the complex survey design and generalizability of the findings was also used. In this study, we relied on the Strengthening the Reporting of Observational Studies in Epidemiology’ (STROBE) statement in writing the manuscript .
Ethical clearance was obtained from the Ethics Committee of ORC Macro Inc. as well as Ethics Boards of partner organisations of the various countries such as the Ministries of Health. The DHS follows the standards for ensuring the protection of respondents’ privacy. Inner City Fund (ICF) International ensured that the survey complies with the U.S. Department of Health and Human Services regulations for the respect of human subjects. This study employed a secondary analysis of data and therefore no further approval was required because the entire data is available in the public domain. Further information about the DHS data usage and ethical standards are available at http://goo.gl/ny8T6X.