Type of study
This was a secondary analysis of data from the Guinea Demographic and Health Survey (DHS) conducted in 2018 which is a cross-sectional survey. The collection methodology as well as the reports are available and accessible on the DHS program website (http://dhsprogram.com). For this study, the data from the women's individual questionnaire were used. For this study, the data from the individual questionnaire of women were used in particular information on the nutritional practices of children, including breastfeeding.
Study population and sampling
We considered as the study population all women aged 15 to 49 from the individual woman questionnaire. The number of women surveyed was 10,506. Our analysis focused on women who had their last birth six (6 months) before the collection and who consented to the survey during the collection period.
This study included all mothers of children who had their last birth six months before the data collection, and whether or not they practiced exclusive breastfeeding; present during the visit of the investigators and who answered the question on breastfeeding.The following were not included in this study: Mothers whose children were not alive at the time of collection; those whose newborns did not live with them and mothers whose children are over 6 months of age (Figure 1).
Study variables
Our dependent variable was exclusive breastfeeding. This variable was obtained from a set of variables related to infant feeding.
First, we created the dichotomous variables that corresponded to the types of food that the mother gave the child the 24 hours preceding the survey.
Secondly, we grouped together all the variables whose modality is to have ingested a liquid or a solid other than breast milk. This variable constituted the modality "not having been exclusively breastfed for the variable exclusive breastfeeding. After the construction of the dependent variable, we proceeded to the following recoding:
All mothers who answered that they had not given their child any food other than breast milk were coded 1 and the modality which corresponds to the ingestion of foods other than breast milk for the child was coded 0.
Variables at the individual level (level 1) were the individual characteristics of the mother and child: mother's age, education level, marital status, parity, region of residence, occupation of woman, type of household, number of antenatal consultations, place of delivery, type of delivery assistance, mode of delivery, household size, type of wealth quintile, the child's breastfeeding age and gender.
Community level variable (level 2) is represented by the enumeration area or community level. The contextual variable in our study is place of residence.
Data analysis
For the entire analysis, STATA 15.1 software was used. Before the analysis started, data cleaning, labeling, and recoding were done for all of the selected variables. The analysis was carried out in three stages.
The descriptive analysis of all the selected variables was the first step. The socio-demographic and maternal characteristics of the subjects studied were described. The percentages were calculated for each variable.
The univariate analysis consisted in measuring the association between the dependent variable and each of the independent variables selected by performing a simple logistic regression. A variable is retained for the construction of the multilevel model when its degree of significance p-value is less than 20%.
For multivariate analysis, a binary multilevel logistic regression with random effect was performed using the ascending step-by-step method. The adjusted SDGs ratios were estimated with their p-value and their confidence interval. Successive models were compared using the likelihood ratio test. This involved multivariate hierarchical analysis to determine the fixed and random effects of the characteristics associated with the practice of exclusive breastfeeding. Using this same analytical technique, we looked for associations between the practice of Exclusive breastfeeding and variables at the individual and community level. Three two (2) level statistical models have been implemented:
The first model did not contain any independent variables. This allowed us to compare our multi-level regression model to a classic logistic model and to appreciate the variance of exclusive breastfeeding between the different EAs.
The second model was obtained after the introduction of the explanatory variables of the individual level.
The third model (3) or the complete model, which contains in addition to the variables of the individual level, the variable of the contextual level which is the place of residence.
The intra-class correlation coefficient was calculated for each statistical model and the discriminating power of the model was calculated using the Roc table.
Ethical consideration
This survey benefited from the approval of the Health Research Ethics Committee before its implementation. The database was obtained after a request to the EDS database management program (DHS-Program) which gave us its favorable opinion for the use of this database.