Burundi is a country in East Africa. It borders Rwanda to the north, Tanzania to the south and east and the Democratic Republic of Congo (DRC) to the west. It is part of the Great Lakes region. Its surface area is 27,834 km², including 2700 km² of lakes and 23,500 km² of potentially agricultural land. The total population, according to the RGPH 2008 projections, is estimated at 11,215,578 inhabitants, of whom 5,500,949 are men and 5,714,629 are women. This population is distributed in 18 provinces, 119 communes and 2911 hills.

The commune of Buhiga, on which our study focused, is made up of three zones spread over 30 hills and 132 sub-hills.

The population of children aged 6 months to 59 months is 12,764 inhabitants in 2024 according to ISTEEBU’s 2010-2050 municipal projections.

This is a cross-sectional study with an analytical purpose that was carried out in the different households of the commune of Buhiga. The study covered a 4-month period from March 1 to June 31, 2024. A survey questionnaire was sent to each survey and completed by the interviewer during data collection.

The choice of the Buhiga commune was motivated by the results of the SAMART 2022 survey which showed that the Buhiga commune is among the communes with a high rate of children under 5 years of age suffering from global acute malnutrition, i.e. 6.4% which is higher than the WHO standards of 5%.

The study population consisted of children aged 6 to 59 months at the time of the survey who were in different households in the hills of the Buhiga commune ( Figure 1 ).

The prevalence of children aged 6 to 59 months with Global Acute Malnutrition assessed by weight-for-height and/or MUAC ratio is known and is 6.4% according to data reported in the 2022 SMART survey. This proportion was used to calculate the sample size.

The sample size was determined using the following Schwartz formula.

$n=\frac{z^{2}Xp\left(1-p\right)}{e^2}$ ( Schwartz, 1969 )

n = sample size;

Z = 95% z-score (1.96);

p = prevalence of acute malnutrition (0.064);

e = margin of error (4.3%);

n = 124.

Applying the 5% non-response, the sample size becomes 130, i.e. 130 households with children aged 6 to 59 months.

$n=\frac{124.4X105}{100}=130$

The sample size of our study was 130 mothers or their guardians with children aged 6 to 59 months, i.e. 130 households in the commune of Buhiga.

1) Inclusion criteria

2) Non-inclusion criteria

Our study took place in all the areas of responsibility of the health centers located in the commune Buhiga using the technique of proportional allocation stratification, each area of responsibility being a stratum. To determine the number of children to be surveyed in each stratum, the 130 children in our sample were allocated proportionately based on the average number of children aged 6 to 59 months in each area of responsibility ( Table 1 ).

The sample by area of responsibility was identified using the following formula:

$n_i=\frac{n}{N}*N_i$ ( York House et al., 2024 )

With:

N_i: number of children to be included in the sample for each area of responsibility;

n: sample size to be drawn in all areas of responsibility of the Buhiga Health District;

N: Total number of children aged 6 - 59 months who are in the Buhiga community;

N_i: Total number of children aged 6 to 59 months in each area of responsibility of the Buhiga Health District.

The information was collected through a questionnaire that was sent to the children’s mothers or their guardians. It is a structured questionnaire comprising 4 dimensions: a dimension related to sociodemographic factors, a dimension related to environmental and behavioral factors, and a dimension related to the child’s morbidity.

The questionnaire has been translated into the national language to facilitate the respondent’s understanding. Before completing the questionnaire, the interviewer began to explain the purpose of the survey and assured the respondent of the confidentiality of the information provided.

Prior to the interview, informed and written consent was obtained from the mothers who consented on behalf of their children. In cases where the mother was illiterate, verbal consent was obtained from the mother according to a detailed description of the protocol and written consent was obtained from her husband who consented in writing on behalf of the mother and child.

The questionnaires have been coded to maintain anonymity and confidentiality. Data collection was carried out from Monday to Friday (8 a.m. to 3 p.m.) on the identified hills until the required sample was obtained. After collection, data entry was performed in Excel version 2016.

To test the understanding of the questionnaire (pre-test), the questionnaire was first sent to some households with children aged 6 to 59 months in the Bugenyuzi commune, then some questions were adapted and reformulated. The pre-test also made it possible to estimate the time needed to investigate a person.

The dependent variable for this study is global acute malnutrition which is a dichotomous variable that was determined using the mid-upper arm circumference using a tape measure coded by red, yellow, green colors, so we retained those with a MUAC < 115 mm, we also assessed the presence or absence of bilateral edema.

The independent variables which has been tested and have significantly influenced the dependent variable on the one hand and other independent variables which have been associated with the dependent variable on the other hand are grouped in 3 headings, namely:

1) Socio-economic Characteristics;

2) Factors related to child morbidity;

3) Behavioral and environmental factors.

A database was built using Excel 2016 and the data was exported to STATA for analysis. After the data was entered, the data was pre-processed to assess the quality of the information entered, including checking the normality of the data. If quantitative data were missing, they were replaced by the mean or median for variables with a normal and abnormal distribution, respectively. For qualitative data, they have been replaced by a frequent modality. Outliers were also corrected prior to the actual analysis.

Subsequently, the data analysis was carried out as follows:

Different strategies have been set to minimize or avoid possible biases:

The results of the study were extrapolated to all children aged 6 to 59 months in the Buhiga commune.

Table 3 shows that the variables Father’s Education Level (p = 0.023), Mother’s Education Level (p = 0.000), Child Age (p = 0.008) and Household Size (p = 0.000) are associated with acute malnutrition.

Table 4 above shows that the variables Digestive disorders (p = 0.008) and Recurrent malaria (p = 0.001) are associated with acute malnutrition.

Table 5 above shows that only the variable Meal frequency (p = 0.000) is associated with acute malnutrition.

In Tables 6-8 below, the variables significantly associated (p < 0.05) with acute malnutrition in children aged 6 to 59 months in the commune of Buhiga are: digestive disorders, compliance with the vaccination schedule, the father’s level of education, the mother’s level of education, the age of the child, the adapted latrine system, domestic violence, marital status, household size and frequency of meals eaten.

Table 7. Analysis of the association between acute malnutrition and morbidity factors in children aged 6 to 59 months in Buhiga commune of Karusi province.

Table 8. Analysis of the association between acute malnutrition and behavioural and environmental factors.

Indeed

The following Table 9 shows the explanatory variables retained in a multivariate logistic regression model as significantly associated with acute malnutrition in the Buhiga commune.

All independent variables tested that had a significant influence on the rate of acute malnutrition in the bivariate analysis with a p < 5% (Marital status of the mother in a common-law union, mother’s level of education, father’s level of education, age of the child, household size, digestive disorders, frequency of meals consumed, adherence to the vaccination schedule, modified latrine) were taken and their association with the dependent variable (global acute malnutrition) was tested and eliminated confounding factors with p value > 5% (Mother’s level of education, father’s level of education, child’s age, digestive disorders, adherence to the vaccination schedule, adapted latrine) and remained with the variables significantly associated with acute malnutrition in children aged 6 - 59 months in the Buhiga commune of the Karusi Province with a p < 5% (frequency of meals consumed, household size, marital status of the mother in common-law union) which constitute the determinants of acute malnutrition in children aged 6 - 59 months in the Buhiga commune of the Karusi Province.

At the end of the analysis, logistic regression eliminated confounding factors. Variables with a significant p-value (<5%) and those with a p-value of less than 20% were integrated into the initial model with top-down elimination of variables with p > 0.05.

Thus, the variables with p < 0.05 at this stage are then significantly associated with acute malnutrition in children aged 6 to 59 months in the commune Buhiga and therefore constitute the determinants of acute malnutrition in children aged 6 - 59 months in the commune Buhiga, and it is the frequency of meals consumed (p = 0.017), household size (p = 0.011) and marital status (p = 0.043).

During our study, 3 types of data analysis were performed:

A database was built using Excel 2016 and the data was exported to STATA 15.1 for analysis. After the data was entered, the data was pre-processed to assess the quality of the information entered, including checking the normality of the data. If quantitative data were missing, they were replaced by the mean or median for variables with a normal and abnormal distribution, respectively. For qualitative data, they have been replaced by a frequent modality. Outliers were also corrected prior to the actual analysis.

Subsequently, the data analysis was carried out as follows:

At the end of the analysis, the variables significantly associated (p < 0.05) with acute malnutrition among children aged 6 to 59 months in the commune of Buhiga were the frequency of meals consumed, household size and marital status ( Figure 4 ).

The graph above shows the ROC curve obtained from the results of the final model. The area under the ROC curve is 0.8627. The model’s ability to correctly classify observations is 86.2%. We can conclude that this model has predictive power, with very good discrimination.