Data source, study design, and sampling

Using the most recent DHS data from 21 SSA nations, collected between 2015 and 2022, a cross-sectional pooled dataset was used. Angola (2015-16), Benin (2017-18), Burundi (2016-17), Ethiopia (2016), Gabon (2019-21), Ghana (2022), Gambia (2019-20), Guinea (2018), Kenya (2022), Liberia (2019-20), Mali (2018), Malawi (2015-16), Nigeria (2018), Rwanda (2019-20), Sierra Leone (2019), Senegal (2019), Tanzania (2022), Uganda (2016), South Africa (2016), Zambia (2018), and Zimbabwe (2015) were among the 21 SSA countries whose demographic and health surveys were used. The data were appended to determine the pooled prevalence of PEB in SSA and identify the factors associated with it. Each country’s survey has different datasets, such as those for males, females, children, births, and households. The kid’s record (KR) file was employed in this investigation. The DHS is a national survey that is primarily conducted in LMICs every five years. By using common methods for sampling, questionnaires, data collection, cleaning, coding, and analysis, it enables cross-national comparison [23]. A total weighted sample of 63,172 children aged 6 to 23 months who are living with their mother were included in the present study (Table 1). The DHS uses a two-stage, stratified sampling method [24]. The first step is creating a sample frame, which is a list of enumeration areas (EAs) or primary sampling units (PSUs) that encompass the entire nation. This list is typically created using the most recent national census that is available. The systematic sampling of the households included in each cluster, or EA, is the second step. More details on survey sample techniques are available in the DHS guidelines [25].

Table 1 Sample size for prolonged exclusive breastfeeding and its associated factors among children aged 6–23 months in sub-saharan African countriesVariables of the studyOutcome variable

According to the WHO, exclusive breastfeeding is defined as feeding exclusively on breast milk or expressed breast milk and avoiding any other liquids or solids, with the exception of drops or syrups containing vitamin or mineral supplements or medications [20]. The study’s outcome variable was prolonged exclusive breastfeeding (no, yes), which is defined as a child’s exclusive breastfeeding intake between the ages of 6 and 23 months [22].

Explanatory variables

The current study took into account both the individual and community levels in order to accommodate the hierarchical nature of DHS data. At the individual level, factors like child age (6–8 months, 9–11 months, 12–23 months), child sex (male, female), household wealth (poor, middle, rich), maternal education (no education, primary, secondary, higher), maternal age (15–24 years, 25–34 years, 35–49 years), marital status (unmarried, married), media exposure (no, yes), place of delivery (home, health facility), postnatal checkup (no, yes), breastfeeding initiation (≥ 1 h of birth, < 1 h of birth), drinking water source (unimproved, improved), sanitation facility (unimproved, improved), and antenatal care visits (< 4, 4–7, ≥ 8) were included. Community-level factors: place of residence (urban, rural), community literacy (low, high), community-level poverty (low, high), and community media exposure (low, high).

Description of explanatory variablesHousehold wealth

Categorized to three by combining poorest and poorer into one category, “poor,” middle wealth level into the second category, “middle,” and richer and richest into the third category, “rich.”

Drinking water source

Improved (use of piped water into dwelling, piped water to yard/plot, public tap/standpipe, tube well or borehole, protected well, protected spring, and rainwater collection) and unimproved (use of unprotected wells, unprotected springs, surface water (river, dam, lake/ponds/stream/canal/irrigation channel), tanker trucks, and carts with small tanks) [26].

Sanitation facility

Improved (flush or pour-flush to piped sewer system, septic tank or pit latrine, ventilated improved pit latrine, pit latrine with slab, and composting toilet) and unimproved (flush or pour-flush to elsewhere, pit latrine without slab or open pit, bucket, hanging toilet or hanging latrine, and no facilities or bush or field (open defecation) [26].

Media exposure

A variable that is coded as “yes” if the mother was exposed to at least one of these media and “no” otherwise. It is generated by combining the respondent’s preferences for reading newspapers or magazines, listening to the radio, and watching television.

Community media exposure

The percentage of women who have been exposed to at least one media outlet, such as a newspaper, radio, or television. It is classified as low (communities where ≤ 50% of women are exposed) or high (communities where > 50% of women are exposed) based on the national median figure.

Community literacy

The percentage of women who have completed at least primary school, as determined by survey respondents’ educational attainment. Next, it was divided into two groups based on the national median value: low (communities where ≤ 50% of women have completed primary school) and high (communities where > 50% of women have completed primary education).

Community poverty level

This was recoded as low and high community poverty level based on an aggregated variable from household wealth status: low (communities where ≤ 50% of women were poor) and high (communities where > 50% of women were poor).

Data management and analysis

STATA/SE version 14.0 statistical software was used to clean, recode, and analyze data that was taken from the most recent DHS data sets. To control for non-responses and sampling errors, a sample weight was used. After categorizing continuous variables, categorical variables underwent additional reclassification. The results were presented in frequencies and percentages using descriptive analysis. Descriptive statistical methods were used to portray the variables at the individual and community levels. The variables in the DHS data were arranged into clusters; households were nested within 1692 clusters, and 63,172 children are nested inside households. In order to use the conventional logistic regression model, the presumptions of independent observations and equal variance across clusters were broken. This suggests that accounting for between-cluster effects requires the use of a complex model. Multilevel mixed-effects logistic regression was therefore employed to identify the variables associated with PEB. The null model (outcome variable only), model I (only individual-level variables), model II (only community-level variables), and model III (both individual and community-level variables) are the four models that multilevel mixed effect logistic regression uses. The null model, which is devoid of independent variables, was employed to examine the variation in PEB within the cluster. Evaluations were conducted on the relationships between the outcome variable (Model I) and the factors at the individual and community levels (Model II). The link between the community- and individual-level variables and the outcome variable was fitted simultaneously in the final model (Model III). Through the use of the intra-class correlation coefficient (ICC) and proportional change in variance (PCV), the magnitude of the clustering effect and the extent to which community-level factors explain the unexplained variance of the null model were assessed. The best-fitting model was determined to be the one with the lowest deviance. Ultimately, factors were deemed statistically significant when they had a p-value of less than 0.05 and an adjusted odds ratio (AOR) with a 95% confidence interval (CI) associated with PEB.

Random-effect analysis results

The methods of estimating random effects or measures of variation of the outcome variable were the PCV, ICC, and median odds ratio (MOR). The variation between clusters was measured by the ICC and PCV. Taking clusters as a random variable, the ICC reveals that the variation of PEB between clusters is computed as ICC = VC/(VC + 3.29) ×100%. The MOR is the median value of the odds ratio between the area of the highest risk and the area of the lowest risk for PEB when two clusters are randomly selected using clusters as a random variable; MOR = ? 0.95√VC. In addition, the PCV demonstrates the variation in the prevalence of PEB explained by factors and computed as PCV = (Vnull-VC)/Vnull×100%, where Vnull = variance of the null model and VC = cluster level variance [27]. The fixed effects were used to estimate the association between the likelihood of PEB and individual and community-level independent variables.