Recently collected standard cross-sectional DHS data set of the SSA countries within 10 years (2010–2020) were used as our data source. To obtain a representative sample of the recent standard DHS data from each region of SSA, 10 years of DHS data (starting from 2010) were taken for analysis. The standard DHS data set was used for each country to get all parameters and a large sample size, which can be representative of the source of population and have a particular advantage in a pooled analysis (7).

The sub-Saharan is the area in the continent of Africa that lies south of the Sahara and consists of four vast and distinct regions, i.e., Eastern Africa, Central Africa, Western Africa, and Southern Africa. A total of thirty-three SSA countries were represented in this study in the four regions, namely, Eastern Africa (Burundi, Comoros, Ethiopia, Kenya, Malawi, Mozambique, Rwanda, Tanzania, Uganda, Zambia, Zimbabwe), Southern Africa (Lesotho, Namibia, South Africa), Central Africa (Angola, Cameroon, Chad, the Democratic Republic of the Congo, Republic of the Congo, Gabon), and Western Africa (Benin, Burkina Faso, Ivory Coast, Gambia, Ghana, Guinea, Liberia, Mali, Niger, Nigeria, Senegal, Sierra Leone, Togo) (29) (Figure 2). Together, they constitute an area of 9.4 million square miles and a total population of 1.1 billion inhabitants (30). The datasets are publicly available from the DHS website www.dhsprogram.com (29). DHS collects data that are comparable across countries. The surveys are nationally representative of each country and population-based with large sample sizes. All surveys use a multistage cluster sampling method (7).

All children aged 6–23 months preceding 5 years of the survey period across 33 SSA countries were our source population, whereas children aged 6–23 months in the selected Enumeration Areas (EAs) or primary sampling units of the survey clusters were our study populations. The mother or the caregiver was interviewed for the survey in each country and mothers who had more than one child within the 2 years preceding the survey were asked questions about the most recent child (6). Moreover, from the data set of included countries, children in the age category of 6–23 months, which is not assessed for MDD based on the DHS guideline, and having the missing value of the outcome variable, were excluded.

Of the total of 47 countries located in SSA, only 41 countries had Demographic and Health Survey Report. Of these, five countries, namely, Central Africa Republic (DHS report 1994/95), Eswatini (DHS report 2006/07), Sao Tome Principe (DHS report 2008/09), Madagascar (DHS report 2008/09), and Sudan (DHS report 1989-90) have a survey report before the 2010 survey year and were excluded from further analysis. Moreover, three countries (Botswana, Mauritania, and Eritrea) were excluded due to the DHS dataset not being publicly available. Finally, a total of 33 SSA countries were included in this study.

In general, the most recent standard census frame was used in all of the surveys conducted in the selected countries. Typically, DHS samples are stratified by administrative geographic region and by urban/rural areas within each region. DHS sample designs are usually two-stage probability samples drawn from an existing sample frame. In the first stage of sampling, EAs were selected with probability proportional to size within each stratum. In selected EAs, a fixed number of households is selected by the systematic sampling method in the second stage of sampling. Finally, a fixed number of households is selected by equal probability systematic sampling in the selected cluster. The detailed sampling procedure was available in the Measure DHS program (7).

The children’s records or kid’s records (KR) DHS datasets were used. Weighted values were used before using the DHS dataset to restore the representativeness of the sample data, as the overall probability of selection of each household is not constant. DHS guidelines set four sampling weighting methods, and from that, we used the individual weight for women (v005), which is the household weight (hv005) multiplied by the inverse of the individual response rate for women in the stratum. Individual sample weights are generated by dividing (v005) by 1,000,000 before using it to approximate the number of cases (31). From 81,123 total eligible households, 78,483 had complete data for inadequate MDD intake with a response rate of 95.5%. Finally, a total weighted sample of 77,887 children in the age category of 6–23 months from all 33 countries was included in this study (Table 1).

Individual and community-level independent variables have been studied. The individual-level factors included sociodemographic characteristics such as the age of the mother, mother’s employment, marital status, family size, maternal education, and household wealth status. Child-related factors such as the age of the child, sex of the child, birth order, the plurality of birth, and breastfeeding status of the child are all taken into account.

Behavioral characteristics like media exposure were studied. Media exposure status was created from the frequency of watching TV, listening to the radio, and reading a newspaper or a magazine. If a woman has at least one yes, she has considered media exposure. Health service utilization-related factors such as the place of delivery and antenatal care (ANC) visit were also considered.

The community-level factors included the place of residence, region in SSA, survey year, income level of the country, community-level media exposure, and community-level women’s education (Table 3).

This study was performed based on the DHS data obtained from the official DHS measure website www.measuredhs.com after permission has been obtained via an online request by specifying the objectives. Standard DHS datasets were downloaded and then cleaned, integrated, transformed, and appended to produce favorable variables for the analysis. The Microsoft Excel and STATA 16 software were used to generate both descriptive and analytic statistics of the data of appended 33 countries to describe variables in the study using statistical measurements. The pooled estimate of inadequate MDD intake among children in SSA and sub-group analysis was estimated using a metan STATA command.

The DHS data have a hierarchical nature, and children aged 6–23 months and mothers were nested within a cluster. This might violate the standard logistic regression model assumptions such as the independence and equal variance assumptions. Therefore, a multilevel binary logistic regression model was fitted. Four models were fitted for multilevel analysis. The model without exposure variables (null model) was used to check the variability of inadequate minimum DD intake across the cluster. The association of individual-level variables with the outcome variable (Model I) and the association of community-level variables with the outcome variable (Model II) were assessed. In the final model (Model III), the association of both individual and community-level variables was fitted simultaneously with the prevalence of inadequate minimum DD intake.

Random effects or measures of variation of the outcome variables were estimated by the median odds ratio (MOR), intraclass correlation coefficient (ICC), and proportional change in variance (PCV). Taking clusters as a random variable, the MOR is defined as the median value of the odds ratio between the area at the highest risk and the area at the lowest risk area when randomly picking out two clusters: MOR=e0.95V⁢A. In contrast, the ICC reveals the variation of inadequate minimum DD between clusters is calculated asI⁢C⁢C=V⁢AV⁢A+3.29*100%. Moreover, the PCV reveals the variation in the prevalence of inadequate minimum DD intake among children aged 6–23 months explained by factors and calculated as P⁢C⁢V=V⁢n⁢u⁢l⁢l-V⁢AV⁢n⁢u⁢l⁢l*100%, where Vnull = variance of the initial model and VA = area/cluster level variance (35–37).

The fixed effects or measure of association was used to estimate the association between the likelihood of prevalence of inadequate minimum DD intake and individual and community-level explanatory variables. It was assessed and the strength was presented using adjusted odds ratio (AOR) and 95% confidence intervals with a p-value of < 0.05.

where πij is the probability of inadequate minimum DD intake use and 1−πij is the probability of adequate minimum DD intake use. ß0 is the intercept that is the effect of feeding inadequate minimum DD when the effect of all explanatory variables is absent. β1xij are individual and community-level variables for the i^th individual in group j, respectively. The ß’s are fixed coefficients indicating that a unit increase in X can cause a ß unit increase in probability feeding inadequate minimum DD. The uj shows the random effect (effect of the community/clusters on the mother’s decision to provide inadequate minimum DD) for the j^th clusters (35, 37, 38).

Model comparisons were done using the deviance test and log likelihood test and the model with the highest log-likelihood ratio and the lowest deviance was selected as the best-fitted model. The variance inflation factor (VIF) was used to detect multicollinearity, and all variables had VIF values less than 10 and the mean VIF value of the final model was 1.54.

The concentration index and concentration graph approach are used to examine socioeconomic inequalities in inadequate minimum DD intake (39, 40). The concentration curve is used to plot the cumulative percentage of the health variable (y-axis) against the cumulative percentage of the population ranked by living standards beginning with the poorest and ending with the richest (x-axis) (41). The concentration index is used to quantify and compare the degree of socioeconomic-related inequality in inadequate minimum DD intake (42, 43). The concentration index is twice the area between the concentration curve and the diagonal line and ranges from –1 to + 1. The sign of the concentration index indicates the direction of the relationship between the health variable (inadequate minimum DD) and the distribution of living standards (wealth status) (40, 41), whereas the magnitude of the concentration index reflects both the strength of the relationship and the degree of variability in the health variable (41, 44).

A total weighted sample of 77,887 children aged 6–23 months were included in this study. Three-fourths (75.77%) of mothers/caregivers were found in the age group of 20–35 years, with a median age of 27 (IQR: 22, 33) years. More than three-fifths of women (62.26%) had formal education. Most (70.52%) of the mothers were not working. Moreover, 67.01% of children were delivered at the health facility.

Almost equal proportions of male (50.46%) and female (49.54%) children were studied. Nearly two-thirds (65.25%) of the children were found in the age group from 12 to 23 months with a median age of 14 (IQR: 10, 18) months. Almost all (96.94%) the newborn children were singleton. From the community-level variables, most (65.36%) of the respondents were rural inhabitants of whom 80.71% had inadequate MDD intake. Notably, 64% of the SSA countries included in the study were lower-income countries (Table 4).

The overall pooled estimate of inadequate MDD intake among children aged 6–23 months in the SSA countries was 76.53% (95% CI: 73.37–79.70), with I² = 99.2% and ranging from 50.5% in South Africa to 94.40% in Burkina Faso (Figure 1).

As the I² value was large, which shows the true variabilities (the variability not by chance) of inadequate MDD intake among 33 countries, further subgroup analyses were performed to treat this heterogeneity effect based on the region in SSA, level of income of the country, and the DHS survey year. Based on the subgroup analysis using regions in SSA, the pooled magnitude of inadequate intake of MDD ranges from 66.84% (95% CI: 49.51–84.17) in Southern Africa across three countries to 80.94% (95% CI: 75.66–86.22) among 13 West Africa countries.

Moreover, the pooled magnitude of inadequate intake of MDD across country income levels was determined. Among 21 low-income level countries, the pooled estimate of inadequate MDD intake was 76.53% (95% CI: 73.37–79.70), whereas it was 64.91% (95% CI: 56.99–72.83) across four upper middle-income countries. In addition, the pooled estimate of inadequate MDD intake in 15 countries whose DHS survey conducted before and in 2015 was 78.44% (95% CI: 72.88–84.00), whereas it was 74.97% (95% CI: 71.77–78.18) in 18 countries whose DHS survey was conducted after 2015 (Table 5).

The magnitude of inadequate dietary intake for each seven food groups was present in the following figures. Most of the children (86.95%) and four of five children (80.78%) children aged 6–23 months in SSA cannot access eggs and other fruits, respectively (other than vitamin A rich fruits listed in the DHS guidelines). But children aged 6–23 months in SSA have better access to grains, roots, and tuber food groups (74.74%) (Figure 2).

In the multilevel analysis, the random effect analyses ICC, PCV, and MOR were assessed. The ICC in the null model indicates that approximately 16% of the variations in inadequate MDD intake among children aged 6–23 months were attributed to cluster differences. The MOR value of 2.05, in the null model, also revealed that the odds of inadequate MDD among study participants were two times different between lower and higher risk clusters. Furthermore, the PCV valve in the final model indicates that about 16% of the variation in inadequate MDD intake among study participants was explained by the final model (model four). As the model is nested, log-likelihood and deviance tests were used for model comparison. The model with the highest log-likelihood and the lowest deviance was the better-fitted model, which was model four (72,558) and selected for interpretation (Table 6).

Based on the multivariable analysis of the final selected model (model III) results, the chances of having primary and above primary educated women were 8 and 34% less to have inadequate minimum DD intake of their children than women with no formal education (AOR = 0.92; 95% CI; 0.86–0.96) and (AOR = 0.66; 95% CI; 0.62–0.70), respectively. The odds of having children with inadequate minimum DD were 24% reduced among women who were working than those who were not working (AOR = 0.76; 95% CI; 0.72–0.79). Women who were not married were 1.27 times more likely to have children with inadequate minimum DD as compared to those who were married (AOR = 1.27; 95% CI; 1.21–1.33).

The odds of having children with inadequate minimum DD intake were 31% less likely among households, which have exposure to at least one media (radio, TV, or newspaper) than those not exposed (AOR = 0.69; 95% CI; 0.66–0.72). When we consider from poorer to the richest households, inadequate minimum DD intake among children decreased by 12% (AOR = 0.88; 95% CI; 0.83–0.94), 23% (AOR = 0.77; 95% CI; 0.72–0.82), 35% (AOR = 0.65; 95% CI; 1.61–0.69), and 56% (AOR = 0.46; 95% CI; 0.43–0.50) for poorer, middle, rich, and richest households, respectively. Moreover, children from lower middle-income and higher middle-income level countries are also less likely 21 and 58% to have inadequate minimum DD than a child from lower-income level countries (AOR = 0.79; 95% CI; 0.75–0.83) and (AOR = 0.42; 95% CI; 0.38–0.46), respectively.

Children aged 12–23 months were 0.69% less likely to have inadequate minimum DD as compared to a child with 6–8 months of age (AOR = 0.31; 95% CI; 0.29–0.33). Multiple births were 23% less likely to have inadequate minimum DD than those who were singleton (AOR = 0.77; 95% CI; 0.69–0.87). The odds of having inadequate minimum DD among children who are currently breastfed were 1.53 times higher than those who are non-breastfed (AOR = 1.53; 95% CI; 1.46–1.60).

The odds of having inadequate minimum DD of the child were 26 and 13% less likely among mothers who have at least one ANC follow-up and delivery at a health institution (AOR = 0.74; 95% CI; 0.69–0.78) and (AOR = 0.87; 95% CI; 0.83–0.91), respectively.

Children who live in rural residences were 1.29 times more likely to have inadequate minimum DD access than those who live in urban areas (AOR = 1.29; 95% CI; 1.23–1.36). Children who live in the East Africa region were also 41% times less likely to have inadequate minimum DD as compared to those who live in Central Africa (AOR = 0.59; 95% CI; 0.55–0.63) (Table 6).

In this study, the overall wag staff normalized concentration index (C) analyses of the wealth-related inequality of inadequate minimum DD intake showed the pro-poor distribution with (C = −0.24; 95% CI: −0.22, −0.0.26). This shows that inadequate minimum DD intake among children aged 6–23 months was disproportionately concentrated in the poor groups (pro-poor). The concentration index is twice the area between the concentration curve and the diagonal line. Then, when multiplying the C by 75, (−0.24 × 0.75) = 0.18 showed that 18% of the inadequate minimum DD intake would need to be (linearly) redistributed from the richer half to the poorer half of the population to arrive at a distribution with an index value of zero (perfect equality). Even though the wealth-related inequalities of inadequate MDD intake are high among urban residents than rural, it is not statistically significant (C = −0.24; 95% CI: −0.22, −0.0.26) with a p-value = 0.175. The finding from the indices is in agreement with the results of the concentration curves which showed that the concentration graph of inadequate minimum DD usage was above the line of equality, which indicated that the distribution of inadequate minimum DD used by children was concentrated in poor households (pro-poor distribution) (Figure 3).