Introduction

Malnutrition which “refers to deficiencies, excesses, or imbalances in a person’s intake of energy and nutrients” [1], has continued to be a public health concern world over and especially in developing countries [2]. Over 200 million children under-five are either undernourished or overweight [1]. Undernutrition in children is contributing about 45% of under-five mortality worldwide [1]. The World Health Organisation (WHO) member states recently ratified a commitment to nine global targets by 2025, including a 40% decrease in childhood stunting, a prevalence of childhood wasting of less than 5%, a guarantee that the number of overweight children will not rise, and the elimination of all forms of malnutrition by 2030 [3]. With less than three years till the goal date, the development has remained minimal, and none of the nine targets have been fully attained by any nation [4]. For instance, the case of ‘double burden of malnutrition’ (which goes beyond studying malnutrition as a separate health problem), is a condition where a child cohabits with undernutrition (such as stunting, and wasting) and overnutrition (such as overweight or obesity), is becoming a highly prevalent condition among children under-five years in Low and Middle Income Countries (LMICs) [5]. However, in the last two decades, stunting in children has witnessed significant reduction, while overweight is on the increase [3]. In Nigeria, the 2013 Nigeria Demographic and Health Survey (NDHS) reported that 37%, 18% and 29% of the children under-five years are stunted, wasted and underweight, respectively [4]. Understanding how different socio-economic, demographic, and contextual factors determine child malnutrition is a great way to improve the distribution of scare resources for the desired interventions. Variations also exist in the indicators of malnutrition as per the place of residence where urban children are likely to be more nourished than children in rural areas [6, 7]. To reach the nine targets, set forth by the WHO on malnutrition by 2025, researchers have committed to investigating malnutrition in its entirety, including the interconnection between undernutrition and overnutrition (particularly among children under the age of five years). A recent scoping review of predictors of malnutrition in its separate multifaceted form in SSA has identified both individual and household determining factors, and is reported elsewhere [8]. But, effective interventions cannot be created until a solid evidence base is built [9]. This study aims to assess the prevalence, and its individual and contextual predictors of malnutrition among children aged 6–59 months in Nigeria.

Methods

Source of data

This study is a secondary analysis of data from two separate, independently collected, nationally representative cross-sectional surveys. We took the Gender Inequality Index (GII), Human Development Index (HDI) and Multi-dimensional Poverty Index (MPI) out of the National Human Development Report (NHDR 2018) to use as the state variables and were included in the 2018 Nigeria Demographic and Health Survey (2018 NDHS), which is the primary data set. Using a two-stage stratified cluster design on each stratum obtained from the demarcation of the identified enumeration in the 2006 census, samples were independently chosen from each stratum in the NDHS. 1,400 Enumeration Areas (EAs) were chosen as sampling units at the first stage. At the second stage equal probability sampling was used to select 30 households at random from each EA, yielding a target sample size for the survey of 42,000 (30 × 1400) households. However, a weighted sample size of 7,770 children aged 6–59 months was used in the study.

Variables definition

Outcome or dependent variable

In NDHS, the malnutrition status (MNS) of children under-five years was determined through the measurement of anthropometric indices expressed as (i) Stunting: measured as height-for-age Z-score (HAZ) of less than minus two standard deviations from the median of the reference population. It is an indicator of growth retardation. (ii) Wasting: a measure of weight-for-height Z-score (WHZ) of less than minus two standard deviations from the median of the reference population. (iii) Underweight: a measure of weight-for-age Z-score (WAZ) of less than minus two standard deviations from the median of the reference population [10, 11]; these are measures of undernutrition. (iv) Overweight is also a measure of weight-for-height Z-score (WHZ) of more than plus (+) two standard deviations above the median of the reference population [11], considered as a measure of ‘over-nutrition’. The measure of malnutrition was considered as the outcome of either undernutrition or over-nutrition indicators. A recumbent length for children less than two years old was taken lying down, while children two years and above had their height standing up using Shorr Board measuring instrument. On the other hand, the weights of the children were measured with the SECA scale (model 878U) [12]. In this study, the composite index for malnutrition was computed using the four indicators (stunting, wasting, underweight, and overweight) as a measure of the overall description of the ‘double burden of malnutrition’ status among children aged 6–59 months in Nigeria. Children with no trace of anthropometric failure were classified as ‘0’, labelled as ‘well nourished’, and those that have at least one of the four indicators were classified as ‘1’, labelled as ‘poorly nourished’ [13–15].

Independent variables

The detailed descriptions and classifications of the predictor variables taken into consideration for this study have already been published elsewhere [8]. These variables were grouped into children/individual-related variables to include, child’s sex, age, perceived birth size, order, malaria status, nutritional status, fever, acute respiratory infection status, had diarrhoea, duration of breastfeeding, deworming, iron pills/syrup, and child took vitamin A; parental-related to include, place of delivery; preceding birth interval, maternal religious status, age group, age at first birth, educational status, working status, body mass index, anaemia status, autonomy level, marital status, ante-natal care visit, maternal ethnicity, religious status, maternal iron supplement during pregnancy, and paternal education status and work status; others are household-related to include, household socioeconomic status (wealth index), household size, number rooms for sleeping, number of under-five in the household, age and sex of household head, under-five slept under bed net last; and community-related variables to include, household region of residence, place of residence, community distance to health facility; and area-related variables to include, the state multidimensional poverty index, and the state human development index [8].

Statistical analysis

At the first level of analysis, charts and spatial map were used to describe the prevalence of malnutrition. At the second level, multivariate multilevel logistic regression models were fitted to determine the predictors of malnutrition status among 6–59 months of age in Nigeria due to the complex/hierarchical nature of the data sets, where children/individuals (at level 1) are nested in communities/clusters (at level 2) and nested in states (at level 3). The 3-level model was found to be more appropriate than the 2-level model and single-level model using a likelihood ratio test. Stata 16 SE was used for the computations in this study. Variables were considered significant at 5% level; otherwise, they are stated on the table.

Results

Prevalence of malnutrition

Four indicators were used to establish the nutrition status of the children: stunting, wasting, underweight, and overweight. The ‘composite of anthropometric failure’ was used to classify the children as poorly nourished if the child had any indicators and represented as ‘1’, and well-nourished otherwise. Figure 1 indicates that 56.6% (5931/10,481), 95%CI (55.63–57.53) of the children is well nourished. In addition, about 38% (3942/10,481), 95%CI (36.7–38.5) of the children aged 6–59 months in Nigeria are stunted, 7% (709/10,481), 95%CI (6.30–7.26) are wasted, 22% (2296/10,481), 95%CI (21.12–22.71) are underweight, while about 2% (172/10,481), 95%CI (1.42–1.91) are overweight.

Fig. 1
figure 1

Prevalence of malnutrition status by indicators

Full size image

Figure 2 presents the spatial variations in the proportion of children aged 6–59 months in Nigeria that were poorly nourished by states and Federal Capital Territory (FCT). The proportions of poorly nourished children were generally highest in the Northern Nigeria- Kebbi, Zamfara, Katsina, Kano, Jigawa, Bauchi and Gombe states. Edo and Lagos states are among the states with the least poorly nourished children.

Fig. 2
figure 2

Spatial maps describing the proportions of poorly nourished children by states & FCT

Source: Data computed from Nigeria DHS 2018.

Full size image

Multilevel analysis of malnutrition status

Multilevel multivariate models of predictors of malnutrition status

In the first instance, a multicollinearity test was conducted to check for highly correlated predictors. The outcome has already been reported elsewhere [16]. The backward and forward stepwise procedures were used to test all 48 variables that scaled through the multicollinearity checks. The removal was by p ≥ 0.20 for backward, and entry was p < 0.20 for stepwise forward methods, and then combined the outcomes of both forward and backward stepwise methods. Furthermore, the goodness of fits (Table 1) from these three methods using AIC and BIC shows that the combination of the backward and forward stepwise variables yielded the least AIC (9124.917).

Table 1 Evaluation of goodness of fit for variables selection methods in malnutrition
Full size table

Given above, 31 potential variables (child’s age, sex, birth size, preceding birth interval, child took iron syrup in the last six months before the survey, duration of breastfeeding, malaria status, anaemia status, cough status, diarrhoeal status, and child’s place of delivery, mother lives with partner, maternal education status, maternal ethnic group, maternal religious status, paternal education status, maternal body weight status, maternal anaemia status, household wealth, household head age group, number of bedrooms, low cluster wealth level, cluster distance to a health facility is no big problem, low cluster maternal education level, low cluster household with bed net, state human development index, state gender inequality index, region of residence, and place of residence), were finally retained for the multilevel model building.

Model set-up

Given the complex/hierarchical nature of the data sets, such that children/parental/household in individual units at Level-1 (since children from the same parent, and household tend to be more similar than children from other households because they share the same characteristics) [16], are nested in communities/clusters at Level-2 and nested in states at Level-3, multivariate multilevel logistic regression models were fitted to determine the predictors of malnutrition status among 6–59 months of age in Nigeria. Furthermore, a likelihood ratio test was carried out to establish that the three-level model was more appropriate than the two-level model (the likelihood-ratio test is LR χ2 = 448.73, p < 0.001 for Level-2 nested in Level-3).

Model building

The analysis built four multilevel logistic models. Model 1 is a null model (or empty model) with no predictors. The essence is to measure the variations across the communities and the states. Model 2 contains level-1 factors only. Model 3 contains level-1 and level-2 factors only (that is, community-related variables were added to Model 2); Model 4 (full model) was derived for all the selected variables, including the area-related variables (level-3 variables). Intra-class correlation coefficients (both the communities and the states), variance partition coefficients (both for the communities and states), and median odds ratios (MOR) were the post-estimation techniques used to evaluate the models. The goodness of fit was determined using Log-likelihood (LLH), Akaike’s information criteria (AIC) and Bayesian information criteria (BIC), such that the model with the highest LLH and lowest AIC and/or BIC was chosen as the best fit [16].

Multilevel model results

This section presents the results from the analysis using multilevel mixed-effects logistic regression. A weighted total sample of 7770 children was analysed. These result into 1361 communities (level 2) with an average of 6 persons per community, nested in 37 states (level 3) with the number of children ranging between 92 and 373 and on average of 210 children per state.

Measure of random effects

Model 1 is the null model (without predictors), and the random effect reveals that the proportion of the total variations attributable to differences in the communities and the states were, respectively, 0.322 (95%CI: 0.234 to 0.443) and 0.595 (95%CI: 0.367 to 0.966), while the variance attributable to individual level is 3.29 (2/3), which is fixed for logit. Therefore, the intrastate correlation coefficient of 0.1415 (95% CI: 0.092–0.210) and the intracommunity correlation coefficient of 0.218 (95% CI: 0.165–0.281) were used to evaluate the variations in the prevalence of malnutrition status among children aged 6–59 months in Nigeria due to the three-level factors. These results showed that respectively, state and community levels were responsible for 14.15% and 21.8% of the overall variation in the odds of malnutrition. As a result, the ICC and variance partition coefficient (VPC) at the state level are equivalent. The VPC at the community level, however, is 0.076, which means that a combined 7.6% of the overall variance is attributable to both the state and community levels.

AIC and likelihood ratios were used to evaluate the effectiveness of the models. Model 4 (complete model), with AIC = 9118.5 and log-likelihood = 4490.3, showed improvements in model fit. As a result, this study interpreted Model 4’s results. The ICC at the community level in the null model has decreased from 21.8 to 4.6% (95% CI: 17–25%) from the selected model (Model 4 in Table 2), meaning the correlation between two children/individuals (unit of analysis) within the same community and the same state is 0.046. The ICC at the state-level has also decreased from 14.1 to 0.1% (95% CI: 0.1–4.4%), both of which have maintained significance.

The variance partition coefficient (VPC) is the same as the ICC at the state level (0.002). Nevertheless, the VPC at the community level is 0.045, meaning that 4.5% of the total variance is collectively attributed to both the state and the community levels.

Similarly, in model 4, the median odds ratio (MOR) computed for states was 1.077, signifying that there exists a mean difference of the risk of being poorly nourished for two children with the same level-1 characteristics and picked randomly from two states. It means there is a 7.7% increased risk of a child being poorly nourished if he/she moves from one state to another state of increased risk of poorly nourished. Additionally, there is a 46% increased risk of a child being poorly nourished if he/she moves to another community with a higher risk of being poorly nourished.

Measures of fixed effects

Table 2 displays the outcomes of the adjusted odds ratios (AOR) for the variables included in the analysis after accounting for the other factors in the model. The three levels of variables—individual variables at level 1, community variables at level 2, and state/area variables at level 3—were all included in Model 4, which was the model that best suited the data. Variables at level-1 are child’s age, sex, birth size, preceding birth interval; the child took iron syrup in the last six months before the survey, anaemia status, the child had diarrhoea in the last two weeks before the survey, mother currently works, maternal education status, maternal ethnicity, paternal education status, maternal body weight status, household wealth, and several bedrooms in the household were statistically significant predictors of malnutrition among children 6–59 months of age in Nigeria. Among the community-related variables (level 2-related), cluster distance to a health facility is no big problem and was a statistically significant predictor of malnutrition. Gender inequality index and region of residence among the area variables were also significant predictors of malnutrition. On the other hand, the child who took iron syrup in the last six months before the survey, the child’s duration of breast-feeding status, malaria RDT status, the child had cough two weeks before the survey, and place of delivery were not statistically significant predictors of malnutrition. Also, maternal religious status, maternal anaemia status, household head age group, number of people in the household, community wealth status, the proportion of community maternal education status, proportion of community households with no net, and type of place of residence, were not statistically significant predictors of malnutrition among children aged 6–59 months in Nigeria.

The odds of female children being poorly nourished (AOR = 0.74, 95% CI 0.67–0.82) are significantly lower than their male counterparts. Also, the odds of children aged 24–35 months developing malnutrition were 2.22 times the odds of children 6–11 months of age (AOR = 2.22, 95% CI: 1.72–2.86). The smaller the birth size of the children is, the more likely they will be poorly nourished. Children who were born with average birth size (AOR = 1.26, 95% CI: 1.05–1.51), and born with small birth size (AOR = 1.79, 95% CI: 1.45–2.26), had 26% and 79% increased odds of contracting malnutrition when compared with children born with large birth size. Anaemic children (AOR = 1.33, 95% CI: 1.18–1.51) and those who had diarrhoeal (AOR = 1.27, 95% CI: 1.08–1.49) two weeks preceding the survey were more prone to being poorly nourished compared with children who do not have any of these conditions.

Table 2 Multilevel multivariate logistic models of predictors of malnutrition with adjusted odds ratios (AOR) among children 6–59 months in nigeria
Full size table

Similarly, children of working-class mothers have increased odds (AOR = 1.14, 95% CI: 1.01–1.29) of being poorly nourished compared with children whose mothers do not work. The educational levels attained by mothers and fathers are inversely proportional to the odds of their children being poorly nourished. Children of mothers with secondary education (AOR = 0.71, 95% CI: 0.58–0.86), tertiary education (AOR = 0.51, 95% CI: 0.38–0.70), and fathers with tertiary education (AOR = 0.78, 95% CI: 0.62–0.98), were respectively, 29%, 49%, and 22% reduced odds of being poorly nourished compared with children whose mothers and fathers do not have any formal education. The results also show that the richer the household, the less likely the children will be poorly nourished compared with their counterparts in the poorest household. Also, the odds of children residing in a 3-bedroom household (AOR = 1.23, 95% CI: 1.02–1.47) are 1.23 times more likely to be poorly nourished than children residing in a one-bedroom household. Furthermore, among the community-related variables, children from a community where the proportion of the community distance to the nearest health centre is not a big problem are high; there are 17% reduced odds of being poorly nourished compared with children from a community with a low proportion. Moreover, from the area-related variables, the odds of children from a state where the gender inequality index is low (AOR = 1.45, 95% CI: 1.14–1.85) is significantly 1.45 times more likely to be poorly nourished when compared with children from the state with lowest gender inequality index. The odds of children from the North-east (AOR = 2.27, 95% CI: 1.77–2.94), and the North-west (AOR = 3.11, 95% CI: 2.40–4.14), are 127% and 211% significantly more likely to be poorly nourished respectively.

Discussion

In the random effect analysis, the intrastate correlation coefficient of 0.142 and the intracommunity correlation coefficient (ICC) of 0.218 were used to assess the variations in malnutrition status among children aged 6–59 months in Nigeria due to the three-level factors. These results showed that 14.2% and 21.8% of the total variation in the odds of malnutrition were due to state and community levels, respectively. In addition, the ICC at the state level also decreased from 14.1 to 0.1%, which was statistically significant. Furthermore, the ICC at the community level in the null model decreased from 21.8 to 4.6% in the choice model, meaning the correlation between two children/individuals (unit of analysis) within the same community and the same state is 0.046. Similarly, model 4’s median odds ratio (MOR) calculated for states was 1.077, indicating that if a child travels from one state to another there is a greater chance of 7.7% of increased risk of being poorly nourished. In addition, moving to a community with a higher risk of malnutrition increases a child’s probability of being poorly nourished by 46%.

In the fixed effect analysis, the study found that female children have considerably decreased odds of being poorly nourished than their male counterparts. The same findings were reported in previous studies on malnutrition [8, 17–30], The risks of children aged 24–35 months cohabiting malnutrition were 2.22 times higher than those of children aged 6–11 months. Similar conclusions were reached in previous studies that an increase in the age of the children has a harmful effect on poor nourishment for the children [31–43]. The possible explanation for this difference is that as the child grows older, less attention is paid to their welfare compared with the youngest child. Also, children who had diarrhoea two weeks before the survey were more likely to be poorly nourished than children who did not have the illnesses. This finding is supported by previous studies [32, 33, 42,44,45,46]. Similarly, children of working-class mothers are more likely to be poorly nourished than children whose mothers do not work. This agrees with another study [47]. The possible reason for this is that working class mothers do not breast feed their children much longer compared to mothers who do not work. The odds of being poorly nourished are inversely proportional to the educational status that mothers and fathers have obtained. Other studies substantiated the result on maternal education [30,32,33,34, 36, 37,39,40, 42, 43,45,46, 47,49,50,51,52], and for paternal educational attainment was supported in [33, 39, 51]. The results also show that the richer the household, the less likely the children from such a home will be poorly nourished compared with their counterparts from the poorest household. This finding agrees with other studies [30,32,33,34, 35,37,38,39,40, 41,43,44,45,46, 48, 49,51,52,53,54], and contrary finding with [55].

In addition, children from communities where many residents do not consider distance to the nearest health facility to be ‘no big problem’ are 17% less likely to be poorly nourished than children from communities with a low proportion do exist. The odds of children from a state with a low gender inequality index being poorly nourished are significantly 1.45 times higher than those with the highest gender inequality index. The odds of children from the North-east, and the North-west, are 127% and 211%, significantly more likely to be poorly nourished, respectively. Children living in urban areas are more protected from being malnourished than children from rural areas. This result agrees with other previous studies [36, 38, 42, 44, 46, 48, 50, 51].

Strengths and limitations of the study

This study is based on the analysis of the merged data sets whereby contextual variables from 2018 NHDR were incorporated into the 2018 NDHS. Moreover, the study took into consideration the hierarchical nature of the data set and conducted a 3-state multilevel mixed effect logistic analysis to account for individual, community, and state variabilities. In addition, the study recognised overweight as part of the indicators (stunting, wasting, and underweight) that formed the proxies for malnutrition, otherwise known as the index of double burden of malnutrition.

There are several restrictions on this study. First, because the survey was cross-sectional, the study could only look at the relationships between factors. Consequently, it was impossible to determine causality. It’s also important to note that the random coefficient model in a three-level model was not examined in this study to see if it was superior to the variance component and random intercept model. Additionally, only the individual level of weighting was considered. These are appropriate for further research.

Conclusion

In general, the Northern Nigerian states of Kebbi, Zamfara, Katsina, Kano, Jigawa, Bauchi, and Gombe had the highest proportions of poorly malnourished children aged 6–59 months. Most of these areas are highly affected by high rates of constant insecurities that have distorted farming activities [56]. Efforts should be made to end these age long insecurities in these areas. The results reported female children are well-nourished compared to their male counterparts. Therefore, gender-sensitive policies and community mobilisation against gender-based bias are required to overcome the discrepancies in malnutrition status between male and female child.