Individual and Contextual Factors Associated with Malaria among Children 6–59 Months in Nigeria: A Multilevel Mixed Effect Logistic Model Approach

Background/Purpose: Over the last two decades, malaria has remained a major public health concern worldwide, especially in developing countries leading to high morbidity and mortality among children. Nigeria is the world most burdened malaria endemic nation, contributing more than a quarter of global malaria cases. This study determined the prevalence of malaria among children at 6–59 months in Nigeria, and the effects of individual and contextual factors. Methods: This study utilized data from 2018 Nigeria Demographic and Health Survey (NDHS) involving a weighted sample size of 10,185 children who were tested for malaria using rapid diagnostic test (RDT). Given the hierarchical structure of the data set, such that children at Level-1 were nested in communities at Level-2, and nested in states and Federal Capital Territory (FCT) at Level-3, multilevel mixed effect logistic regression models were used for the analysis. Results: The proportion of children 6–59 months of age in Nigeria that had malaria fever positive as assessed by RDTs was 35.5% (3418/10,185), (CI: 33.9–37.1). Kebbi State had 77.7%, (CI: 70.2–83.5), which was the highest proportion of 6–59 months who were malaria positive, next in line was Katsina State with 55.5%, (CI: 47.7–63.1). The Federal Capital Territory (FCT), Abuja had the proportion of 29.6%, (CI: 21.6–39.0), malaria positive children of 6–59 months of age. Children between the age of 48 and 59 months were 2.68 times more likely to have malaria fever than children of ages 6–11 months (AOR = 2.68, 95% CI: 2.03–3.54). In addition, children from the rural area (AOR = 2.12, 95% CI: 1.75–2.57), were more likely to suffer from malaria infection compared to children from urban area. Conclusion: The study identified some individual and contextual predictors of malaria among children in Nigeria. These factors identified in this study are potential areas that need to be considered for policy designs and implementations toward control and total elimination of malaria-related morbidity and mortality among children in Nigeria.


Introduction
The past twenty years and up till now, malaria has persisted as a major global public health concern [1,2], with over 300 million cases reported in 2018 [3] and has remained a leading cause of morbidity and mortality. Low and medium-income countries (LMIC), especially in Sub-Sharan Africa (SSA), contribute to more than 80% of the global malaria burden [4,5]. Nigeria, with a population of over 200 million people, is the world most burdened malaria endemic nation, contributing more than a quarter of global malaria cases. The risk of malaria infection cut across all age segments with women, (especially the pregnant), and children (especially under-five years), the most vulnerable. Malaria is a deadly disease that kills an estimated number of 30 children every hour worldwide. There were

Sampling Techniques
NDHS, being the primary data set for this study, samples were selected separately from each stratum using a two-stage stratified cluster design on each stratum derived from the 2006 census identified enumeration demarcation. At the first stage, 1400 enumeration areas (EAs) were selected as sampling units. At the second stage, 30 households were randomly selected from each EA using equal probability sampling, leading to a target total sample of 42,000 (30 × 1400) households used for the survey. In total, 11 EAs were excluded from being captured because of in-security. Therefore, a total of 41,668 households was earmarked for sampling, but only 40,427 households representing a response rate of 99.4% were finally captured in the survey [11].
The target groups for the survey were women age 15-49 years from all the randomly picked households and men age 15-59 years in one-third of all randomly selected households across Nigeria [11]. Besides, children 6-59 months in 14,000 households (i.e., in the households where men questionnaire was administered) had their blood sample taken for malaria (via rapid diagnostic test-RDT), anaemia, and genotype testing. In addition, 75% of the children tested for malaria using RDT, were randomly selected for confirmatory test for malaria using macroscopic blood smear in a laboratory. Only 97.2% of all eligible children for RDTs test were successfully captured, resulting to 2.8% missing values. In this study therefore, the unit of analysis was children 6-59 months of age in Nigeria. Further details of sampling procedures have been published elsewhere [11].

Outcome Variable
In 2018 NDHS diagnostic tests for malaria parasite were carried out for children 6-59 months of age in approximately a third of the selected households. Children in this age bracket have not developed enough immunity as such and they are more prone to contracting malaria [29]. Two testing methods were adopted: (i) Rapid diagnostic tests (RDTs) were conducted on blood samples taken from pricking the finger or heal of the child using SD Bioline Ag Pf (HRP-II). The RDTs detect the Histidine-rich protein-II (HRT-II) human whole blood (antigen). The results were either classified as positive or negatives for Pasmodium falciparum (Pf). (ii) Laboratory microscopy investigation on thick blood smears was performed for a three quarter of the households where RDTs was done. Malaria results were also classified as either positive or negative.
There was a strong positive correlation 0.581 (95% CI: 0.57-0.60; p < 0.0001), between these two test results. However, the microscopy laboratory test was performed essentially as a confirmatory test for the RDT that was carried out on the field [9][10][11]. Therefore, in this study, the malaria status of children 6-59 months of age in Nigeria using RDTs was classified as "one" if the result was positive and "zero" when the result was negative.

Independent Variables
The predictor variables considered for this study were identified from previous scoping reviews [1,30,31], and categorized in line with previous studies, and as reported in NDHS 2018 final report [21]. Two files in 2018 NDHS, children under-five years (KR), and household members (PR), were merged using a common identifier to obtain variables that meet this analysis [32]. The health status (healthy or not) of a child in a population is a function of interrelated factors at both individual and environmental level [33]. These factors are called determinants and are divided into three main groups: individual characteristics; physical and social; and health services [33]. As for individual determinants, these are on one hand, biological uncontrolled inherent traits from birth that distinguish the health status of one child from another, such as age, sex, parental affiliation. On the other hand, they could be behavioural factors that pertains to the child which are subject to modification through some control measures such as diet, immunization, or food supplements, etc. Other determinants of child's health outcomes relate to the household's physical and social environment [33]. In this study, the determinants of the risk of malaria infection among children 6-59 months of age in Nigeria were grouped into child-, parental-householdcluster-, and area-related variables. The definitions and classifications of these variables are given as follows:

Child-Related Variables
Age of the child was classified into quintiles of one year interval each, with the reference group (6-11 months) corresponds to the period the child had not begun to walk, 12-23 months is another important landmark (completion of immunization), by 24-36 months, some children would have had another sibling following, 36 months and above connotes immediate preschool age. Further, sex of the child, maternal perceived birth size of the child classified into, large, average, and small categories. Birth order of the child defines a child's rank among other children of same mother. However, the effect of this on child's health outcome is not well established [34]; duration of breastfeeding, breast milk is very important to a child in the first few years of life [35]. It gives the child all the nutrients needed, and thus strengthens the child's natural immunity against diseases, this was classified as: (i) ever breastfed, not currently breast-feeding (ii) never breastfed (iii) still breastfeeding; whether the child had taken iron, vitamin A, and deworming treatment in the last 6 months before. In addition, also considered were if a child had diarrhoea, fever, or acute respiratory infection (ARI), 2 weeks before survey. Nutritional status was derived through a composite index computed from the four nutrition indicators (stunting, wasting, underweight, and overweight), using the composite index of anthropometric failure (CIAF) [36][37][38]. Children with no trace of anthropometric failure were classified as "well-nourished" and those that have at least one of the four indicators were classified as "poorly nourished"; anaemia status derived from the record of whether the child was mildly, moderately, or severely anaemic and were collapsed into one group, such that a child with haemoglobin level less than 11.0 g per decilitre was regarded as being "anaemic" and classified as "one", otherwise, for 11.0 g or more was "not anaemic" and classified as "zero".

Parental-Related Variables
Mother's age (year) at last birthday was classified as reported in similar studies to distinguish younger mothers (below 25 years), middle aged, and older mothers, above 34 years [39][40][41]. The mother's age when she had her first child was also important. Younger mothers lack baby-care experience, and their babies are very likely to have poor health outcomes. Age groups of mothers are often supported in 10 s of years brackets [42], age at first birth was classified as (i) 10-19 years, (ii) 20-29 years, and (iii) 30 years+. Mother working status, paternal working status, and mother's educational status is very important in maternal and child health. Maternal literacy can enhance the level of knowledge, altitude, and practice (KAP) of some common childhood diseases. Paternal education status is worthwhile investigating the role it plays in child's health outcome. This is likely going to be moderating the effect of maternal characteristics on child's health. Both maternal and paternal education statuses were classified as: (i) no education, (ii) primary, (iii) secondary and above; marital status (never in union, in union, widow/divorced/separated); mother slept under mosquito net; mother's body weight status (kg/m2), (i) underweight when BMI <18.5 (kg/m2), (ii) healthy when BMI is 18.5-25.0 (kg/m2), and (iii) overweight/obese when BMI >25.0 (kg/m2). Mother's anaemia status, and maternal anaemia statuses (whether mildly, moderately, or severely anaemic), were collapsed into one group, such that mild, moderate, or severe were regarded as "anaemic" and classified as "one", otherwise, it was "not anaemic" and classified as "zero". Number of antenatal care visit mother attended during the child's pregnancy was also provided. The WHO as at the time of the survey for this study recommended a minimum of four visits [11,43]. Maternal autonomy was derived using the composite index score from the mother's level of participation in decision concerning her own health, large purchases, and visits to family [36]. The score ranged from three to six such that three and four represent "less" autonomy, and five and six represent "more" autonomy. Maternal ethnicity affiliation was classified into four groups; and maternal religion status, since it defines people's belief system and can affect several maternal and child's health-related outcomes [44].

Household-Related Variables
Household wealth status is the measure of household economic status derived as the composite score from some durable goods using principal component analysis (PCA). The wealth status was originally classified into quintiles. In this study, three categories were derived, such that poor and poorest groups were collapsed into poor, while rich and richest were collapsed into rich categories [41,45]; household had mosquito bed net; household number size measures the number of people that stay in a household, and was originally a scale variable, but for the purpose of this analysis, the variable was classified into four categories in line with average typical family sizes of four to five members in Nigeria [11]; number of bedrooms in household, though in the survey the number of rooms available for sleeping in a household were given in scale values with less than 10% indicated having five or more rooms for sleep. A typical building in Nigeria has either one-, two-, three-, or four-bedrooms, hence in this analysis, we categorized the variables into five groups; number of children under-5 year in household were also treated as categories as in previous studies [2,45]; improved source of drinking water, toilet facilities, floor materials, roofing materials, and wall materials. These materials are either natural, rudimentary, or finished. Unimproved are the natural and rudimentary, and the improved are the finished materials [46]; sex of household head; household head age (years) in groups. In some developing countries, the age status of the household head is very important as it relates to maturity with which final decision making are taken, particularly on the issue of health. This was also classified as, ((i) less than 34 years, (ii) 35-44 years, (iii) 45-55 years (iv) 56 years +); whether the household had electricity. The presence or absence of electricity in household can impact food storage, cooling at certain season can affect the health of children in particular; type of cooking fuel; under-five slept under a bed net last night, (i) no child, (ii) all children, (iii) some children, (iv) no net in the house [11].

Cluster-Related Variables
Proportion of cluster's household with no bed net; distance to a health facility is no big problem; and proportion of low cluster wealth status. These were derived from the household characteristics. The classification into low and high was based on the cut-off point of 50th percentile (median) score for each variable [47].

Area-Related Va31riables
Region of residence are the six geopolitical zones in Nigeria where the child resides in. There is seasonal variation across the zones, and this can impact differently on health outcome particularly on children under-five years; place of residence is the type of place where the child's household is located and is classified as either urban or rural; state human development index (SHDI). The human development index by each state defines the average on key indicators of human development of the communities. The higher the HDI, the more protective the children from these communities will be from diseases. These categorizations of HDI by states in this study as given in NHDR 2018 [28]; state multidimensional poverty index (SMPI), is the measure of the socioeconomic status of the state and is taken as proxies to the multidimensional poverty index of the communities. The highly deprived a community's MPI is, the more harmful the people are to diseases. This categorizations of HDI by states are also given in NHDR 2018 [28].

Statistical Analysis
Descriptive analysis using percentage frequencies was used to establish the prevalence, distribution, and association of the malaria status among children 6-59 months of age in Nigeria with the predictor variables considered in this study. "Svyset" command was used to adjust for under-and over-reporting in the survey using a weighting factor of (v005/1000000), where v005 is the sample weight [45,48]. 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, are nested in communities/clusters (at Level-2), and nested in states (at Level-3), multiple multilevel logistic regression models were fitted to determine the predictors of malaria status among 6-59 months of age in Nigeria. 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 χ 2 = 30.21, p < 0.001 for Level-2 nested in Level-3).
In handling the missing values, the listwise deletion technique was applied. In the first instance, all variables having more than 20% missingness were removed, furthermore, any variable with incomplete observations, and those with responses "I don't know" or refusal to answer the questions were deleted [45].

Multilevel Model Description for the Three-Level Survey on Malaria Status
The dependent variable of interest is binary and follows the Bernoulli (π ijk ) distribution with a logit link function: where η ijk is the predicted log odds of individual child i (Level-1) in community (com) j (Level-2), and in state (sta) k (Level-3) having a positive RDT for malaria. β * 0,0 represents the overall intercept (the grand mean of Level-3), β * a,0 , β * b,0 , and β * c,0 are respectively the mth, the nth, and the pth coefficients associated with W (Level-1), X (Level-2), and Z (Level-3) predictors. Further, ε 0,jk represents the random effect of jth community in kth state, while ε 0,k denotes the state-level random effect, with the assumption that ε 0,jk ∼ N 0, σ 2 com and ε 0,jk ∼ N 0, σ 2 sta are identical and independently distributed [49,50]. Equation (1) has a logistic transformation, with and it denotes the probability that an ith child in the jth community and the kth state will be RDT malaria fever positive.

Model Building
In this study, five multilevel logistic models were considered. Model 1, a null model (or empty model), without any predictors. The essence is to measure the variation across the communities and the states. Model 2 included only child-related variables; Model 3, adjusted for/parental-related variables, while for Model 4, household-related variables were added to Model 3; Model 5 (full model) was derived for all the selected variables including the area-related variables. Goodness of fit was determined using Akaike's information criteria (AIC), such that the model with the lowest AIC was chosen as the best fit [2].

Measure of Association
The measures of association (i.e., fixed effects) were described using adjusted odds ratio (AOR) with their corresponding p-values and 95% confidence intervals (CIs).

Measures of Variations
The measures of variation (i.e., random effects) were captured using intra-cluster correlation (ICC), and variance partition coefficient (VPC).

Intraclass Correlation Coefficient (ICC)
Intraclass correlation coefficient (ICC) represents the proportion of the total variation in the model that can be accounted for by variations across the different level of clusters. In our model (three-level model), we identified two intraclass correlation coefficients: the one pertaining to children/individuals nested in community-level, and community-level groups nested in the state-level group [50,51]. Therefore: ICC com is the correlation between two children/individuals (unit of analysis) within the same community and the same state [52,53].
However, Equation (3), in terms of the variance partition coefficient (VPC) differs, as it does not have corresponding interpretation, therefore: refers to the proportion of the total variance in the same state, but different communities [52] ICC sta is the correlation between two children/individuals within the same state, but different in community clusters. In VPC, it refers to the proportion of the total variance that is attributable to between state-level [52].
From (3), (4), and (5), σ 2 com is across community variance, σ 2 sta is the across the state variance, and π 2 3 3.29 is between children/individuals' variance with scale factor one, and for logistic regression [49]. The values of ICCs help to establish the need for multilevel analysis as against the single-level analysis. The rule of thumb could be that when the ICC is less than 5% at the null model, hierarchical modelling may not be necessary [54].
All computations were performed in Stata 16 SE (StataCorp LP: College Station, TX, USA). In recognition of the complexity of the survey design, weight proportion as specified in Stata was used to account for over-and under-estimation. The listwise deletion, the default missing values handling technique in Stata was applied.

Prevalence of Malaria Fever
There was a total of unweighted sample of 10,152 and weighted sample of 10,185 children 6-59 months of age in Nigeria considered in this analysis. In Figure 1, the proportion of children 6-59 months of age in Nigeria that had malaria fever positive as assessed by RDTs was 35.5% (3418/10,185), (CI: 33.9-37.1). The proportion of six states from the northern part of Nigeria, and one state from the southwest are in the highest quintile proportion group of malaria among children 6-59 months in Nigeria. Of these seven states   Table 1 displays the descriptions of the background characteristics, the associatio with malaria fever status of children 6-59 months in Nigeria and presented under; child , parental-, household-, and area-related factors, respectively. In (a) part of the table, ther were more children 12-23 months of age, 23.8% (2421/10,185) among the age group males. In total, there were 51.2% (5216/10,185) compared to females, and 1.7% (171/10,185 were never breastfed. The prevalence of malaria fever among children 24 months an above is more than the national prevalence of 35.5%. Furthermore, the result shows tha malaria status was strongly associated with child's, age, birth order, vaccination, malari status, nutritional status, fever, diarrhoea, duration of breastfeeding, deworming, iro pills/syrup, and vitamin A intake.  Table 1 displays the descriptions of the background characteristics, the association with malaria fever status of children 6-59 months in Nigeria and presented under; child-, parental-, household-, and area-related factors, respectively. In (a) part of the table, there were more children 12-23 months of age, 23.8% (2421/10,185) among the age groups, males. In total, there were 51.2% (5216/10,185) compared to females, and 1.7% (171/10,185), were never breastfed. The prevalence of malaria fever among children 24 months and above is more than the national prevalence of 35.5%. Furthermore, the result shows that malaria status was strongly associated with child's, age, birth order, vaccination, malaria status, nutritional status, fever, diarrhoea, duration of breastfeeding, deworming, iron pills/syrup, and vitamin A intake.

Bivariate Analysis of Proportion and Associations between Predictors and Malaria Status
However, malaria status was not statistically associated with child's sex, perceived birth size, wasting, and the presence of acute respiratory infection two weeks before the survey.
The results in (b) part show the relationship between parental characteristics and malaria status of children 6-59 months of age in Nigeria, only marital status and paternal occupational status were not strongly associated with the malaria status at 5% level of significance. Whereas child's 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, ante-natal care visit, maternal ethnicity, religious status, maternal iron supplement during pregnancy, and paternal education status were all strongly associated with the child's malaria fever status. The proportion of children with malaria fever (38%) was highest among mothers in the youngest age groups (15-24 years) compared to other age groups. More than 50% of children whose mother and father had no education reported RDT positive results. Additionally, 38%, 38%, 36%, and 32% of children whose mother had a preceding birth interval of 12-24 months, 25-35 months, 36-59 months, and 60+ months were RDT positive, respectively. In addition, the prevalence of malaria fever among children aged 6-59 months was higher for those whose mothers were underweight compared to normal and overweight mothers.
On the household-related categories of predictors, (i.e. the (c) part of Table 1), indicates that household socioeconomic status (wealth index), household size, number of bedrooms available, number of under-five in the household, age and sex of household head, number of under-five who slept under bed net the night before the survey, the various household characteristics were statistically associated with RDT positive status among children 6-59 months of age in Nigeria at 5% level of significance, while disposal of youngest child's stool methods and household sharing toilet facilities were not statistically associated with malaria status. The proportion of malaria fever among children 6-59 months in Nigeria varied inversely with the level of household wealth index. The highest was recorded among the poor household (53%) followed by middle (38%) and rich (18%) household.
Accordingly, the household where only some under-five years slept under a bed net the night before the survey witnessed the highest prevalence of malaria fever among children 6-59 months of age in Nigeria when compared to the household where "no net in the house" (30%), "no child" (34%), "all children" (37%), slept under a bed net the night before the survey.
In addition, (d) of Table 1 displays the results of the univariate analysis and the association between the cluster-related factors and the malaria fever status of children 6-59 months of age in Nigeria. All the three variables were strongly associated with malaria fever status of children 6-59 months in Nigeria.
Furthermore, (e) of Table 1 displays the results of the univariate analysis and the association between the area-related factors and the malaria fever status of children 6-59 months of age in Nigeria. All the area variables (human development index, multidimensional poverty index, regions of residence, and place of residence) were strongly associated with malaria fever status of children 6-59 months in Nigeria.

Multilevel Multivariable Models of Predictors of Malaria Fever Status
In the first instance, all the variables that serves as proxies to nutritional status and household wealth were excluded from the multilevel analysis. Furthermore, a multicollinearity test was conducted to check for highly correlated predictors. Out of the variables included two factors: "under-five slept under a bed net the night before the survey" and "household had bed net" were perfectly correlated with variance inflation factors (VIF) of 7.08 and 11.18, respectively such that the mean VIF was 2.23. The variable "household had bed net" was dropped resulting in a mean VIF of 1.79. We used a forward stepwise variable selection procedure by entering all variables that were statistically associated with the malaria status of children 6-59 months of age in Nigeria at a 5% level of significance, and removal was by p > 0.20. Because of this, 25 variables (child's age, duration of breastfeeding, anaemia status, nutritional status, fever status, deworming, maternal age in group, age at first birth, maternal education status, paternal education status, maternal anaemia status, ethnic group, religious status, household wealth, number of under-five in household, household head age group, under-five slept under a bed net, number of bedrooms, low cluster wealth level, cluster distance to health facility is no big problem, low cluster household with bed net, state multidimensional poverty index, state human development index, region of residence, and place of residence), were finally retained for the multilevel model building.

Multilevel Model Results A Measure of Variation (Random Effects)
Model 1 is the null model (no predictors) with the fixed effect showing that the proportions of the total variations due to differences in the communities and the states were respectively, 1.266 and 0.614, while the variance due to individual level is 3.29 (π 2 /3), which is fixed for logit. Therefore, the variations in the prevalence of malaria status due to the three-level factors were assessed through intrastate correlation coefficient of 0.1188 (95% CI: 0.75-0.183) and intracommunity correlation coefficient of 0.3636 (95% CI: 0.318-0.412), indicating that 11.9% and 36.4% of the total variation in the odds of malaria positive were respectively due to state and community levels. The variance partition coefficient (VPC) at the state level corresponds with the ICC at the state level. However, the VPC at community level is 0.249, meaning that 24.9% of the total variance is collectively attributed to both the state and community levels. However, from the chosen model (Model 4) in Table 2, the ICC at the community-level has dropped from 36.4% in the null model to 21.0% (95% CI:17-25%), meaning the correlation between two children/individuals (unit of analysis) within the same community and the same state is 0.21, and the ICC at state-level dropped from 11.88% to 4.8% (95% CI: 3-8%), both had remained significant. The performance of models was established using AIC and likelihood ratio. Improvements in model fit was achieved at Model 4 (full model), with AIC = 9646, and log likelihood = −4763.47. Table 2 shows the results of the adjusted odds ratios (AOR) for each of the variables considered in the analysis after adjusting for the rest variables. Model 2 represents the model fitting with child/individual-level variables only: age of the child, the child's duration of breast feeding, had fever two weeks before the survey, dewormed in the last six months before the survey, anaemic status of the child, maternal age at first birth, secondary education and above of maternal and paternal, mother's anaemia and ethnicgroup, household wealth, number of under-five in the house is more than four children, and the household head is between 35 and 44 years of age were statistically significant predictors of malaria status among children 6-59 months of age in Nigeria. The child's nutritional status, maternal age and religious status, number of under-five who slept under a bed net the night before the survey, and the number of bedrooms in the household were not statistically significant predictors. However, after including all the predictor variables (Model 4), the significant status of the child-specific factors remains.

Measures of Association (Fixed Effects)
The odds of a child having malaria increased as the child's age increased. The odds of children between the age of 48-59 months experiencing malaria fever were 2.68 times the odds of children 6-11 months of age (AOR = 2.68, 95% CI: 2.03-3.54). Children who were still breastfeeding (AOR = 0.61, 95% CI: 0.51-0.76), and dewormed (AOR = 0.75, 95% CI: 0.65-0.87), had 39% and 25% reduced odds of contracting malaria infection. Similarly, children whose mother had their first birth after the age of 20 years had reduced odds of experiencing malaria fever than their counterparts whose mothers had their first birth earlier than 20 years.
In addition, children whose mothers (AOR = 0.67, 95% CI: 0.55-0.82), or fathers (AOR = 0.79, 95% CI: 0.66-0.95) had secondary education and above had 33% and 21% significantly reduced odds, respectively, of being malaria fever positive. The wealthier the household (AOR = 0.61, 95% CI: 0.49-0.76), the less likely the child can be malaria positive. From among the clusters-related variables, children from a community with high proportion of mothers who said distance to the nearest health centre is "no big problem" had lower odds of malaria fever infested. The result from among the area-specific variables shows that children from south-south geopolitical zone (AOR = 0.50, 95% CI: 0.25-0.98), had 50% reduced odds to contract malaria fever when compared with children from the north central geopolitical zone. On the contrary, children from rural area (AOR = 2.12, 95% CI: 1.75-2.57), were more likely to suffer from malaria infection compared to children from urban area.
Finally, the result shows that the state-level multidimensional poverty index and state human development index were not statistically significant predictors of malaria status among children 6-59 months of age in Nigeria.

Discussion
This study assessed the prevalence of malaria across the states and the federal capital territory and examined its predictors among children 6-59 months of age in Nigeria. Some researchers use either rapid diagnostic test or microscopy thick blood smears result to classify the presence or absence of Pf. However, Azikiwe et al. (2012) found that RDTs and microscopy laboratory investigation of malaria Pf yield similar results with RDTs being more precise [55]. In addition, in a recent scoping review [1], RDT was found to be more frequently used in studies compared to microscopic smears. Additionally, in 2018 NDHS, RDTs account for more samples than microscopic blood smear laboratory test. RDT result was used in this study. The study compliments the findings in previous studies [15,22], to show impacts of some potential effects of contextual variables at both cluster and state levels. About one-third of children were found to be malaria positive. This result shows that substantial reduction in malaria prevalence among children 6-59 months of age in Nigeria have been achieved since 2010 when the first national representative survey was conducted, (52%), and 2015 recorded 45% prevalence [9,10]. Furthermore, the study established that malaria status among children 6-59 months of age in Nigeria is determined by both child-, parental/household-, community-, and state-related variables. The results revealed that children in the higher age group are more prone to malaria infection than their younger counterparts, this agrees with other studies [5,15,16,21]. The reason for this may be that the younger the child, the more attention he/she gets from caregivers, i.e., priority is often given to the younger ones in the use of bed net when available. However, children less than 12 months may not have started walking, and as the age increases and the likelihood to walk it becomes more difficult to protect them. The study also discovered that currently breastfeeding children had reduced odds of malaria infections of which this agrees with another study [13]. The possible reasons for these are that a breastfeeding child gets more antibodies from the mother to help fight infection, and a breastfeeding child often sleeps alongside the mother who normally ensures that the child is unexposed to mosquito bites and this could provide additional protection [5]. Furthermore, there are evidence that comorbidity in anaemia and non-malaria fever increased the chance of malaria infection [5] but had decreased influence with anaemia [13].
As the maternal age at first birth increases, the odd of child's RDT outcome decreases. The study also found that maternal and paternal education statuses are significantly associated with child's malaria status. The odds reduced with an increase in educational level [5,13,21]. However, it is not clear how these two factors interact to affect child's health outcomes, (this can be interesting for future analysis). This finding further demonstrates the important role parental education plays in a child's health [5]. In addition, the result shows that the anaemia status of the mother is significantly associated with the child's malaria prevalence.
Household wealth plays a significant role in child's health, children from rich household are less likely to be infected with the malaria parasite. Similar conclusions were reached in recent studies [5,13,15,56]. Furthermore, the study revealed that the higher the number of rooms available in the household for sleep, the less likely the child will be malaria RDT positive, this agrees with the result found in a similar study [19]. This may relate to the fact that when fewer people sleep in a room, the tendency for reduced cross-infection. Children from rural areas were found to be more vulnerable to malaria infection compared to their urban counterparts, this finding is consistent with similar studies [13][14][15][16][17]. The study also shows that the higher the proportion of respondents whose distance getting to health centre "is no big problem", the less likely is the children would contract malaria infections. In other words, when immediate action for prompt medical attention becomes a big problem in a community, the children are highly exposed to be bitten by an infected mosquito and other childhood diseases.

Strengths and Limitations of the Study
A recent scoping review conducted has revealed that there are very few studies that analysed the influence of contextual factors on the risk of contracting malaria infection among children under-five years in SSA countries [1]. This study has contributed to bridging this knowledge gap. Moreover, the few studies available adopted a two-level multilevel analysis procedure, but in this study, we found that a three-level multilevel analysis was more appropriate, as such, this study is one among the few to carry out such analysis at a country level. The study involved large data sets, which might make it possible to draw inference over the country.
However, there were some limitations: (i) The data sets were cross-sectional and could not ascertain the remote causes of RDT malaria positive among children 6-59 months of age in Nigeria. Information regarding the causes is better obtained from a longitudinal study which requires periodic follow up of participants [41,48]. (ii) The accuracy of the information provided at the survey time were not ascertained to be correct because of high maternal illiteracy in Nigeria which could have resulted in recall errors while responding to some questions. (iii) some variables were dropped for missingness, and listwise deletion method was used to handle incomplete data in the remaining variables. Other methods such as multiple imputation could have been used for the incomplete observation. (iv) The classification of the determinants of RDT malaria prevalence among children 6-59 months of age were tailored through child-, parental-household, and contextual framework. However, it is argued that some variables may not directly increase or decrease the risk of contracting malaria (indirect), and some may do (direct), and the interactions effects may be interesting. This was not performed in this study and could be a subject of future study.

Policy Implications
Between 2010 when a national representative survey on malaria was first conducted to determine the prevalence of malaria among children 6-59 months of age in Nigeria, and 2018 when the current survey was conducted, the prevalence of malaria had dropped significantly by 15% from 50% in 2010 to 35% in 2018. This drop is commendable but was not enough to meet the expectations of total elimination of malaria-related death in Nigeria by the year 2020 [6,9]. Despite the achievement obtained so far, Nigeria remains the most malaria-endemic nation in the world. In view of this, the current efforts regarding the Government, need a boost to ensure that the scare resources are channelled to areas that will require more and urgent attention. The findings from this study will aid informed decisions that will help reduce the incidence of malaria among children 6-59 months of age in Nigeria.
In line with some findings in this study, we have proffered some suggestions for policy implementation.
(i) The findings show that older children are often at risk compared to younger siblings. These children are allowed by their parents to walk about freely unattended to, perhaps to obtain more feeding opportunities elsewhere. These children have been weaned from breast feeding, nutritious foods are often not provided at home because some households cannot afford it. Most of these children scavenge about to find food. It is recommended that the school feeding program of the Government should also be extended to what may be called "community feeding program" where children of pre-school age can access food and boost their immunity. (ii) Parenting is a very serious matter in our society. The age at which most women first become a mother is often very low, thereby making them "baby-mothers". Some of these mothers are not even mature enough to take care of themselves, particularly becoming mothers at that tender age, and not even able to care of their babies. Policies and the political will should be in place to discourage early girl-child marriage that often result into early childbirth. Girl-child education often delay the age at which most females become married and give birth. Therefore, female education should be made free and compulsory for up to secondary education. Any parent who withdraws their girl child from school for early marriage or whose daughter gets pregnant while still in school should be liable for prosecution. (iii) Most families sleep in over-crowded apartments. The governmental housing policies in some states see some buildings constructed but are never allotted to anyone until they are eventually vandalized. The Government's policies should encourage early allocation of these buildings to those that need them to ease off over-crowding households and communities. This will reduce cross-infections in communicable diseases such as malaria (iv) Due to the problem of accessibility to most rural areas (especially during the raining season when malaria infection rates are usually very high), these communities are often neglected in the distribution of scare palliatives from the Government and agencies (such as medicals and bed nets), to addressing the issue of malaria. The Government can also invest in the use technologies available for logistics such as drone to transport these items to the hard-to-reach communities. (v) When community distance to health facility is not a "big problem" to the people, they become prompt in getting the much-needed timely medical attention for their children. Therefore, having identified areas of high malaria prevalence, the governments for these areas should among other suggestions ensure that increased proportion of the people do not travel long distances before they can access prompt medical attention. Therefore, more functional health centres are available in such localities. (vi) Lastly, the results also indicates that children from a high category among low cluster wealth areas had significantly higher proportion of malaria positive status compared to the proportion among the low category ((d) in Table 1). Therefore, the Government can investigate and implement ways to bring many homes out of poverty lines. Many of the governmental programs in the past toward poverty alleviation have ended up without achieving their aims. These measures do not find their way into the hands of those that need them. Sometimes they are distributed as political campaign "juices" for political party supporters. There should be sincerity in the part of both the program implementers and the beneficiaries in driving the program to success.

Conclusions
Despite the huge resources committed to eliminating malaria induced morbidities and mortalities, Nigeria has remained the most burdened malaria-endemic nation in the world. This study has identified some important individual and contextual predictors of malaria among children 6-59 months of age in Nigeria. These predictors are areas that need to be considered for policy designs and implementations toward control and total elimination of malaria-related morbidity and mortality among children in Nigeria.
Author Contributions: The conceptualization of this study was performed by P.E.O. and K.K.; the formal drafting of manuscript was carried out by P.E.O.; while S.J.W., R.J. and K.K. supervised, revised, and edited the manuscript. All authors have read and agreed to the published version of the manuscript.