Multilevel modelling of malaria risk in children under ve years old in Nigeria

Background: Malaria remains a major public health threat in sub-Saharan Africa. Despite efforts to eradicate the disease in Nigeria, it is still a major cause of morbidity and mortality; especially among children under ve (U5) years. This study assessed individual, household, and community risk factors for malaria in Nigerian children U5 years of age. Methods: Data from the Nigerian Malaria Health Indicator Survey 2015 were pooled for analyses. This comprised of national survey of 329 clusters. Children aged 6-59 months who were tested for malaria using microscopy were retained. Multilevel logit model accounting for sampling design was used to assess individual, household and community factors associated with malaria parasitaemia. Results: A total of 5742 children were assessed for malaria parasitaemia with overall prevalence of 27% (95% CI 26-28%). Plasmodium falciparum constituted 98% of the Plasmodium species. There was no signicant difference in parasitaemia between older children, and those ≤ 12 months. In adjusted analyses, rural living, Northwest region, household size of >7, dependence on river and rain water as primary water source were associated with higher odds of parasitaemia; while higher wealth index, all U5s who slept under bed net and dependence on packaged water were associated with lower odds of parasitemia.


Introduction
Malaria is a major public health problem in developing economies, especially in sub-Sahara Africa (SSA) where it is responsible for substantial morbidity and mortality (1) as well as other socioeconomic losses (2). The disease is caused by Plasmodium parasite, and transmitted through the bite of infected female Anopheles mosquitoes (3). It has been argued that the observed decline in All Cause Child Mortality is attributable to a commensurate trend in child malaria mortality in otherwise high risk areas (4)(5)(6).
However, malaria still accounted for over 400,000 mortality in 2018, plus the fact that about 219 million cases of malaria occurred in 2017 compared to 217 million cases in 2016 (7). This report emphasises that no signi cant progress was made in reducing global malaria between the year 2015-2017.
The African region of the World Health Organisation (WHO) contributes about 92% of malaria cases with South-East Asia (5%) and the Eastern Mediterranean region (2%) constituting the remaining fraction.
Nigeria accounts for a quarter of the global malaria burden, far greater than Democratic Republic of the Congo, Mozambique, India and Uganda (7). Despite signi cant investments in the control of malaria, it remains a threat in Nigeria, especially among under-ve (U5) children (8). In 2018, U5 children accounted for 67% and 61% of all malaria deaths worldwide and SSA, respectively. It is estimated that every two minutes, an U5 child dies of malaria, and SSA dominates in the death toll (9).
Understanding country-speci c malaria risk, especially in the most vulnerable populations like U5 children will in no small measure form the prerequisite to establishing a long-term control intervention. In countries like Nigeria with less promising indices, the focus is still on case reduction, which requires investment in both human and material resources for surveillance and health system strengthening for any attempt at elimination (10). This study therefore assessed the relationship of demographic, socioeconomic and some environmental determinants of malaria transmission/prevention among U5 children at individual, household and community levels of disease intervention.

Study area:
The study was conducted in Nigeria, which is a country in West Africa. Nigeria shares border with Niger in The sampling frame was based on the 2006 National Population and Housing Census of the Federal Republic of Nigeria, conducted by the NPC. Administratively, Nigeria is divided into states. Each state is subdivided into local government areas (LGAs), and each LGA is further divided into communities. In addition to these administrative units, during the 2006 census, each community was subdivided into convenient areas called census enumeration areas (EAs). The primary sampling unit, referred to as a cluster for the 2015 National Malaria Indicator Survey (NMIS), was de ned based on EAs from the 2006 EA census frame. A two-stage sampling strategy was adopted. Nine clusters (EAs) were selected from each state, including the Federal Capital Territory (FCT) during the rst stage in a manner that is representative of each state. The result was a total of 333 clusters throughout the country, 138 and 195 in urban and rural areas, respectively.
In the second stage of the selection process, 25 households were selected in each cluster by equal probability systematic sampling. All women aged 15-49 who were either permanent residents of the households in the 2015 NMIS sample or visitors present in the households on the night before the survey were eligible to be interviewed. In addition, all children age 6-59 months were eligible to be tested for malaria and anaemia. This sample size was selected to guarantee that key survey indicators could be produced for each of the country's six geopolitical zones, with approximately 1,338 women in each zone expected to complete interviews. In order to produce some of the survey indicators at the state level for each of the 36 states and the FCT, interviews were expected to be completed with approximately 217 women per state. A more detailed description of survey design and microscopy procedures can be found in the 2015 NMIS report (11).

Malaria risk factors
Individual level explanatory variables: Gender, age and number of children who slept under insecticide treated bed net (ITN) a night before survey was considered potential individual level risk factors of malaria risk. We grouped age in months into ve categories: ≤ 12 months, 13-23 months, 24-35 months, 36-47 months and 48-59 months.

Household level explanatory variables:
This includes household size, comprising number of household members, mother's (caregiver's) level of education, number of U5 children in the household, wealth index level, household source of drinking water, the location of source of water and whether dwelling has been sprayed with insecticide(s) against mosquitoes. Household source of drinking water such as piped into dwelling, yard/plot, neighbour and public tap/standpipe were grouped together as piped; protected/unprotected well, protected/unprotected spring were grouped as well/spring; river/dam/lake/ponds/stream/canal/irrigation channel were grouped as river/stream; sachet water, bottled water, cart with small tank were grouped as packaged water; while borehole or tube well and rainwater stood alone forming six recategorised groups.
Community level explanatory variables: The study considered place of residence (rural/urban) and region of residence as potential community level explanatory variables. The region of residence is based on the six geopolitical zones that exist in Nigeria-Northcentral, Northeast, Southwest, Southsouth and Southeast regions.

Ethical clearance
Secondary data approved by ICF Macro International were used for this study. The ICF Macro International obtained ethical approval from ethical review boards of the Nigerian Ministry of Health before the survey commenced. Informed consent was received from the study participants before data collection. Also, participants were assured of the con dentiality and anonymity of the information provided during data collection.

Statistical analysis
Individual level data were retrieved from the nationally representative 2015 NMIS. The survey design estimation command (svy) in Stata 16.1 (StataCorp, College Station, TX) was used to conduct descriptive analysis, accounting for the sampling weight. Categorical variables are represented as counts and percentages. To assess the association between malaria in U5 children and individual, household, and community risk factors, we speci ed a three-level random intercept logistic regression model to account for contextual within-household and within-cluster correlations. Given the complex survey design of the data, sampling weight was included in the xed effects portion of the model. The model is represented as below: Where Y ijk is the outcome (presence of malaria) for i th under 5-year-old child at household j in cluster k. β 0 is the mean log-odds of malaria across household and cluster. X ijk is a level 1 covariate for the i th child in household j and cluster k, while β 1 represents the slope associated with X ijk which represents the relationship between the level 1 covariates and the log odds of malaria. η k is the random effect for cluster, while ξ ij is the household random effect. The random effects are independent and identical to each other and are assumed to be normally distributed with mean of zero and variances σ η and σ ξ .
A univariate multi-level logistic regression was conducted, using each individual, household, and community risk factors as predictors and malaria risk as outcome. Individual predictors with any level less than p = 0.10 were considered for inclusion in the multivariable multi-level logistic regression. The multivariable analysis was conducted in a sequential process. Firstly, we tted a null model (Model 0)-an unconditional model-to decompose the total variance of malaria prevalence between the cluster and level 1 covariates. Model 1 consist of individual level factors, Model 2 included household level factors, while model 3 included community level factors to constitute a fully adjusted model.
The in uence of contextual household and cluster effects on the prevalence of malaria were explored, and expressed as the area variance, intracluster correlation and the median odds ratio (12,13). Estimates of the univariate association between outcomes and individual, household or community level covariates are reported as odds ratio (OR) and the 95% con dence interval, while the multivariable models (Models 1, 2 and 3) are reported as adjusted odds ratio (AOR) and 95% CI. The goodness of t of the sequential multivariable models were assessed using the Akaike Information Criterion (AIC). Statistical signi cance was set at an alpha level 0.05.

Results
Characteristics of study population and prevalence of malaria Table 1. There was a total number of 5742 (weighted=5814) children in 3685 households, and 326 clusters with valid record for malaria parasitaemia assessed through microscopy. The mean age of U5 children assessed for malaria parasitaemia was 36 months (range: 0 -59 months), of which 50% were females. About 43% of the children were reported to have slept under mosquito bet net last night before the interview, while 42.6% of the caregivers have completed no education. Majority of the household source of drinking water were from borehole/ tube well (34.4%) and well/spring water (30.4%). Overall, only 1.3% had dwelling place sprayed against mosquito in the last 12 months, and 65.9% of the children lived in rural areas. The overall malaria prevalence was 27.3% (95% CI: 24.8-30.0%). The prevalence of Plasmodium species -falciparum, malariae and ovale were 98.2%, 5% and 1.1%, respectively. Figure 1 shows the prevalence of U5 malaria based on source of drinking water across regions. Well/ spring and borehole/tube well are the predominant sources of drinking water in the northern regions amongst U5 children with malaria. About 23.8% prevalence of U5 malaria in Northwest was as a result of use of well/spring sources of drinking water. Association of malaria parasitaemia with individual, household and community level factors   At the household-level, children in households with members >7 had 47% greater odd of malaria, compared to those with <5 members after accounting for other individual, household and community level factors. Children whose caregivers have primary or secondary education had lower risk of malaria compared to those born to a non-educated caregiver. The higher the wealth index of the household, the lower the odds of U5 malaria parasitaemia. Also, households whose source of drinking water is from well/spring water, rainwater, and river had 2.45, 2.79, and 3.03-fold increase in the odds of malaria parasitaemia compared to those who drink from pipe borne water, respectively. Conversely, children from household that use packaged water as source of drinking had signi cantly lower odds of malaria. At the community level, children who live in the rural areas had 2.68-fold increase in the odds of malaria compared to those who live in urban areas. However, children from the Northwest region had signi cantly 2.35-fold increase in odds of malaria compared to those from Northcentral region. Figure 2 shows the predicted probabilities of malaria risk for children living in households with varying wealth index and in rural/urban areas across the six geopolitical regions. Although, children in Northwest region had the highest probability of malaria parasitaemia infection for each household wealth index and place of residence, the trend seem to be the same for children at all regions.

Measures of household and cluster-level variation in malaria risk
Estimates of the cluster and household random effects (Table 4)

Discussion
This representative study of Nigerian children U5 years old shows a high prevalence of malaria in this vulnerable group. Our result shows 27% prevalence of U5 malaria diagnosed by microscopy was found in Nigeria. This reveals a fraction of the huge national burden of malaria in Nigeria. Over 93% of malaria cases and related deaths were recorded in Africa, with Nigeria ranking up there as the most endemic country with 25% and 24% of global malaria cases and deaths, respectively (7).
While no association was recorded for age and sex, we found that the highest prevalence of malaria among U5 children in our sample population was recorded in children aged 4-5 years [48-59 months] (33.4%). The absence of de nite strati cation within this age group suggests that all children under 5 years of age are at high risk. In fact, a seasonal malaria chemoprevention program has been piloted, and shown to signi cantly reduce malaria in under 5 children participating in the mass drug administration program in different areas of diverse malaria transmission intensity (14)(15)(16)(17). Some economic bene t has further been attributed to seasonal malaria chemoprevention of all under 5 children (16,18), with recommendation for this approach to be integrated into the health system in endemic areas.
The study did not observe any sex difference in U5 malaria prevalence in Nigeria. Our nding is in agreement with other studies in U5 children in Ethiopia (19) and Burkina Faso (20). In addition to demographics, another individual level factor assessed was the availability and use of ITNs by U5 children, the night before the survey. Although the prevalence of malaria infection decreased when at least some children slept under the ITNs, there was no de nite association between use of ITNs the night before and U5 malaria in the study population. While other studies had reported up to 10-fold decrease in under 5 malaria due to use of ITNs (19), our data suggest that the use of ITNs alone may not be su cient for effective malaria control and elimination as opined by others (21).
At the household level, we found strong association between several assessed demographic and socioeconomic determinants. High number of household members but not the number of U5 children signi cantly increased malaria risk, especially when this number exceeds seven members. As previous studies have found that Anopheles mosquitoes are attracted to higher concentration of carbon dioxide as well as human odour (22), it is therefore plausible that larger households in our data are more likely to have an increased risk of mosquito bites.
Furthermore, our data suggest that children with educated mothers had less odds of malaria. As mother's education is often used as a proxy for household wealth (23), previous studies have found a reduced risk of malaria in wealthier households (24). There is the consistent trend in association between the household wealth index, an indicator related to other household determinants that impact on malaria incidence, with the poor suffering the brunt of malaria cases and perhaps mortality. Maternal education profoundly affects the household perception of malaria preventive measures, including acceptability and practice of malaria control interventions (25). Putative causal relationship has been reported for the impact of mother's level of education to U5 malaria (26). Wealth index is closely related to educational level and has been shown in this study to reduce malaria risks, consistently showing a negative association between household wealth and risk of under 5 malaria. Because wealth impacts other indices like education, housing, household nutrition, area of residence and health seeking behaviour, it is arguably a major determinant of U5 malaria (25,27). Although malaria can be rightly described as a disease of poverty (28), no association was found in some studies between chronic malnutrition and U5 malaria (29), suggesting a more complex relationship.
Also, household domestic water sources that promote availability of stagnant water will directly impact the breeding capacity of malaria vectors, indirectly worsening malaria transmission among U5 children.
Potable water resources and indoor residual spraying are key malaria intervention efforts targeted at the disease vectors that has recorded signi cant success in malaria transmission in several diverse settings (25,30,31). Indeed, termination of indoor residual spraying intervention was reported in a study to result in rebound of prevalence to epidemic proportion (32).
At the community level, residence in urban or rural areas can impact malaria infection in U5 children mainly related to associated factors like population density, proximity to favourable vector breeding sites, agricultural projects, and closeness to health facilities. Like other studies, we also found that residence in rural areas and suburbs can worsen malaria transmission among U5 children (33,34). The low prevalence in urban areas may be related to population density and its impact on the e cacy of malaria control interventions. Stebbins et al (35) reported variation in the e cacy of ITNs in urban and rural areas, concluding that ITNs use in urban areas offer bene ts beyond individual protection. This further implicates population density factors, which were not observed in rural settings with more sparse Averaging of the cluster-level variations by State shows that the contextual variations in U5 malaria risk are lower in most States in the southern region and higher in States in the Northern region. The sharp decrease in the cluster level median odds ratio after controlling for individual, household and communitylevel factors, again signify the important role these factors play in malaria transmission/intervention.
While a spatially explicit analysis was not performed in this work, there is likely a spatial pattern in our prevalence estimates, given that Nigeria has distinct climate, socio-economic and environmental characteristics between its region. We therefore believe that these are signi cant contributing factors to regional variations in U5 malaria transmission. It will be interesting to thoroughly investigate these regional variations in U5 malaria transmission within a spatial framework.

Strengths And Limitations
The large and nationally representative study population is an important strength of the study.
Parasitaemia was diagnosed through microscopy, which is the gold standard. Also, the study was able to identify individual, household and community level predictors of U5 malaria in Nigeria, thus provides information that will favour design of suitable intervention. Given that the data resulted from crosssectional designs, causal relationship between explanatory variables and U5 parasitaemia cannot be assumed.

Conclusions
Malaria remains a substantial public health burden in Nigeria, especially in U5 children despite sustained investment in malaria control and prevention. A quarter of the overall national burden are in the U5s.
Across the six geopolitical zones, the highest burden was in children living in poorest rural households.
Based on the variations in U5 malaria risk across regions in Nigeria, region-speci c interventions will be more bene cial in reducing the burden of malaria than a one size ts all approach.  Predicted probabilities of malaria across region and wealth index (from tted model 3) Figure 3 Cluster (A) and State (B) level variations in prevalence of malaria from the null model