Determinants of childhood anaemia in Nigeria: a public health perspective using quantile regression analysis Authors: Talani Mhelembe, Shaun Ramroop, Faustin Habyarimana Published: 2025 ABSTRACT BackgroundIn Nigeria, anaemia is still a major public health issue for children under five, and its high prevalence is caused by a variety of risk factors. Conventional regression techniques frequently ignore the ways in which these risk factors affect various population segments. The distributional effects of environmental and sociodemographic factors on childhood haemoglobin levels are investigated in this study using quantile regression.MethodData from the 2021 Nigeria Integrated Malaria Survey (NIMS) and the Demographic and Health Survey (DHS) household dataset were combined to produce a pooled sample. Using SPSS, a quantile regression analysis was performed to evaluate the relationship between haemoglobin levels in children aged 6–59 months and a number of risk factors, such as altitude, child age, region, housing materials, household wealth index, mosquito net usage, sex, recent fever, and malaria status. Quantile regression has an advantage over binary logistic regression in that it makes it possible to ascertain how the independent factors affect the full distribution of the dependent variable.ResultsThe results of the investigation showed that risk factors had different effects on haemoglobin levels depending on the distribution. Children with substandard housing materials, those from the poorest households, and those from certain locations (north-east, north-west, and south-east) had considerably lower haemoglobin levels, especially at the lower quantiles. Higher haemoglobin levels were linked to protective factors such as growing older, being a woman, sleeping beneath mosquito nets, not having had a recent illness, and having a negative malaria status; these effects were frequently more noticeable at quantiles. These results demonstrate how the most vulnerable children bear a disproportionate amount of the anaemia burden.ConclusionQuantile regression provided a nuanced understanding of the determinants of childhood anaemia, revealing that socio-economic disadvantage, regional disparities, and malaria infection exert the greatest impact among children with the lowest haemoglobin levels. Universal and targeted interventions addressing poverty, housing quality, malaria prevention, and equitable access to health resources are essential to reduce the burden of anaemia and promote health equity.RecommendationsInterventions for the most vulnerable children should be given top priority by policymakers and medical professionals, especially for those living in high-burden areas and the quintiles with the lowest incomes. Improving socioeconomic conditions, expanding access to high-quality healthcare and nutrition, and improving malaria control measures (such universal mosquito net coverage) are all important strategies. It is advised that more studies employ quantile regression to track the reduction of childhood anaemia and inform focused public health initiatives. Introduction Anaemia is considered a most life-threatening public health concern [6, 9, 19, 24], especially in developing countries, where children are more vulnerable [6, 29]. In Nigeria and around the world, the incidence of anaemia in children, especially those under five, remains a serious problem [24, 29]. It is a severe burden on underprivileged children with long-term health and socio-economic consequences [24]. Anaemia can be defined in many ways [5, 9, 14, 21]. This current study defines anaemia as the low blood haemoglobin concentration [5, 29]. Childhood haemoglobin (Hb) concentration level is commonly classified based on the cut-off point [9] as follows \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Hb<7.0g/dL$$\end{document} as severe anemia, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$10.0g/dL\le Hb\le 10.9g/dL$$\end{document} as mild, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$7.0g/dL\le Hb\le 9.9g/dL$$\end{document} as moderate anemia, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Hb\ge 11.0g/dL$$\end{document} as non-anemic [9, 29]. According to several research in the literature, raising the mother's household income and educational attainment will lower her likelihood of developing anaemia [6, 8, 9, 13, 17]. The likelihood of childhood anaemia is also increased by the child's age, family size, and exposure to malaria, according to additional studies [5, 12]. A study by Ogunsakin et al. [24] suggested that the policymakers in Nigeria should apply a pragmatic approach to help address the menace of childhood anaemia [24]. According to the Nigeria Indicator Malaria Survey report [23], the percentage of low haemoglobin levels among children aged between 6–59 months old decreased from 13% in 2010 to 8% in 2021 [23]. This suggests that research like the current study is necessary to uncover some of the risk factors connected to anaemia in order to help policymakers and government health organisations further build measures to reduce the prevalence of anaemia in Nigerian children. Several studies in literature used the binary as a response variable where a child is either anaemic or not anaemic [5, 6, 14], to name a few. All these studies considered the mean average of the dependent variable or Gaussian distribution [9]. However, the upper and lower quantiles of the outcome variables may depend on the covariates differently from the centre [9]. The inability to analyse the quantiles of interest is a drawback of the aforementioned models. The goal of the current study is to examine the influence of predictor variables on the various quantiles of interest in childhood haemoglobin levels using quantile regression, which offers flexibility. This study's primary goal is to use quantile regression models to identify the risk factors linked to the prevalence of childhood anaemia. At the same time, multiple imputations using chained equations (MICE) are used to address missing observations. Methods and materials Data source Study design This population-based cross-sectional study used secondary data from demographic health surveys (DHS). Study population The Nigeria Malaria Indicator Survey (NMIS) data from 2021 were used in this investigation. The National Malaria Elimination Programme (NMEP) of the Federal Ministry of Health (FMoH), in partnership with the National Population Commission (NPC) and the National Bureau of Statistics (NBS), conducted the 2021 Nigeria Malaria Indicator Survey (NMIS), which is used in this study [23]. For the 2021 NMIS, a two-phase sampling approach was used. A probability proportional to the enumeration area (EA) size was used to choose 568 EAs in the first stage. The number of households that call the EA home determines its size. To make the sample representative of each state, it was carefully chosen. As a consequence, there were 568 clusters nationwide, 195 in cities and 373 in rural areas [23]. The 2021 NMIS employed three questionnaires: the Biomarker Questionnaire, the Woman's Questionnaire, and the Household Questionnaire [23]. The questionnaires were modified to represent the population and health issues pertinent to Nigeria, based on the model questionnaires developed by the DHS Program. Following their completion in English, the surveys were translated into Hausa, Yoruba, and Igbo [23]. The National Health Research Ethics Committee of Nigeria (NHREC) and the ICF Institutional Review Board reviewed and approved the survey protocol [23]. The data from the DHS was from the 2021 NIMS, and the children sample was selected from the household recode [23]. The total population of children was 9648,according to the NIMS report, the selected children for anaemia were between 6–59 months old [23]. A thorough case study was carried out, and some predictors had missing observations. After that, quantile regression was used to model the data, and imputation was performed to address the missing observations. The imputation techniques that can be used with missing data are described in the section below. Missing data It is usual to encounter missing data in epidemiological studies [3, 10, 15]. Most survey data contain missing observations [20], and different methods can be employed to handle missing observations [10, 20]. The method employed can affect the analysis outcome [10]. This could compromise the conclusion drawn from the results [27] and lead to wrongful inference. In most cases, missing data is managed by dropping the cases that are not fully measured [10]. However, this study compared the complete case and imputed data based on statistics of the standard error resulting from the model. Multiple imputation, one of the techniques employed for missing data, is discussed below: Table 1 clearly shows 3.20% missing values when the response variable is crossed with each covariate. Therefore, this influenced the use of imputation.Table 1Distribution of valid and missing valuesCategoryCasesTotalP-valueValidMissingFrequencyPercentNPercentNType of residence932596.70%323.1923.30%9648 <.001Mother's education level932896.70%320.1923.30%9648 <.001Type of toilet facility934596.90%303.1923.10%9648 <.001Household has electricity933196.70%317.1923.30%9648 <.001Main floor material933796.80%311.1923.20%9648 <.001Main wall material933096.70%318.1923.30%9648 <.001Main roof material934396.80%305.1923.20%9648 <.001Wealth index933696.80%312.1923.20%9648 <.001Children slept under a mosquito net931396.50%335.1923.50%9648 <.001Malaria prevention knowledge933896.80%310.1923.20%9648 <.001Sex of child932796.70%321.1923.30%9648 <.001Had fever in the past two weeks933196.70%317.1923.30%9648 <.001Malaria RDT Results932896.70%320.1923.30%9648 <.001Altitude933196.70%317.1923.30%9648 <.001Age in months933896.80%310.1923.20%9648 <.001Region933696.80%312.1923.20%9648 <.001 Table 1Distribution of valid and missing valuesCategoryCasesTotalP-valueValidMissingFrequencyPercentNPercentNType of residence932596.70%323.1923.30%9648 <.001Mother's education level932896.70%320.1923.30%9648 <.001Type of toilet facility934596.90%303.1923.10%9648 <.001Household has electricity933196.70%317.1923.30%9648 <.001Main floor material933796.80%311.1923.20%9648 <.001Main wall material933096.70%318.1923.30%9648 <.001Main roof material934396.80%305.1923.20%9648 <.001Wealth index933696.80%312.1923.20%9648 <.001Children slept under a mosquito net931396.50%335.1923.50%9648 <.001Malaria prevention knowledge933896.80%310.1923.20%9648 <.001Sex of child932796.70%321.1923.30%9648 <.001Had fever in the past two weeks933196.70%317.1923.30%9648 <.001Malaria RDT Results932896.70%320.1923.30%9648 <.001Altitude933196.70%317.1923.30%9648 <.001Age in months933896.80%310.1923.20%9648 <.001Region933696.80%312.1923.20%9648 <.001 Distribution of valid and missing values Multiple imputation via chained equations The assumption was that the data was missing at random (MAR) [20]. Unbiased results with accurate estimates for the standard errors (SE) are obtained, when multiple imputation is applied to a dataset, that is MAR,[25]. In this study, multiple imputation by chained equations was used. Chained equations are also known as fully conditional specification (FCS) [10], and it is the algorithm adopted by the Statistical Analysis System (SAS) Enterprise. This approach is more flexible to imputation since it is designed to handle different types of variables (binary, continuous, categorical, and ordinal) and does not assume the multivariate normality of the data [10, 18]. Assuming the multivariate normality can often lead to truncation and inflated standard errors or wrongful inference with the loss of information [10]. As applied in SAS, imputation by FCS is also an iterative process that starts by imputing every missing value with random draws from the distribution of the non-missing values [10]. SAS Enterprise Guide version 8.3 was used to perform the imputation process. Variables Dependent variable The outcome variable of interest is the haemoglobin concentration in blood measured in grams per decilitre (g/dL). The NIMS 2021 provided adjusted and unadjusted childhood haemoglobin concentration levels to altitude; this study selected the adjusted haemoglobin. Independent variables The independent variables used in this study were also used in various studies modelling childhood anaemia [6, 9, 14, 24, 26] among others. This serves as the theoretical foundation for the current investigation. Demographic, geographic, and socioeconomic factors were used to categorise these independent variables: the results of the malaria rapid diagnostic test (RDT), the mother's educational attainment, the household's electricity status, the wealth index, the number of children who slept under a mosquito net the night before the survey, the knowledge of malaria prevention methods, the child's sex, the status of child fever in the two weeks prior to the survey, the altitude level, the child's age in months. Statistical analysis Box-plot Based on a study by Hossain and Majumder [11], a boxplot is very important if the analysis is required to deal with a large dataset. The study by Talukder et al. [28] explored the boxplot for the response variable used in the study. The boxplot uses a five-number summary to give information about the dataset being used. The boxplot summarizes the data measured on an interval scale that can be used to identify if the distribution is skewed and if there are outliers in the data being used [28]. The outliers can be visualized by the graphical plot that is produced from the analysis. This study used the boxplot to detect normality from the dataset. Based on the findings of the boxplot, quantile regression was used to analyse the association between the independent and dependent variables. The analysis was carried out using IBM SPSS Statistics Version 24. Quantile regression Koenker and Bassett [16] first introduced the classic quantile regression model 1978 to extend the notion of ordinary quantiles in a location model to a more general class of linear models in which the conditional quantiles have a linear form [16]. Consider a random valued variable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$X$$\end{document} that is characterized by the following distribution function [16, 28]:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G\left(y\right)=\mathrm{Pr}(Y\le y)$$\end{document} Then, for any \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau \epsilon (\mathrm{0,1})$$\end{document}, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\tau }^{th}$$\end{document} quantile of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y$$\end{document} is defined as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R\left(\tau \right)=inf\left\{y:G(y)\ge \tau \right\}$$\end{document} The median is the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R(\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$2$}\right.)$$\end{document}, the first quantile \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R(\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$4$}\right.)$$\end{document}, and the third quantile \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R(\raisebox{1ex}{$3$}\!\left/ \!\raisebox{-1ex}{$4$}\right.)$$\end{document} (Buhai, 2004). The quantile function provides a complete characterization of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$Y$$\end{document}, just like the distribution function \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$G$$\end{document} [2]. The quantiles can be written as solutions to the following optimization problem [2]: for any \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tau \epsilon (\mathrm{0,1})$$\end{document}, define the piecewise linear function3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rho }_{\tau }\left(v\right)=v(\tau -I\left(v<0\right))$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v$$\end{document} represents the residual for each observation, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I(.)$$\end{document} is the indicator function. This study used the piecewise linear function to determine the estimates for the model. Let \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left({x}_{i},{y}_{i}\right), i=\mathrm{1,2},\dots ,n,$$\end{document} be a random sample from the population, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{i}$$\end{document} is the response variable, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{i}$$\end{document} is a set of selected independent covariates. The following function describes the quantile regression model, which has been fully described by Buchinsky [1]:4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${y}_{i}={x}_{i}{\theta }_{\tau }+{\varepsilon }_{\tau }$$\end{document}where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varepsilon }_{\tau }={y}_{i}-{x}_{i}{\theta }_{\tau }$$\end{document}, which is the difference between the observed value and the value predicted by the quantile regression model. IBM SPSS version 29.0.0.0 (241) was used to run the quantile regression model. Results Exploratory data analysis The current study used the imputed data over the complete case data. The missing values were not deleted when related to the response variable. Comparison of the complete case and imputed data Results from the three different analyses show that, in general, the results are similar. The results from the current study are similar to the study by Hendry et al. [10]. A comparison of the results of the complete case analysis (CC) with the other analysis (imputation) shows that the standard errors (SE) of the estimated coefficients for the CC are larger in all the predictor variables. There is little difference in the magnitude when the estimated coefficients are compared. Based on the analysis, the SE found for the data where imputation has been employed is lower compared to those with the complete case; therefore, it shows higher precision in the estimates found when using the imputed data. In the current study, the imputation was done in SAS using the three iterations of the FCS method. The final data was used in SPSS to model the quantile regression model. Table 2 demonstrates that approximately 38.10% of the youngsters tested positive for malaria according to the fast diagnostic test. There are four percent of mothers who have no education, fourteen percent who have elementary education, thirty-two percent who have secondary education, and ten percent who have further education. Children from the lowest, poorest, middle-class, richer, and richest families made approximately 20.80%, 19.40%, 20.50%, 20.80%, and 19.10% of the total. All children slept under a mosquito net, some slept under one, and children from households without a net contributed roughly 16.50%, 34%, 12.90%, and 36.50%, respectively, to the total number of children who did not sleep under one. It was found that 63.20%, 36.60%, and 0.30% of the chosen children did not have a fever, had a fever within the two weeks before to the survey, and that 0.30% of their mothers were unsure if their child had a fever.Table 2Distribution of children within the selected covariatesVariableCategoryFrequencyPercentageP-ValueMalaria Rapid Diagnostic Test ResultsNegative17,78461.90% <.001Positive10,96238.10%Mother's education levelNo Education11,77541.00% <.001Primary420314.60%Secondary981934.20%Higher294910.30%Household has electricityNo16,73458.20% <.001Yes11,82041.10%Not a Resident1920.70%Wealth IndexPoorest579920.20% <.001Poorer558919.40%Middle588320.50%Richer597620.80%Richest549919.10%Children slept under a mosquito net last nightNo473416.50% <.001All Children978534.00%Some Children372212.90%No net in the house10,50536.50%Knowledge of Malaria preventionNo389413.50% <.001Yes23,04080.20%Don't know18126.30%Sex of ChildMale14,93752.00% <.001Female13,80948.00%The child had a fever in the last two weeksNo18,15963.20% <.001Yes10,50936.60%Don't know780.30%Altitude LevelBetween 500 and 1000 m331211.50% <.001Greater than 1000 m4861.70%Less than 500 m24,94886.80%Age in monthsBetween 13 to 24651922.70% <.001Between 25 to 36614421.40%Between 36 and 48799227.80%Between 49 and 59744925.90%Between 6 to 126422.20%RegionNorth Central535218.60% <.001North East555019.30%North West814528.30%South381313.30%South East306310.70%South West28239.80%Main floor material usedFurnished Material16,11656.10% <.001Natural Material12,17442.40%Other2400.80%Rudimentary2160.80%Main wall material usedFurnished Material17,23860.00% <.001Natural Material23678.20%Other6002.10%Rudimentary854129.70%Main roof material usedFurnished Material23,99483.50% <.001Natural Material27039.40%Other4381.50%Rudimentary16115.60%Toilet FacilityFlush to piped sewer system19116.60% <.001Flush to septic tank516618.00%Flush to pit latrine20857.30%Flush to somewhere else990.30%Flush, do not know where150.10%Ventilated Improved Pit Latrine10233.60%Pit Latrine with slab612621.30%Pit Latrine without slab/open pit490517.10%No facility/bush/field673523.40%Compositing toilet1050.40%Bucket toilet720.30%Hanging toilet/latrine2610.90%Other510.20%Not a resident1920.70% Table 2Distribution of children within the selected covariatesVariableCategoryFrequencyPercentageP-ValueMalaria Rapid Diagnostic Test ResultsNegative17,78461.90% <.001Positive10,96238.10%Mother's education levelNo Education11,77541.00% <.001Primary420314.60%Secondary981934.20%Higher294910.30%Household has electricityNo16,73458.20% <.001Yes11,82041.10%Not a Resident1920.70%Wealth IndexPoorest579920.20% <.001Poorer558919.40%Middle588320.50%Richer597620.80%Richest549919.10%Children slept under a mosquito net last nightNo473416.50% <.001All Children978534.00%Some Children372212.90%No net in the house10,50536.50%Knowledge of Malaria preventionNo389413.50% <.001Yes23,04080.20%Don't know18126.30%Sex of ChildMale14,93752.00% <.001Female13,80948.00%The child had a fever in the last two weeksNo18,15963.20% <.001Yes10,50936.60%Don't know780.30%Altitude LevelBetween 500 and 1000 m331211.50% <.001Greater than 1000 m4861.70%Less than 500 m24,94886.80%Age in monthsBetween 13 to 24651922.70% <.001Between 25 to 36614421.40%Between 36 and 48799227.80%Between 49 and 59744925.90%Between 6 to 126422.20%RegionNorth Central535218.60% <.001North East555019.30%North West814528.30%South381313.30%South East306310.70%South West28239.80%Main floor material usedFurnished Material16,11656.10% <.001Natural Material12,17442.40%Other2400.80%Rudimentary2160.80%Main wall material usedFurnished Material17,23860.00% <.001Natural Material23678.20%Other6002.10%Rudimentary854129.70%Main roof material usedFurnished Material23,99483.50% <.001Natural Material27039.40%Other4381.50%Rudimentary16115.60%Toilet FacilityFlush to piped sewer system19116.60% <.001Flush to septic tank516618.00%Flush to pit latrine20857.30%Flush to somewhere else990.30%Flush, do not know where150.10%Ventilated Improved Pit Latrine10233.60%Pit Latrine with slab612621.30%Pit Latrine without slab/open pit490517.10%No facility/bush/field673523.40%Compositing toilet1050.40%Bucket toilet720.30%Hanging toilet/latrine2610.90%Other510.20%Not a resident1920.70% Distribution of children within the selected covariates The mean of the children’s haemoglobin level was 10.2, with a standard deviation (SD) of 1.59. The minimum value is 2.0, while the maximum is 16.1. The skewness of the response variable indicates that the distribution of haemoglobin levels is normally distributed Table 3. Outliers were detected using the boxplot, which then gave an in-depth understanding of the connection between factors outside the mean of the data Fig. 1. The data is normally distributed (Fig. 2). It is further noted that the graph shows that the data is normally distributed with a few outlier, indicating that the quantile regression analysis would be appropriate for further analysis [28]. The quantile regression for different quantiles was employed in the dataset 0.25, 0.50, 0.75, and 0.95.Table 3Summary statistics for the Haemoglobin Level variableNumber of observationsMinimumMaximumMeanStd. ErrorStd. DeviationVarianceHaemoglobin Level92622.0016.110.20.01651.5925.2Fig. 1Detecting outliers using Box PlotFig. 2Children’s Haemoglobin level distribution Table 3Summary statistics for the Haemoglobin Level variableNumber of observationsMinimumMaximumMeanStd. ErrorStd. DeviationVarianceHaemoglobin Level92622.0016.110.20.01651.5925.2 Summary statistics for the Haemoglobin Level variable Detecting outliers using Box Plot Children’s Haemoglobin level distribution The selected model uses imputed data, mainly because the standard errors for the complete case were more significant (value-wise) than those for the imputed case (FCS = 3). Therefore, the results interpreted in this section are for those covariates that were significant at each quantile based on the p-values; for the overall tables please refer to Appendix A, B, C and D. IBM SPSS version 29.0.0.0 (241) was used to run the quantile regression model. The results from the analysis represented in Table 4 show that the baseline values for the haemoglobin level at the different quantiles consistently increases from the lower quantile (q = 0.25) to the upper quantile (q = 0.95) when all the other covariates are kept at a value of zero. At the 25th percentile, for each unit of altitude unit increase, haemoglobin increases by 0.001 units. The same outcome is observed at the 75th, and 95th percentiles. However, there is no measurable change in haemoglobin with altitude at the median (50th percentile). These results indicate that the altitude level across all the percentiles except for the 50th percentile, the haemoglobin is positively impacted by altitude. The age of a child in months also showed a significant effect on haemoglobin level; for each additional month of age, haemoglobin increases by 0.023, 0.021, 0.022 and 0.018 units at the 25th, 50th, 75th, and 95th percentiles, respectively. Age in months consistently contributes to haemoglobin levels across the different quantiles, however, the effect slightly decreases at the higher quantiles.Table 4Estimated coefficients and standard errors for the predictors selected in the different analysisParameterq = 0.25q = 0.5q = 0.75q = 0.95Intercept6.6097.8598.51010.336Altitude Level0.0010.0000.0010.001Age in Months0.0230.0210.0220.018Region (reference = South-West) North-East−0.300−0.239****** North-West−0.484−0.264−0.255−0.233 South-East−0.321−0.306−0.314−0.444Wealth Index (reference = Richest) Poorest−0.292−0.362******How many children slept under a mosquito net (reference = Some children) All Children******0.132***Gender (reference = Male) Female0.1010.131******Did child have fever in the last 2 weeks prior to the survey (reference = Yes) Not sure***0.6730.6581.237 No0.2370.2090.1740.237Child's malaria status (reference = Positive) Negative1.1591.0590.9270.6701*** represent all the values that were not significant at the 95% confidence limit2the confidence intervals of the estimates and the p-values, can be found in the Appendices of this manuscript3all the reported categories in Table 4, were the only ones significant at the 95% confidence limit Table 4Estimated coefficients and standard errors for the predictors selected in the different analysisParameterq = 0.25q = 0.5q = 0.75q = 0.95Intercept6.6097.8598.51010.336Altitude Level0.0010.0000.0010.001Age in Months0.0230.0210.0220.018Region (reference = South-West) North-East−0.300−0.239****** North-West−0.484−0.264−0.255−0.233 South-East−0.321−0.306−0.314−0.444Wealth Index (reference = Richest) Poorest−0.292−0.362******How many children slept under a mosquito net (reference = Some children) All Children******0.132***Gender (reference = Male) Female0.1010.131******Did child have fever in the last 2 weeks prior to the survey (reference = Yes) Not sure***0.6730.6581.237 No0.2370.2090.1740.237Child's malaria status (reference = Positive) Negative1.1591.0590.9270.6701*** represent all the values that were not significant at the 95% confidence limit2the confidence intervals of the estimates and the p-values, can be found in the Appendices of this manuscript3all the reported categories in Table 4, were the only ones significant at the 95% confidence limit Estimated coefficients and standard errors for the predictors selected in the different analysis 1*** represent all the values that were not significant at the 95% confidence limit 2the confidence intervals of the estimates and the p-values, can be found in the Appendices of this manuscript 3all the reported categories in Table 4, were the only ones significant at the 95% confidence limit For the different regions, it has been noted that the haemoglobin levels for children from the north-eastern region of Nigeria at the 25th percentile decreases by 0.300 units as compared to the haemoglobin levels of children from the south-western region of Nigeria. The haemoglobin levels for children from the north-eastern region of Nigeria at the 50th percentile decreases by 0.239 units as compared to the haemoglobin levels of children from the south-western region of Nigeria. The 75th and 95th percentiles were not significant for the current study. The haemoglobin levels for children from the north-western region of Nigeria at the 25th, 50th, 75th, and 95th percentiles decrease by 0.484, 0.264, 0.255, and 0.233 units, respectively, as compared to the haemoglobin levels of children from the south-western region of Nigeria. The values indicate consistent lower haemoglobin levels in north-western region compared to the south-western region, with the gap narrowing at the higher quantiles. The haemoglobin levels for children from the south-eastern region at the 25th, 50th, 75th, and 95th percentiles decrease by 0.321, 0.306, 0.314, and 0.444 units, respectively, as compared to the haemoglobin levels of children from the south-western region. The values indicate consistent lower haemoglobin levels in south-eastern region compared to the south-western region, with the gap widens at the highest quantile (q = 0.95). The north-east, north-west, and south-east regions show negative coefficients, meaning lower haemoglobin levels for children in those regions compared to those from the south-western region. The haemoglobin levels for children from a poorest household decrease by 0.292, and 0.362 units at the 25th, and 50th percentiles, compared to children from the richest household. All children who slept under a mosquito net, their haemoglobin increases by about 0.132 units, at the 75th percentile compared to those from a household where some of the children slept under a mosquito net. The haemoglobin levels for female children increase by 0.101, and 0.0.131 units at the 25th, and 50th percentiles, compared to male children. Children whose mother was not aware whether their child had fever in the last two weeks prior to the survey, showed a positive unit increase of the haemoglobin level, however the increase steadily decreases at the 50th, and 75th percentiles, and sharply increases from the 75th to the 95th. The units increase are as follows: 0.673, 0.658, and 1.237, at the 50th, 75th, and 95th, percentiles, respectively, compared to children who had fever in the last two weeks prior to the survey. Furthermore, children who did not have fever in the last two weeks prior to the survey, showed a positive unit increase of the haemoglobin level, however the increase steadily decreases at the 25th, 50th, and sharply decrease from 50th to the 75th percentile, and sharply increases from the 75th to the 95th. The units increase are as follows: 0.237, 0.209, 0.174, and 0.237, at the 25th, 50th, 75th, and 95th, percentiles, respectively, compared to children who had fever in the last two weeks prior to the survey. Lastly, children who tested negative for malaria showed a positive haemoglobin level increase at the different percentiles. However, there is a steady decrease on how the haemoglobin level is affected at different percentiles. The units increase at the different percentiles are as follows: 1.159, 1.059, 0.927, and 0.670, respectively, compared to children who tested positive for malaria. Discussion The advantage of quantile regression over logistic regression (binary) is that it allows for the independent variables' impact on the dependent variable's complete distribution to be determined [9]. Furthermore, outliers and extreme data are usually less influential in quantile regression due to the inherent robustness of the model [9, 28]. The prevalence of anaemia in children under 5 years in Nigeria influenced the study that was carried out to determine some of the risk factors that might influence the high prevalence of anaemia in children. The DHS household and NIMS 2021 were merged to create a pooled sample. The results were obtained from SAS and SPSS software. A quantile regression model was performed in SPSS to determine the risk factors associated with anaemia levels in children between 6 and 59 months old. The results from the model showed that altitude level, age of the child in months, region, primary wall material, main roof material, main wall material, household wealth index, number of children who slept under a mosquito net the night before the survey, sex of the child, whether the child had fever or not in the past two weeks before the survey, and child’s malaria status. In general, this study's findings are consistent with those of comparable research. According to the study, children from households with different altitudes have a positive (marginal) steadily increasing haemoglobin levels, suggesting that they are less likely to suffer from anaemia. This suggests that altitude levels have a minimal influence in childhood haemoglobin levels. It was observed that as the child's age increases, the child's haemoglobin level also increases; this implies that as children grow, the risk of experiencing anaemia also decreases. The finding is similar to that in other studies [6, 14]. According to the current study's findings, children from the north-eastern, north-western, and south-eastern regions of Nigeria have lower haemoglobin levels than those from the south-western region. This suggests that the geography also affects haemoglobin levels. This further suggests that anaemia is more common in children from north-eastern, north-west, and south-east Nigeria. The region was significant in the current study, showing a consistent result similar to another study conducted by Lemoine and Tounian [19]. The haemoglobin levels of children from families with primary floor material that is natural or other have lower haemoglobin levels at the highest (95th), and lower (25th) percentile, respectively, as compared to children from a household where the primary floor material is rudimentary. The haemoglobin levels of children from families with primary wall material that is finished have higher haemoglobin levels at the highest percentile as compared to children from a household where the primary wall material is rudimentary. Children from households where the principal roof material is composed of natural or finished materials, respectively, have higher levels of haemoglobin than children from homes where the predominant roof material is rudimentary, putting them at lower risk of having anaemia. The results further show that children from the poorest households show a decreasing haemoglobin level at the lowest percentiles, compared to those from the richest households. This implies that children from the poorest households are most likely to have anaemia than those from the richest households. According to the study's findings, male children are more likely than female children to suffer from childhood anaemia. This result is consistent with other research findings in the literature [21, 24]. On the contrary, studies by Habyarimana et al. [9] and Khulu & Ramroop [14] found that anaemia was higher among female children than males. The results show that fever negatively impacts haemoglobin levels,therefore, a child with a fever is more likely to have anaemia. This is consistent with results found in other studies [5, 9]. In contrast, the studies by Gaston et al. [5] and Habyarimana et al. [9] compared the children with fever to those without fever. The positive outcome of fever negatively impacts the haemoglobin level. It could be influenced because fever is mainly associated with other morbidity, such as malaria, cough, and diarrhoea, which also affect childhood anaemia [9]. The results show that a mosquito net positively impacts haemoglobin levels; therefore, a child from a household where all the children slept under a mosquito net the night before the survey is less likely to have anaemia than a child from a family where only some children slept under a mosquito net. The result is consistent with what Gaston et al. [6] found. This might be influenced by mosquito nets being used to prevent mosquito bites, which leads to malaria. Malaria is also one of the comorbidities of anaemia [9]. The results of the study further imply that households that are favourable in terms of socio-economic can easily have proper facilities and can afford a comfortable lifestyle. The result is similar to other studies [6, 9, 24]. The study's findings further showed a significant effect of the household wealth index on childhood haemoglobin levels. Children born to low-income families have been found to have an increased risk of childhood anaemia compared to children born in wealthy families. This might be influenced by the fact that wealthy families can afford to buy nutritious foods, which may help to prevent anaemia [7–9, 21, 22]. Lastly, the results showed that the results from malaria RDT significantly affect childhood haemoglobin levels. A child who tested negative for malaria had a reduced chance of being anaemic. A study that was focused on malaria risk factors showed that an anaemic child had a high chance of being diagnosed with malaria compared to those who were not anaemic [4]. Study Limitations The current study used cross-sectional data, and one of the disadvantages of this type of dataset is the inability to determine causality. Therefore, a longitudinal study is suggested to address this problem. Recommendations The study suggests improving household facilities such as toilets and building materials. Furthermore, they suggest the prevention of fever in children under 5 years, eradicating illiteracy within the mothers/caregivers, and poverty among the different households. The above can be achieved through encouragement of caregivers to attend mandatory training at clinics or nearby schools, where the different factors can be discussed and how to better prevent anaemia. Media announcements, podcasts, and flyers can also be the best way to get the message across within the high-risk areas. This recommendation improves on the previous work conducted, since the Quantile Regression model has the following:The analysis is for all the pre-defined quantiles, unlike the mean model that focuses on the mean.There is no assumption on the variance, it does not assume a constant variance.Provides a better understanding of the data. The analysis is for all the pre-defined quantiles, unlike the mean model that focuses on the mean. The analysis is for all the pre-defined quantiles, unlike the mean model that focuses on the mean. There is no assumption on the variance, it does not assume a constant variance. There is no assumption on the variance, it does not assume a constant variance. Provides a better understanding of the data. Provides a better understanding of the data. Therefore, the above points support the recommendations based on the outcomes of Quantile Regression, to have improved as compared to the outcomes found from the mean model. Conclusion The current study’s main aim was to identify the risk factors that influence anaemia in children from Nigeria who are between 0 and 59 months old. Quantile regression was used to determine the different impact on different percentiles of haemoglobin levels i.e. 25th, 50th, 75th, and 95th percentiles. The findings of the study suggest that the following factors influence the haemoglobin levels in children, which then leads to anaemia: altitude level, age of the child in months, region, primary wall material, main roof material, main wall material, household wealth index, number of children who slept under a mosquito net the night before the survey, sex of the child, whether the child had fever or not in the past two weeks before the survey, and child’s malaria status. This study demonstrates the value of quantile regression in uncovering the nuanced and heterogeneous effects of socio-demographic and environmental factors on childhood haemoglobin levels in Nigeria. Unlike traditional binary logistic regression, quantile regression enabled the identification of risk factors whose impact varies across the entire distribution of haemoglobin, providing a more comprehensive understanding of childhood anaemia. Key findings indicate that children from the poorest households, those living in regions such as the north-east, north-west, and south-east, and those with suboptimal housing materials are disproportionately affected at the lower quantiles of haemoglobin distribution, highlighting their heightened vulnerability to anaemia. Conversely, protective factors such as increasing age, female gender, absence of recent fever, sleeping under mosquito nets, and negative malaria status are associated with higher haemoglobin levels, with their effects often more pronounced at specific quantiles. Clinically, these results underscore the importance of targeted interventions. Efforts to reduce childhood anaemia should prioritize the most disadvantaged groups, particularly those at the lowest haemoglobin quantiles by addressing poverty, improving housing conditions, enhancing malaria prevention, and ensuring equitable access to health-promoting resources. The quantile-specific approach also suggests that universal interventions, such as promoting mosquito net use and malaria testing, can yield substantial benefits across the population, but may be especially impactful for those at greatest risk. Overall, this analysis provides actionable insights for policymakers and health practitioners, advocating for both targeted and population-wide strategies to effectively reduce the burden of childhood anaemia and promote health equity in Nigeria.