 Research article
 Open Access
 Open Peer Review
 Published:
Exploring the association of anthropometric indicators for underfive children in Ethiopia
BMC Public Healthvolume 19, Article number: 764 (2019)
Abstract
Background
Child undernutrition is a global health concern. Many studies have focused on the association of childhood undernutrition indicators with their predictors. A few studies have looked at relationship between the undernutrition indicators. This study aimed at investigating the possible association structures of childhood undernutrition indicators.
Methods
A loglinear model of cell counts of a three way table of stunting, wasting, and underweight was fitted based on the 2016 Ethiopia demographic health survey data. The variables of interest were generated based on the 2006 WHO Child Growth Standards as: stunted, wasted and underweight if zscores of heightforage, weightforheight and weightfor age are below2, respectively; otherwise not stunted, wasted and underweight.
Results
This study showed that 36.34, 12.09 and 24.87% were stunted, wasted and underweight out of sampled children respectively and the prevalence of total undernutrition in children was about 45.96%.The fitted loglinear model showed that underweight was associated with both stunting (Pvalue< 0.001), and wasting (Pvalue< 0.001). There was no association between stunting and wasting (Pvalue = 0.999). Furthermore, the model showed that there is no a three way interaction among stunting, wasting, and underweight (Pvalue = 1.000).
Conclusion
The authors conclude that there is lack of three way association of stunting, wasting, and underweight. This confirms that the three anthropometric indicators of children have multidimensional nature. Thus, the concerned body should consider the three undernutrition indicators simultaneously to estimate the actual burden of childhood undernourishment as they are not redundant of each other.
Background
Child undernutrition is a global health concern. Infants and young children are the most vulnerable to undernutrition due to their high dietary requirements for growth and development. Undernutrition affects all aspects of children’s life; its effects are not limited to physical health but extend to mental, social, and spiritual wellbeing [1].
The indicators of childhood under nutrition are stunting, wasting and underweight. Stunting refers to a child who is too short for his or her age (low heightforage); wasting refers to a child who is too thin for his or her height (low weightforheight); and underweight refers to a child with low weightforage [2]. Heightforage, weightforheight and weightforage Zscores are calculated using the 2006 WHO child growth standards recommended for international settings [3]. Children are considered stunted when they have heightforage zscore below two compared with the WHO Child Growth standards median of same age and sex. Wasting is defined by a weight for height zscore (WHZ) below − 2 and suggests acute undernutrition or rapid weight loss. Underweight is defined by a weight for age zscore (WAZ) below − 2 [2].
The effects of childhood undernutrition are classified as short term and longterm. Mortality and morbidity are parts of short term effect as it is indicated to magnify the disease status [4]. The risk of death from childhood illness like pneumonia, diarrhea and malaria is higher for a child who is severely undernourished. The mortality of children under 5 years of age because of undernourishment is around 45%. These usually occur in developing countries [5]. The longterm effects of childhood undernourishment are linked to poor educational outcomes, unhealthy and economically unproductive adult population [6].
According to 2016 global data, 22.90 and 7.70% of children under 5 years of age are stunted and wasted, respectively [2]. Out of one third of all undernourished children under 5 years of age are found in SubSaharan African countries based to the 2015 MDG report. In this SubSaharan Africa, the prevalence of stunting, wasting and underweight for children under 5 years of age are 39%, 10 and 25%, respectively [7, 8]. These reports suggest that although undernutrition has been decreased globally, it remains a series problem in SubSaharan Africa.
Ethiopia has made solid progress against undernutrition in the past decade, but the country remains an extremely undernourished country. According to 2006 WHO child growth norms, 55.70% of Ethiopian preschool children were stunted in 2000. This declined to 43.4% in 2011 and to 40% in 2014 [9]. Child wasting is also relatively higher than 10%and child underweight prevalence is around 25% in 2014, down from 41% in 2000 [9]. Since children are the economic assets to the world and their future development outcomes can be influenced by their nutrition and health status, the mechanism and consequences of undernutrition and health problems need to be understood better. This is true in a country like Ethiopia where undernutrition and health problems are common.
Several studies have assessed determinants of childhood undernutrition [10,11,12] and have not focused on the association of stunting, wasting and underweight. A few studies tried to explore pair wise association and the three dimension association of stunting, wasting and underweight [13, 14]. Exploring the three dimension association is useful to check whether stunting, wasting and underweight have valid multidimensional nature or not. This study aimed at investigating the association of childhood undernutrition indicator variables by applying loglinear model for the threeway table to assess pair wise and three dimension association of undernutrition indicators. The interaction term would be considered to represent possible association [15].
Methods
Data source
This study was conducted based on the database that has been compiled as part of the 2016 Ethiopia Demographic and Health Survey (EDHS). It is the fourth Demographic and Health Survey conducted in Ethiopia [16]. The 2016 EDHS sample was stratified and selected in two stages. Each region was stratified into urban and rural areas. In the first stage, a total of 645 enumeration areas (EAs) were selected with probability proportional to EA size, of which 202 in urban areas and 443 in rural areas. An EA is a geographic area covering on average 181 households. In the second stage of selection, a fixed number of 28 households per enumeration area were selected from the newly created list of household listing using systematic sampling. In all the selected households, height and weight measurements were collected from children age 0–59 months [16]. The data set used in the analysis was Children’s Data set which was based on woman and household questionnaires. The analysis was based on children with complete anthropometric and valid age data. The overall number of child records with complete anthropometric and valid age data was 8768.
Variables
The three anthropometric indicators are measured through zscores for heightforage (stunting), weightfor height (wasting) and weightforage (underweight) and are defined as: \( {Z}_i=\frac{AI_i\mu }{\sigma } \), where AI_{i}_{I} is the individual (child) anthropometric indicator, μ and σ refer respectively to median and standard deviation of the reference population [17].
The variables of interest generated were: stunted (0 = No if HAZ ≥ 2 and 1 = Yes if HAZ <  2), wasted (0 = No if WHZ ≥ − 2 and 1 = Yes if WHZ < − 2), and underweight (0 = No if HAZ ≥ 2 and 1 = Yes if HAZ <  2). The methodology for computing the indicators was based on the 2006 WHO Child Growth Standards [17].
Statistics analysis
Loglinear model for the threeway table
A loglinear model is a statistical model for the natural logarithm of the expected frequency. The loglinear models are fitted to see the significant associations based on the cross table. Loglinear models are interpreted as generalized linear models which treat the cells counts as independent observations from the Poisson distribution with corresponding means equal to the expected cell counts. They are useful when all the three factors can be treated as response [15]. Consequently, in this study the three factors (underweight, stunting and wasting) were treated as response and the focus is on their structure of association by fitting 2^{3} + 1 = 9 loglinear possible models [14]. Therefore, the nine loglinear models of the mean cell count from the three factors are given bellow.
Note that in the last model of the following table the interactions between wasting and stunting; wasting and underweight; stunting and underweight; wasting, stunting and underweight are included.
Loglinear models are appropriate to test hypotheses about complex interactions, but the parameter estimates are less easily interpreted. Parameter estimates are log odds ratios for associations.
Model  Expression  Description 

Log of mean of cell counts = log(μ_{ijk})  λ + λ_{i}^{W} + λ_{j}^{U} + λ_{k}^{S}  Complete independence among the three factors (pairwise independent). No interaction term is included. 
\( \lambda +{\lambda}_i^W+{\lambda}_j^U+{\lambda}_k^S+{\lambda}_{ik}^{WS} \)  Underweight is partially independent of stunting and wasting. This model contains the interaction between wasting and stunting  
\( \lambda +{\lambda}_i^W+{\lambda}_j^U+{\lambda}_k^S+{\lambda}_{jk}^{US} \)  Wasting is partially independent of stunting and underweight. This model contains the interaction between underweight and stunting.  
\( \lambda +{\lambda}_i^W+{\lambda}_j^U+{\lambda}_k^S+{\lambda}_{ij}^{UW} \)  Stunting is partially independent of wasting and underweight. This model contains the interaction between underweight and wasting.  
\( \lambda +{\lambda}_i^W+{\lambda}_j^U+{\lambda}_k^S+{\lambda}_{jk}^{US}+{\lambda}_{ik}^{WS} \)  Wasting & underweight are conditionally independent of stunting. This model contains the interaction terms between underweight and stunting; stunting and wasting.  
\( \lambda +{\lambda}_i^W+{\lambda}_j^U+{\lambda}_k^S+{\lambda}_{ij}^{WU}+{\lambda}_{ik}^{WS} \)  Stunting and underweight are conditionally independent of wasting. This model contains the interaction terms between wasting and underweight; wasting and stunting.  
\( \lambda +{\lambda}_i^W+{\lambda}_j^U+{\lambda}_k^S+{\lambda}_{ij}^{WU}+{\lambda}_{jk}^{US} \)  Stunting & wasting are conditionally independent of underweight. This model contains the interaction terms between wasting and underweight; underweight and stunting.  
\( \lambda +{\lambda}_i^W+{\lambda}_j^U+{\lambda}_k^S+{\lambda}_{ij}^{WU}+{\lambda}_{ik}^{WS}+{\lambda}_{jk}^{US} \)  Homogenous associations (every factor interacts with each other factor, but there is no interaction between all three factors)  
\( \lambda +{\lambda}_i^W+{\lambda}_j^U+{\lambda}_k^S+{\lambda}_{ij}^{WU}+{\lambda}_{ik}^{WS}+{\lambda}_{jk}^{US}+{\lambda}_{ij k}^{WU S} \)  All possible relationships between the factors are included. 
Key: W, U and S represent wasting, underweight and stunting respectively; λ_{i}^{W}, λ_{j}^{U}, and λ_{k}^{S} are wasting, underweight and stunting effects respectively.
For tables with at least three variables, unsaturated models can include association terms. Then, loglinear models are more commonly used to describe associations through twofactor terms than to describe odds through singlefactor terms [18]. Interaction terms correspond to associations among variables.
Goodnessoffit tests
The model with terms corresponding to all possible main effects and interactions is called the saturated model. Goodness of fit tests whether the reduced model is significantly worse than the saturated (full) model. The tests using Chisquared statistic (χ^{2}) and likelihoodratio chisquared statistic (G^{2}) are goodnessoffit tests of a loglinear model. This Goodness of Fit Tests can be used to select best fitting model. The larger the values of G^{2} and χ^{2}, the more evidence exists against independence. It is also suggested by other criteria, such as minimizing AIC = − 2(Maximized log likelihood – number of parameters in the model) [18]. Data was analyzed using R version 3.5.2 and decision was done based on 0.05 levels of significance.
Results
Table 1 shows cross tabulation of stunting, wasting and underweight. There were 36.34% stunted, 12.09% wasted and 24.87% were underweight out of sampled children.
Figure 1 revealed that the multiple correspondence analysis of the three indicators of child undernutrition. There is no association between wasting and stunting as they are in the first and fourth quadrants. Whereas, underweight is located on the boundary line and indicated that it is associated with both stunting and wasting.
For measuring total prevalence of undernutrition in children, composite index of anthropometric Failure (CIAF) was followed [19]. According to CIAF classification children can be divided into: No failure; Stunted only; Underweight only; Wasted only; Stunted and underweight; Wasted and underweight; and Wasted, stunted, and underweight. By excluding children with no anthropometric failure, CIAF is a better index for estimating prevalence of total childhood undernutrition in a population [19]. Therefore, the categories of CIAF are given in Table 2 below.
Table 2 shows Categories of the Composite Indicator of Anthropometric Failure along with its frequency computed from Table 1 raw data.
According to composite index of Failure about 45.96% {=(16.86 + 1.43 + 4.22 + 15.58 + 3.98 + 3.89) %} of children were diagnosed with undernutrition while 54.04 were not diagnosed with undernutrition out of the sampled children. That is 45.96% children were stunted, underweight or wasted. In other word the prevalence of total undernutrition in children is about 45.96%. The CIAF indicates total undernutrition and does not provide any information on the prevalence of stunting, underweight and wasting relative to total undernutrition.
The likelihood ratios chisquare (G^{2}), Pearson chisquare (χ^{2}) and AIC are reported in Table 3. These are model fit test for the fitted loglinear models of cross tabulation of stunting, wasting and underweight (Table 1). The observed and fitted cell counts are compared in the fit statistics. The null hypothesis is stated as the observed and the fitted cell counts are the same (the data is well fitted by the model). The alternative hypothesis is that the observed and the fitted cell counts are different. A Pvalue less than 0.05 significant level for the fit statistics means stronger evidence against the model fits the data well and a Pvalue greater than 0.05 significant level for the fit statistics shows strong evidence that the model fits the data well.
Models 1 to 8 do not fit the data well because the Pvalues for the fit statistics are much less than 0.05 level of significance, while model 9 fits the data well as the pvalues for the fit statistics are much greater than 0.05 level of significance (Pvalue =1.000) and has the smallest AIC value. Thus, results of the saturated model (model 9) are presented and interpreted below.
Table 4 shows results of saturated model (model 9). The null hypothesis of the individual test of coefficient of the interaction term is stated as there is no interaction between indicators of undernutrition. Thus, there is a high significant interaction between underweight and stunting since the Pvalue of the estimates of interaction term Stunting*underweight is much less than 0.05 (Pvalue< 0.001). There is also a high significant interaction between underweight and wasting (Pvalue< 0.001), but no interaction between stunting and wasting as the Pvalue is much greater than 0.05 (Pvalue = 0.999). Furthermore, the saturated loglinear model shows lack of three factor association among stunting, wasting and underweight (Pvalue = 1.000).
Discussion
In this study, a loglinear model was fitted for the threeway table in exploring association among stunting, wasting and underweight. In the loglinear model, the association of the undernutrition variables is represented by the interaction terms. Loglinear models have the advantage of assessing the threeway interaction in addition to the very common analysis of pair wise association [15]. The data is well fitted by the saturated loglinear model as the pvalues for the fit statistics are greater than 0.05 level of significance (Pvalue =1.000) as compared to the rest of the unsaturated models.
The fitted model indicates underweight is significantly associated with stunting (Pvalue< 0.001). It also shows underweight and wasting are significantly interrelated (Pvalue< 0.001). These two findings are in line with previous studies conducted by Ngwira, Gupta and Borkotoky [13, 14]. In addition, this result is agreed on the fact that underweight is a composite measure of stunting and wasting [20, 21]. The fitted model indicates stunting and wasting are not interrelated (Pvalue = 0.999). This findings is consistent with studies done by Ngwira, Gupta and Borkotoky [13, 14]. Furthermore, the fitted model reveals there is no a three way interaction among stunting, wasting and underweight (Pvalue = 1.000). This finding is also coincides with the findings of Ngwira, Gupta and Borkotoky [13, 14]. The lack of three way interaction indicates that stunting, wasting and underweight have statistically valid multidimensional nature.
Conclusion
The study concludes underweight is significantly associated with both stunting and wasting. The observed association of underweight with both stunting and wasting does not imply one undernutrition indicator causes the other since cross sectional data was used for the analysis. The study also concludes that stunting and wasted are not associated. On top of these, the study reveals that there is no a three way association among stunting, wasting and underweight. From the lack of three way interaction one can conclude that the anthropometric indicators of children have multidimensional nature. Thus, the concerned body should consider the three undernutrition indicators simultaneously to estimate the actual burden of childhood undernourishment as they are not redundant of each other. Finally, authors recommended that further studies can explore whether a causal relationship exists between undernutrition indicator variables or not by using data from prospective or retrospective cohort studies.
Availability of data and materials
The data set supporting the conclusions of this article is held by the authors and the Central Statistical Agency, CSA, Ethiopia, and the derecognized data may be made available if a unique request is crafted from CSA website (http://www.csa.gov.et).
Abbreviations
 AIC:

Akaike’s information criterion
 CIAF:

Composite Indicator of Anthropometric Failure
 CSA:

Central statistical agency
 EAs:

Enumeration areas
 EDHS:

Demographic and Health Survey
 HAZ:

Heightforage zscore
 MDGs:

Millennium development goals
 UN:

United Nations
 WAZ:

Weight for age zscore
 WHO:

World health organization
 WHZ:

Weight for height zscore
References
 1.
WHO: The World Health Organization child growth standards. 2016.
 2.
United Nations Children’s Fund WHO, World Bank Group. Levels and Trends in Child Malnutrition—UNICEF/WHO/World Bank Group Joint Child Malnutrition Estimates. In: UNICEF/WHO/World Bank Group New York, NY; 2017.
 3.
Cintron C, Seff I, Baird S. Dynamics of wasting and underweight in Ethiopian children. Ethiop J Econ. 2016;25(2):113–70.
 4.
Pathak PK, Singh A. Trends in malnutrition among children in India: growing inequalities across different economic groups. Soc Sci Med. 2011;73(4):576–85.
 5.
WHO: World health organization. 2018.
 6.
Black RE, Victora CG, Walker SP, Bhutta ZA, Christian P, De Onis M, Ezzati M, GranthamMcGregor S, Katz J, Martorell R. Maternal and child undernutrition and overweight in lowincome and middleincome countries. Lancet. 2013;382(9890):427–51.
 7.
Organization WH: World health statistics 2016: monitoring health for the SDGs sustainable development goals: World Health Organization; 2016.
 8.
Akombi B, Agho K, Hall J, Wali N, Renzaho A, Merom D. Stunting, wasting and underweight in subSaharan Africa: a systematic review. Int J Environ Res Public Health. 2017;14(8):863.
 9.
Headey D. An analysis of trends and determinants of child undernutrition in Ethiopia, 2000–2011. International Food Policy Research Institute (IFPRI); 2014.
 10.
Asfaw M, Wondaferash M, Taha M, Dube L. Prevalence of undernutrition and associated factors among children aged between six to fifty nine months in Bule Hora district, South Ethiopia. BMC Public Health. 2015;15(1):41.
 11.
Khatri RB, Mishra SR, Khanal V, Choulagai B. Factors associated with underweight among children of formerKamaiyas in Nepal. Front Public Health. 2015;3:11.
 12.
Wolde M, Berhan Y, Chala A. Determinants of underweight, stunting and wasting among schoolchildren. BMC Public Health. 2015;15(1):8.
 13.
Gupta AK, Borkotoky K. Exploring the multidimensional nature of anthropometric indicators for underfive children in India. Indian J Public Health. 2016;60(1):68.
 14.
Ngwira A, Munthali EC, Vwalika KD. Analysis on the association among stunting, wasting and underweight in Malawi: an application of a loglinear model for the threeway table. J Public Health Afr. 2017;8(1):620.
 15.
Agresti A. An introduction to categorical data analysis. USA: Wiley; 2018.
 16.
EDHS: Ethiopian demographic and health survey report. 2016.
 17.
Organization WH. WHO multicentre growth reference study group: WHO child growth standards: length/heightforage, weightforage, weightforlength, weightforheight and body mass indexforage: methods and development, vol. 2007. Geneva: WHO; 2006.
 18.
Agresti A. Categorical data analysis, vol. 482: Wiley; 2003.
 19.
Namburi NS, Seepana M. Assessment of undernutrition using composite index of anthropometric failure among children less than 5 years in an urban slum, Visakhapatnam. Int J Commun Med Public Health. 2018;5(11):4773–7.
 20.
Sellen D. Physical status: the use and interpretation of anthropometry. Report of a WHO expert committee. WHO technical report series no. 854. Pp. 452.(WHO, Geneva, 1995.) Swiss Fr 71.00. J Biosoc Sci. 1998;30(1):135–44.
 21.
Rutstein SO, Rojas G. Guide to DHS statistics. Calverton: ORC Macro; 2006.
Acknowledgments
The authors are indebted to the Ethiopian Statistical Authority (ESA) for giving us permission to use the data for our purpose.
Funding
The authors have no support or funding to report.
Author information
Affiliations
Contributions
GW wrote the proposal, analyzed the data and manuscript writing. DL accredited the proposal with revisions, analysis the data and manuscript writing. Both GL and DL read and approved the very last manuscript.
Corresponding author
Correspondence to Demeke Lakew Workie.
Ethics declarations
Ethics approval and consent to participate
The ethical clearance for the survey was approved by Ethical Review Board of Central Statistical Agency (CSA), Ethiopia and all participants who agreed to take part in the survey signed a consent form. Hence, we authors asked the CSA permission to use data via on line form and the data manager of CSA give permission to use for this article.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
About this article
Received
Accepted
Published
DOI
Keywords
 Loglinear model
 Undernutrition
 Stunting
 Underweight
 Wasting
 Interaction