Significant Risk Factors Associated With Stunting for Children Under the Age of 5-years in Malawi: the Application of Proportional Odds Model Using DHS Datasets.


 BackgroundChild malnutrition is perhaps the one of the main medical condition influencing general human wellbeing, mainly in non-industrial nations. The improvement of legitimate evaluations of malnutrition is one of the difficulties encountered by policymakers in numerous countries worldwide. In this manner, the current study was embraced with the essential goal of evaluating and determining all potential determinants of childhood malnutrition in Malawi, using the Demographic and Health Survey (DHS) data 2015/16. The study seeks to reveal some of the significant factors that are perpetuating the incidence of malnutrition in children of Malawi. It also designed to offer deeper insights on how the probability of being diagnosed with this medical condition (malnutrition) evolves across the different levels of the found significant factors.Methods The proportional odds (PO) model was the best model to utilize, motivated by the design of the current study's data set. The PO model is an alternative to conceptualize how the ordinal designed data can be sequentially into dichotomous groups without losing the ordinal nature of response variables. The model is an extension of logistic regression models with two outcomes, it is one of the best models to deal with ordinal response variable comprising of more than two categories. The PO model, as well as the logistic regression models are common classes of generalised linear models (GLMs) mostly used to model association between dependent variable and independent variables. ResultsThe observations derived from fitting the PO model on the Malawi DHS data to investigate risk factors associated with malnutrition (stunting) suggested that: the age of the child; birth type (singleton/multiple births), parents' level of education, household's type of resident; mother's age at the time of birth, mother's BMI, incident of diarrhoea in the last two weeks before the survey, are the most significant independent risk factors of malnutrition (stunting). ConclusionsAll the aforementioned risk factors are controllable, and they can be improved through intervention strategies. The policies that undergird the country are required to counteract this condition, as the majority of the risk factors need the coherent actions of several governing authorities.

difficulties encountered by policymakers in numerous countries worldwide. In this manner, the current 23 study was embraced with the essential goal of evaluating and determining all potential determinants of 24 childhood malnutrition in Malawi, using the Demographic and Health Survey (DHS) data 2015/16. The 25 study seeks to reveal some of the significant factors that are perpetuating the incidence of malnutrition in 26 children of Malawi. It also designed to offer deeper insights on how the probability of being diagnosed with 27 this medical condition (malnutrition) evolves across the different levels of the found significant factors. 28

29
The proportional odds (PO) model was the best model to utilize, motivated by the design of the current 30 study's data set. The PO model is an alternative to conceptualize how the ordinal designed data can be 31 sequentially into dichotomous groups without losing the ordinal nature of response variables. The model is 32 an extension of logistic regression models with two outcomes, it is one of the best models to deal with 33 ordinal response variable comprising of more than two categories. The PO model, as well as the logistic 34 regression models are common classes of generalised linear models (GLMs) mostly used to model 35 association between dependent variable and independent variables. 36

37
The observations derived from fitting the PO model on the Malawi DHS data to investigate risk factors 38 associated with malnutrition (stunting) suggested that: the age of the child; birth type (singleton/multiple 39 Statistically, nearly one in three persons globally suffer from under-nutrition, micronutrient deficiency, 50 overweight, and/or obesity (1). Sub-Saharan Africa has probably the highest level of child malnutrition 51 globally. Hence, a critical look at the distribution of malnutrition within its sub-regions is required to 52 determine the factors that escalate the worst nutritional status in these areas. 53 Determinants of child malnutrition remain an interest of several researchers. In 2017, according to the 54 World Health Organization (WHO) Statistics data visualizations dashboard (2018), the regional average of 55 the prevalence of stunting in the regions of Africa was found to be 33.6% above the global average, which 56 is 22.2%. Furthermore, in 2017, internationally, there were 51 million who were wasted, 151 million kids 57 under five years old who were stunted, and 38 million kids were overweight (2). These numbers support 58 the fact that the prevalence of stunting remains the most problematic form of malnutrition globally. 59 60 Several studies, internationally, have endeavored to disclose elements that influence childhood malnutrition 61 so that relevant measures can be put in place to improve the circumstance straightforwardly (3,4). The 62 WHO has also implemented a United Nations Decade of Action plan on Nutrition. Also, the WHO members 63 have embraced worldwide targets for improving maternal, baby, and little youngster's nourishment and are 64 focused on supervising the progress. These targets are crucial in recognizing need zones for actions and 65 speeding up worldwide change as specified in the global nourishment targets 2025. For example, expanding 66 select breastfeeding in the initial six months of the infants resulted in a 40% to 50% decrease in the quantity 67 of stunted under-five kids (5). These are a few territories, as per WHO, which should be improved to 68 accomplish the target. Food from the Ocean for Food Security and Nutrition purposes for living for higher need to be given to 75 fisheries and hydroponics in an endeavor to improve worldwide food security (7). Among other activities, 76 part states focused on improving practical food frameworks by creating rational public approaches from 77 creation to utilization across pertinent areas to give all year admittance to food that meets individuals' 78 sustenance needs, advance protected and differentiated sound eating regimens (8). 79 The administrations of Italy, the Russian Federation, Japan, the United Kingdom, and Northern Ireland co-80 supported an uncommon occasion on reinforcing public obligation to end malnutrition in the entirety of its 81 structures. This occasion was upheld by the Food and Agriculture Organization of the United Nations and 82 the WHO and occurred on 20 September 2016 at the UN Headquarters (9). Many other reports presented 83 by the WHO show how fighting malnutrition is being conducted worldwide, and researchers are also 84 working to identify all the key factors (1). 85 The African region has the largest population accounting for 33.6% of the stunted children globally and a 86 minimum of 20.0% prevalence of wasting in all the regions, according to the WHO's 2018 report (10). As 87 a result, several studies on children's maturational status under five years have been carried out across the 88 regions of Africa, using different methods. These studies' objective was to examine the risk components of 89 children's malnutrition within the African continent. Habyarimana (2017) utilized the proportional odds 90 model to discover malnutrition's main factors, less than five children in Rwanda (11). Another recent study 91 conducted by Taluder (2017) endeavored to reveal the related elements of lack of healthy sustenance among 92 under-five Bangladeshi youngsters by utilizing BDHS information to utilize the PO model (12). Several 93 researchers have already revealed that factors such as mother's and father's level of education, mother's 94 BMI, wealth index, residence's location, antenatal consideration administration during pregnancy, and the 95 interval between successive birth are regular reasons for a poor nourishing status among children under five 96 across the African mainland (11)(12)(13)(14). 97 A descriptive and econometric analysis done by Kabubo-Mariara (2008) in Kenya, amplified by policy 98 limitations, was employed to research the effect of the child, parental, family unit, and local area attributes 99 kids' tallness and the likelihood of stunting (15). Critical discoveries from this examination were that male 100 children experience more malnutrition than young ladies, and kids from multiple births are bound to be 101 malnourished than singletons. Various investigations have found that even the age of the kid and youth 102 ailment are critical factors influencing kid dietary status in Africa (13,16) 103 The prevalence of stunting in under-five children is very high in Malawi (37.1% on average), which is 104 above the regional average in Africa (33.6%), according to the joint child malnutrition estimates by the 105 United Nations Children's Fund, the WHO, and the World Bank Group (10). According to the UNICEF 106 global site (2018), in Malawi, 4% of youngsters, particularly those under five years old, experience the ill 107 effects of acute malnutrition, and the more significant part of Malawian kids experience the ill effects of 108 chronic malnutrition, bringing about stunting (17). These numbers are empirical evidence that Malawi is 109 one of the nations with the highest malnutrition occurrence within the Southern and Eastern parts of Africa. survival. The initiative was more coordinated towards sound sustenance for pregnant ladies and infants 120 during their initial two years of life. Some strategies to prevent stunting in the country include, but not 121 limited to, supporting the government to develop the Nutrition Act and educating the people living with 122 pregnant women on how to support and encourage the behavioural change for maternal Nutrition and young 123 child and infants nourishing practices (18). All these efforts seem to be promising in ensuring that the 124 average level of stunting is decreasing in the country. 125 In addition to these strategies aimed at preventing the disease, an alternative and effective approach to this 126 endeavor would be to clearly understand the exact factors that escalate the incidence of malnutrition. This 127 study's primary objective is to reveal those factors associated with childhood malnutrition in Malawi using 128 the PO model. An almost similar study by Chirwa (2008) was conducted in Malawi, the researcher used 129 multivariate analysis to investigate factors that determine child malnutrition. The study was conducted 130 using the three anthropometric measures of malnutrition, WAZ for underweight, HAZ for stunting, and 131 WHZ for wasting. Consequently, it was revealed from the Chirwa (2008) study that stunting is the most 132 significant contributor to malnutrition problems amongst the three measures of malnutrition. The current 133 research is fashioned to narrow deep into understanding causes of the most common symptom of 134 malnutrition in Malawi, which stunting, and through the use of the most recent available Demographic 135 Health Survey (DHS) data. We have also taken a deep dive into the specifically addressing the impact of 136 each factors of stunting across different levels. Possible factors are going to be selected, this factors has 137 been used in the different studies conducted across the African continent (4,11,13,16). The study uses 138 association statistics and further uses the regression analysis to analyse information that cannot be revealed 139 by association. 140

Data and Definition of variables 142
Malawi is a landlocked state in south-eastern Africa. Endowed with extensive lakes and spectacular 143 highlands, it is within a narrow, bending piece along with East African Rift Valley land. Lake Nyasa, also 144 recognized as Lake Malawi within Malawi, represents significantly more than one-fifth of Malawi's overall 145 zone. The nation's exports are comprised of production from both minor landholdings and tobacco and tea 146 bequests. The country has achieved an impressive level of worldwide capital in the sort of advancement 147 help that has contributed essentially toward the abuse of its natural assets and has made it feasible for 148 Malawi to, on occasion, produce a food surplus. Notwithstanding, the nation's populace has suffered from 149 persevering malnutrition, especially children under the age of five years, elevated paces of new-born child 150 mortality, and devastating povertyan oddity regularly connected to an agricultural framework that has 151 supported significant estate proprietors. were weighed and measured. The study utilized 5092 children, whose complete and plausible 155 anthropometric data were available. 156

Dependent variable 157
This study's dependent variable is the anthropometric measure of height-for-age z-score as an indicator of 158 chronic malnutrition known as stunting. Stunting of children is considered as the best general indicator to 159 gauge the level of wellbeing of children. Moreover, stunting is imperially the highest pervasive form of 160 childhood malnutrition, there is an estimation of around 161 million kids globally, falling under 2 Standard 161 Deviation (SD) from the World Health Association Child Growth Standards middle, height-for-age (19). 162 As a result, in the current study, we considered only the height-for-age anthropometric index to represent a 163 child's nutritional status throughout our investigations. The Z-scores extracted from the height-for-age 164 anthropometric index for each child were calculated to understand the association, if any, between 165 nutritional status and all the selected independent variables. Children's nutrition status (represented by the 166 calculated Z-score) was considered to be a categorical variable with three different ordered levels: severely 167 malnourished (for children with Z-score < -3.0), moderate malnourished (for children with -3.0 ≤ Z-score 168 < 2.0), and nourished (for children with Z-score ≥ -2.0). Hence the response variable (nutritional status) 169 was considered an ordinal response variable grouped from a continuous variable. 170

171
The explanatory variables considered are child characteristics and recent child illness, household 172 characteristics, and community characteristics. The child characteristics include the child's age, the gender 173 of the child, birth order, and whether the child is a twin or not. The sex of the child is recorded by a dummy 174 variable equal to 1 for a male and 2 for a female child. If there is gender inequality concerning children's 175 care, we expect female children to be better nourished than male children if female children are favoured 176 and vice-versa. In other studies, female children were found to have a better nutrition status than male 177 children (16). Then we also took into consideration the characteristic of whether a child is from multiple 178 births or not. It was recorded by a dummy variable equal to 1 for a child from multiple births and zero for 179 a single born child. 180 The child's age has an upper boundary of five years, which was measured in months since stunting is the 181 failure to grow optimally and is first detected in children who are considered short for their age group when 182 they are two years old . Therefore, we expect the children who are two years or more to have worse nutrition 183 status than those younger than two years. The child's birth order and recent illness are also characteristics 184 that we are to access against the Nutrition of status of the child. The child's illness was the recent incident 185 of diarrhoea. Whether the child had a fever in the last two weeks or not, both were captured with a dummy 186 variable equal to zero for no incident of diarrhoea or fever. A dummy variable equal to 1 represents a yes; 187 there was recently an incident of diarrhoea or fever. 188 The household characteristics include the household's wealth status, mother's BMI, mother's age at birth of 189 a respective child, mother's working status, mother's highest level of education, father's highest level of 190 education, and father's main occupation grouped. The wealth index of household economic status was 191 constructed in the MWDHS-2015 report by using the information on household ownership of assets and 192 dwelling characteristics. The wealth index was recorded with dummies representing the three categories of 193 poor, middle, and rich. The mother's BMI was captured in two categories, Thin (BMI≤18.5) and Normal 194 (BMI>18.5). The expectation was that an underweight mother would result in worse nutrition status for the 195 child since she would lack the fat needed to produce adequate milk for her child. 196 The female parent's age at the birth of the respective child was also categorized in five different levels. 197 There is evidence elsewhere that children born to mothers below 18 and above 34 years of age are more 198 likely to be malnourished when compared with the children born to mothers aged 18 to 34 (20). 199 The mother's current working status was captured by a dummy variable equal to one for the currently 200 working mother and 0 for a non-working mother. The highest level of education of a mother was measured 201 with dummies representing four categories: no education, primary, secondary and higher education. The 202 highest level of education for a father was captured in the same manner. Education affects caregiving 203 practices through the ability to process information, acquire skills, and model behaviour. It can be 204 hypothesized; therefore, those educated parents are associated with the child's high nutritional status. They 205 can better use healthcare facilities and ensure a high standard of environmental sanitation (16). 206 Five dummy variables of a father's occupation measure different types of employment. Thus, this 207 independent variable is categorized as not working, professional, business, agriculture, and other 208 occupational sectors. The prevalence of stunting was significantly lowest among the children of fathers 209 who were service holders in a study of predictors of chronic child malnutrition in Bangladesh by Das (2011). 210 One would expect children of fathers who hold professional positions (such as managerial positions) to 211 have better nutritional status since in professional occupations; parents usually get maternity leave to care 212 for their new-born baby, which may help them ensure an early healthy lifestyle for the child. 213 The community variables used in this study include water sources and the type of residence of the 214 household. Source of water is captured with dummy variables for piped water, well water, and other sources 215 (such as rain, tank, etc.). The residence household is captured by dummy variables equal to one for urban 216 and two for rural areas. The expectation is that children from urban areas would be associated with better 217 nutritional status since the availability of health and educational facilities is greater in urban areas compared 218 with rural areas. The duration of breastfeeding was also added to the independent variables to assess the 219 effect of breastfeeding. Breastfeeding is captured by dummies categorized into three groups: ever breastfed, 220 then stopped, never breastfed, and still breastfeeding. 221 Recent several other studies (21-23) utilised the following variables were mostly found to be significant 222 predictors of malnutrition; Socio-economic and demographic characteristics, such as family size, monthly 223 family income, mother can read and write, mother's educational status, mother's marital status feeding 224 practices, diarrhoea in the past two weeks, and age of child variables include. The current study have used 225 some of this variables to investigate association with childhood malnutrition in Malawi. 226

Descriptive statistics 228
The study utilized a quantitative research methodology appropriate for investigating the relationship 229 between two or more variables known as cross-tabulation through version 25.0 of the SPSS statistics 230 application. The use of cross-tabulation enabled us to examine associations within our selected independent 231 variables and the ordered categorical response variable (child nutrition status). Pearson's chi-square test or 232 the Chi-square test of association was used to discover if there is a relationship between the level of stunting 233 and each independent variable. The results of these descriptive statistics are displayed in Table 1. The test 234 provided a necessary additional understanding of the data as we were able to isolate the most critical 235 variables (with p-value < 0.005). The chi-square test results show that some of the independent variables 236 are highly significant with the response variables. It is from descriptive statistics results as presented in 237 Table 1 we were able to select variables to be fitted in the PO model (introduced in section 3.2). 238 Table 1 here. 239

Components of generalized linear models 249
A proper kick-off on the discussion of GLMs is to briefly discuss its traditional models from which they 250 emanate, which are known as classical linear models. We consider an observation vector possessing 251 components, which is presumed to be a design of arbitrary variable whose components are distributed 252 independently with µ, the mean of the random variables. The essential technical part of the current model 253 is the classification of the vector µ through the use of a smaller number of unknown parameters 1 ,· · ·, . 254 (where is × , µ is × 1, and is × 1), is the parameters' vector, and is the matrix of 263 the model. This completes the specification of the systematic part of the model. 264

The generalization 265
The simple path to describe the shift from classical linear models to GLMs is through rearranging the first 266 expression in 3.2.1. The first expression in 3.2.1 is required to have three-part characteristics know as: The 267 systematic componentwhich is qualified when the covariates 1 , 2 ,· · ·, produce a linear predictor ,

Model Fitting 280
In general, when one is working with data that possesses a categorical response variable, and the categories 281 are more than two, the GLMs provide two choices to be considered by the model developer. Since the possible outcomes for our response variable consist of three categories, and ordinal in design, the 294 use of logistic regression, which is usually model the probability of one of the outcome, named "success" 295 is not applicable. However, there exist regression models that are specifically designed for this situation; 296 these are the expansions of the logistic-regression-model with a response a binary response. The intricacy 297 in fitting ordinal regression models arises to some degree because there are countless opportunities for how 298 "success" and the resulting likelihood of "success" may be represented in the model. Below we discuss one 299 of the commonly used procedures modelling categorical, ordinal response variables. 300

Model assumptions testing 338
The study has an ordinal response variable which is nutritional status. As discussed above, the ordinal 339 response variable was derived by grouping a continuous variable; the height-for-age anthropometric index. 340 Statistically, when one has such a data design, it is appropriate to formulate a PO model. The chi-square 341 score has then utilized to test the appropriateness of developing the PO model; testing the PO assumption. 342 This test is primarily conducted to confirm whether the main PO model assumptions are violated or not. 343 However, there are empirical concerns about the chi-square score test; it tends to result in a too-small p-344 value. As a result, the current study conducted, above the chi-square score test, a supporting technique for 345 investigating the PO assumption. This technique computes the single score tests for each covariate as an 346 alternative method of testing whether the PO assumption is violated (30). Results of the test are displayed 347 in Table 2. 348 Firstly, from Table 2, the chi-square score test for testing the PO assumption is not significant at a 5% level 349 of significance ( 2 = 36.94, p-value = 0.3347). Therefore, the design of our data set does not violate the PO 350 assumption. In addition, this demonstrates that for every one of the picked covariates, one parameter may 351 be utilized for separate modelling logits of cumulative probabilities. However, to confirm the chi-square 352 test results' correctness, we have conducted the alternative or supporting testing technique; single covariate 353 score test. The outcome of the letter test is displayed in the last column of Table 2. The outcome indicates 354 that all the covariates were insignificant (p-value ≤ 0.005), and hence, they satisfy the PO assumption. From 355 the latter results, we were able to make a final decision that our dataset satisfies the PO assumption for 356 POM. 357

Discussions 358
The observations from fitting the PO model for determining risk factors for childhood stunting in Malawi 359 are displayed in Table 2. We have evaluated the goodness-fit test to assess the PO model's overall adequacy 360 on the data set. The test results suggested that the overall model did not lack any type of fit, with p-value = 361 <.0001 (see the last row of Table 2). The data set used in the current study was collected in 2015/16 DHS. 362 Hence, the study results must be read as a likelihood that similar conditions are still evident in Malawi. This 363 is a strong assumption motivated by the argument of no evidence, which suggests that the risk factors of 364 stunting revealed by the current study have already been addressed. 365 The column labelled odds ratio in Table 2 displays the values of the estimated adjusted odds ratios. 366 Therefore, the interpretation of our results is based on adjusted odds ratios. The PO model results revealed 367 that the odds of possessing worse nutrition status (stunting) are 3.62, 4.19, and 3.91 higher among children 368 belonging to the age group 12-23, 36-47, and 48+ months respectively when compared with the infants. 369 This result supports the earlier hypothesis; stunting can be clearly identified when the child is two years 370 old. Moreover, this result supports evidence from a study on childhood malnutrition conducted by Das 371 (2008) in Bangladesh. It can be surmised that when a child is growing, the mother's milk alone becomes 372 insufficient for feeding. As a result, we can confidently state that the reason for increased odds of stunting 373 in children of 12 or months is attributed to inadequate supplementary foods when breastfeeding is no longer 374 adequate. The odds of stunting in children younger than five years can worsen as age increases; however, 375 this is evident up to a certain age (47 months), beyond which then the odds of stunting improve materially. 376 The results of the PO model displayed in Table 2 also suggest that kids from multiple births have 3.66 (odds) 377 times higher odds of possessing stunting compared to children of single births (with p-value = <.0001). The 378 observation could be attributed to the low birth weight for children from multiple births, insufficient food 379 intake from breastfeeding, and competition for nutritional intake, which will distress the offspring of 380 multiple births. Furthermore, we can attribute this to the fact that, in general, parents can care more for 381 fewer kids, which can be added as a justification for the high risk of stunting in multiple birth children. As 382 an individual, I also applaud the Human Fertilization and Embryology Authority in its efforts to reduce 383 multiple births through the "one at a time" campaign (31); African nations can also follow other countries 384 in imposing a single embryo transfer policythis alone may play a significant role in decreasing the risk 385 of childhood medical complications caused by multiple births, possibly even the prevalence of stunting. 386 The results from Table 2 also reveal that the difference in a mother's education level plays a significant role results support the inherent results that generally, children born to mothers below 18 and above 34 are more 417 susceptible to malnourishment than children born to mothers aged within 18-34 years (20,35). 418 Within the two weeks before the survey, children who suffered from diarrhoea were at a 1.2 times higher 419 risk of being malnourished than children who did not suffer from diarrhoea during the same time period 420 (Table 2). This suggests that the illness history of the child can serve as one of the determinants of 421 malnutrition. Number of studies have also postulated the reciprocal relationship between malnutrition and 422 diarrhoea (36,37). 423 Inset Table 2 here. 424

425
The current research was conducted with the primary goal of revealing factors that might be associated with 426 the condition of high childhood malnutrition in Malawi. The study revealed the risk factor of malnutrition 427 in Malawi and deeper insights on how the probability of stunting evolves across the different levels of risk 428 factors. The risk factors of malnutrition from the current study are found to be; the age of the child; birth 429 type (singleton/multiple births), parents' level of education, household's type of resident; mother's age at 430 the time of birth, mother's BMI, the incident of diarrhoea in the last two weeks before the survey. From the 431 results of the current study, we can deduce that Malawi's economic state is possibly the centre of the causal. 432 However, this can be subjective as it was not proven by any statistical model utilized herein. All the risks 433 mentioned above factors are controllable, and they can be improved one way or another. In US Government 434 launched the Feed the Future Guide, describing its strategy to address global hunger and food security.