A predictive model and socioeconomic and demographic determinants of body mass index in Sudan

This paper aims at determining the socioeconomic and demographic determinants related to Body Mass Index (BMI) for children under-five years in Sudan. This study uses the Sudan Multiple Indicator Cluster Survey (MICS) conducted in the year 2014. The survey was conducted by the Central Bureau of Statistics in cooperation with several national institutions. The objective of the study is to identify the factors of BMI for under-five children. The multinomial logistic regression model was adopted. Results revealed that the prevalence of underweight for infant’s under-five years was 86.3% for females and 85.3% for males, the proportion of the normal weight of infants under-five years of age for males is slightly higher than that of females; there is correlation amid geographic determinants; state, and BMI status. Also, there was a significant association between mother’s education and body mass index status and the wealth index and body mass index status. The variables place of residence and sex did not show a statistically significant relationship with body mass index status for the children under-five years of age in Sudan. In addition, the risk factors significantly associated with body mass were the state, child ever breastfeeds, sex, mother’s education, wealth index, and age in the month.


Background
Medically, Body Mass Index (BMI) is a main Index employed to connect weight and height. It is an individual's mass measured in kilograms (kg) divided by a person's highness in meters. Standard mass and obesity are identified by the National Institute of Health (NIH) based on BMI of 27.3 or over for females and 27.8 for males while obesity is considered a BMI of more than 30 for both genders [12].
Worldwide, it is predicted that 165 million under-five infants were stunned in 2011. This phenomenon occurred Africa (36%) and in Asia (27%); it remains a public health issue, but it is unnoticed. These percentages represent 90% of the glob's undersized infants [8]. [16] found no significant variations in percentage of hefty people. Being skinny was associated with age, number of banquets, level of parent schooling, origin of food, and quantity of stock possessed. [4] revealed that BMI was positively correlated with gender (females), but it was negatively connected to age and educational standard. Hefty, obesity and underweight amid Bulgarian infants and teenagers were connected to nurturing manner and gorge for both genders [13] was found to rise in the five distributions of fortune in both genders. BMI and fortune index were also found to be correlated [9] A lofty incidence of gross and obesity was strongly related to females. This indicates a relation between females, attending commercial schools, good socioeconomic position [2]. [18] concluded that infants under five, mass of an infant at birth, material's age; BMI; social status and area Affar, Dire Dawa, gambela, Harari & Somalia were factors considerably affect underfive infant's nutrition habits in Ethiopia. A study conducted in Khartoum State, Sudan revealed that 20.9% were badly sustained and 79.1% of them were well sustained, but approximately 15.4% of the infants were skinny, 8.8% of them were fairly skinny, and 6.6% were seriously underweight [14]. Another study conducted in Ghana by [6] showed that the occurrence of underweight, lavish, and undersize was 10.4%, 5.3%, and 18.4%, respectively. The infant's age was related to underweight, lavish, and undersize while gender was connected to lavish and undersize. Regular or overweight/obese, Mother BMI class, female's self-sufficiency, and middle class fortune index were related to smaller odd malnutrition [1]. The study conducted by [5] concluded that the incidence of overweight and obesity was 34% and 4.97, respectively. Women possessed a loftier frequency of overweight (38.3%) compared to (30%) among men. Obesity was more common among women (7.4%) than amid men (2.4%); most participants were healthy weight (50.9%). A study conducted by Sen, Mondal & Dutta (2013) found that the existence of overweight and obesity was noticed to be high among both genders; it was (23.67% and 9.67%) amid men and (20.33%% and 29.33%) among women. Gender, age, monthly revenue, spousal status, schooling, and alcohol intake were noticed to have considerable impact on obesity.
Additionally, they had noticeable impact on joined overweight-obesity.

Study variables
The response variable for this study is BMI for the under-five child in Sudan which is a multinomial variable. The independent variables used in this research are the place of residence, state, mother's education, gender, age of the child, the child still breastfeeding, and wealth index. The demographic and socioeconomic factors used in this study were proposed by different [10]; [17]; [7].

Statistical methods
The multinomial logistic regression models are an extension of the ordinary logistic models where we study a categorical response variable with more than two possible outcomes [3].
Let J represents the number of categories for the response Y. let 1 , 2 ,……., represent the response probabilities, satisfying ∑ =1 =1. With independent observations, the probability for the number of responses of the is multinomial distribution and can be expressed as follows: The logit link function will be adopted for the multinomial logistic regression model. A response variable that has more than two nominal categories can be modeled adopting multinomial logistic regression. It is important to notice that whether the response variable Where = 1,2, . . . . , − 1; is the outcome from the baseline category, which can be any category but is commonly the highest one: represent the constants , and 1 1 , 2 2 , ….., are the coefficients of the multinomial regression model. As the model has − 1 comparisons, it estimates the − 1 logit function for each predictor [11].
To predict the multinomial logistic regression model, the maximum likelihood technique will be used. Regarding the nominal categories, one of the categories is considered as a reference or baseline category and the rest of the categories are compared with the reference category [3].

Results
Firstly, bivariate tests were carried out, before performing the multinomial logistic regression analysis. The cross-tabulation, as a basic approach of descriptive analysis, was performed adopting the tests of chi-square to explore the association between the response of body mass index and many categorical socio-economic, demographic, and geographic variables at the 5% level of significance. Therefore, Table 1 displays the relationship between body mass index and selected socio-economic, demographic, and geographic categorical variables. There was a statistically significant association at the 5% level among geographic factors; state, and body mass index status (p-value < 0:0001). Among demographic variables and body mass index, a significant association was noticed, i.e., between mother's education and body mass index status (p-value < < 0.0001). Similarly, significant relationship was found between wealth index and body mass index status ((pvalue < < 0.0001). In contrast, place of residence (p-value = 0.588) and sex (p-value=0.095) did not show a statistically significant association with body mass index status. For this study, the proportional odds model was adopted to the 2014 Sudan Multiple Indicator Cluster Survey. The basis for outcomes with most multinomial regression models is the logit function. The distinction between logit and probit functions is found in small samples. This is because the probit link consider the normal distribution of the probability of an event. But the logit link assumes the logistic distribution. For the data derived from complex survey design, it is unsuitable to run the proportional odds model analysis for the ordinal response variable ignoring the design of survey sample. Disregarding the survey sampling information may provide biased estimates of parameters, incorrect variance, and parameter estimates. This is due to overestimated or underestimated parameters and variance estimates. For this kind of situation, a specialized method to obtain suitable estimates and standard errors for the ordinal outcome variable should be adopted. This technique includes the weight in the survey sampling design.
Therefore, the findings of the analysis are given in Table 2. Table 2  The risk factors significantly associated with body mass index were found to be the state, child ever breastfeeds, sex, mother's education, wealth index, and age in the month ( Table   2). Children of mothers who have attended secondary high school ((OR= 1.45, C.I.  and females underweight. The proportion of the normal weight of males is slightly higher than that of females. This result is contradicted with the result obtained by [16]. The results of the study showed that there is an association between geographic factors; state, the status of body mass index status. A significant relationship between mother's education and body mass index status. Also, a significant association was found between the wealth index and body mass index status. From the other side, place of residence and sex did not show a statistically significant relationship with body mass index status. The risk factors significantly associated with body mass index were found to be the state, child ever breastfeeds, sex, mother's education, wealth index, and age in the month. This result is in the same line as [15].

Conclusion
This study shows that the factors associated with body mass index of Sudanese children under five age were child ever breastfeed, sex, mother's education, wealth index, age in the month, and states. Also. There was a high percentage of underweight children under five age in most Sudanese states. Evidence revealed that the body mass index is associated with a wealth index. The findings of this study will help policymakers to focus on the