On selection of an appropriate logistic model to determine the risk factors of childhood stunting in Bangladesh

Abstract

Stunting is the core measure of child health inequalities as it reveals multiple dimensions of child health and development status. The main focus of this study is to show the procedure of selecting the most appropriate logistic regression model for stunting by developing and comparing several plausible models, which ultimately helps to identify the predictors of childhood stunting in Bangladesh. This study utilizes child anthropometric data collected in the 2014 Bangladesh Demographic and Health Survey. Valid height-for-age anthropometric indices were available for a total of 6,931 children aged 0-59 months, of which about 36% were stunted. Ordinary logistic, survey logistic, marginal logistic, and random intercept logistic regression models were developed assuming independence, sampling design, cluster effect, and hierarchy of the data. Based on a number of model selection criteria, random intercept logistic model is found the most appropriate for the studied children. A number of child, mother, household, regional, and community-level variables were included in the model specification. The factors that increased the odds of stunting are children older than 11 months, short birth interval, recent morbidity of children, lower maternal education, young maternity, lower maternal body mass index, poor household wealth, urban residential place, and living in Sylhet division. Findings of this study recommend to utilize an appropriate logistic model considering the issues relevant to the data, particularly sampling design and clustering for determining the risk factors of childhood stunting in Bangladesh.

Prevalence of child undernutrition measures and their spatio-demographic inequalities in Bangladesh: an application of multilevel Bayesian modelling

“…After accounting for the district and division level random effects, the fixed effects components indicate that children aged 24-35 months, living in Sylhet division and rural domains have higher possibility of being stunted and underweight. These patterns are expected, based on existing studies which have explored risk factors of children undernutrition in Bangladesh [ 16 , 24 , 50 ]. For wasting, most of the fixed effects are found insignificant except for sex (male domains have slightly higher prevalence) and children age-group (12-23 months age-groups has lower prevalence than infants).…”

Differences in Tsimane children’s growth outcomes and associated determinants as estimated by WHO standards vs. within-population references

“…Mean WHO HAZ scores decline systematically across early childhood in resource poor-settings due to nutritional and infectious conditions [21–24], resulting in increased age-related variance in WHO HAZ scores that may bias parameter estimates in mixed-age samples [25,26]. Although researchers often control for child age in statistical models [27,28,29], the systematic deviance in WHO-derived z-scores may bias or confound estimates of size differences associated with locally varying determinants, particularly those correlated with age or developmental changes. In contrast, within-population growth references should minimize the influence of endemic influences in estimating relative size, resulting in more accurate and biologically relevant estimates of local growth determinants in regression models.…”

Estimation of child undernutrition for spatiodemographic small domains in Bangladesh An application of multilevel Bayesian model

“…After accounting for the district and division level random effects, the fixed effects components indicate that children aged 24-35 months, living in Sylhet division and rural domains have higher possibility of being stunted and underweight. These patterns are expected, based on existing studies which have explored risk factors of children undernutrition in Bangladesh (Bhowmik & Das, 2019; M.R.K. Chowdhury et al, 2016;Das & Rahman, 2011).…”