Childhood malnutrition is a real-life and a chronic problem and one of the global major
public health challenges, especially in developing countries like Ghana. Several
attempts from governmental and non-governmental organizations to address the
problem have fallen below expectation. It is recognised that the existing studies and
nutrition intervention strategies are inadequate and hence not working to expectation.
This thesis examines childhood malnutrition in Ghana using appropriate and advanced
statistical methods to help improve the understanding of childhood nutrition and to
better inform targeted public health nutrition interventions in the country.
In this thesis, we provided solutions to five main problems: (1) investigated the major
risk factors for malnutrition; (2) investigated household level variations in nutritional
outcomes of children; (3) explored, modelled and illustrated spatial variations in the risk
of childhood malnutrition over Ghana; (4) explored, modelled, forecasted and illustrated
spatio-temporal variations in the risk of childhood malnutrition over Ghana; (5) jointly
modelled weight-for-age Z-score (WAZ) and height-for-age Z-score (HAZ) to improve
accuracy and reliability in estimates. To answer the first and the second problems,
multilevel models were considered. The results showed strong residual household-level
variations in under-fives nutritional outcomes and that child’s age, type of birth, child’s
experience of diarrhoeal episodes, size of child at birth and months of breast feeding,
mother’s education, current age, BMI and national health insurance status, household
toilet facility ownership and wealth status were predictive of under-fives nutrition.
To answer the third problem, spatial models were employed. The study found
substantial spatial variation in the predicted risk of under-fives malnutrition over Ghana
and also showed that Normalised Difference Vegetation Index (a marker for vegetation
iv
cover), elevation and rural/urban residence status were predictive of under-fives
nutritional outcomes. The study considered spatio-temporal models to answer the fourth
problem. The results showed substantial spatio-temporal variation in the risk of underfives
chronic malnutrition over Ghana. Our forecasted map of chronic malnutrition
showed substantial spatial variation with children from parts of Northern and Western
regions being at the highest risk of malnutrition compared to children from other
regions of the country. In our forecast maps, the effect of increasing the level of
maternal education was shown to reduce the prevalence of malnutrition throughout
Ghana.
To answer the fifth problem, multivariate response multilevel models were considered.
The study found that the residual household effects for WAZ and HAZ are very
strongly correlated and that the correlation was stronger for the residual household
effects than the residual child effects. This also suggests that after adjusting for risk
factors in our model, it is the same as-yet unidentified factors at household level that
influence both WAZ and HAZ. The results also showed that there was more accuracy
and reliability in estimates from the multivariate response multilevel model over
separate multilevel models and showed that the effect of some important risk factors
differed substantially across WAZ and HAZ.
The findings from this thesis are intended to help policymakers responsible for the
health and nutrition of children to design efficient public health policies and targeted
nutrition interventions amidst scarce public health resources available in Ghana to better
understand, target and to reduce childhood malnutrition prevalence closer to the level
expected in a healthy, well-fed population of children under-fives.