Background: Globally, 7.4 million young children are being killed from infectious and treatable diseases, and Saharan Africa accounts for 90% of deaths. In Uganda, Acute Respiratory Infections (ARIs) remain the leading cause of childhood morbidity and mortality among under-five children. The study aims to identify and analyze contributing potential risk factors of childhood ARIs disease among under-five children in Uganda.
Methods: A case-control study was conducted using data for 13,493 sampled under-five children retrieved from a recent 2016 Uganda Demographic and Health Survey. We compared two supervised tree-like structure machine learning algorithms and two logistic regression methods in terms of classification performances in predicting ARIs disease outcomes and analysing various child and parental socio-demographic, behavioural, and environmental characteristics.
Results: The study results revealed that the ARIs prevalence among under-five children accounted 40.3% cases. The Logistic regression findings showed that the risk of developing childhood ARIs disease declined with increase in child's age where the risk of having ARIs was higher in children in one year of birth (AOR=1.27; p < 0:001) and lower in children aged four years old (AOR=0.69; p < 0:001) compared to the infants. Other factors such as the age of mother where children born from teen mothers (15-19 years) were high likely (AOR=1.28; p < 0:001) to have ARIs illness compared to those whose mothers were in the middle age groups, and children whose mothers breastfed showed a lower risk of ARIs disease (AOR=0.83; p < 0:001) compared to those who not breastfed. In the dry season, children were more likely to develop ARIs disease (AOR=1.34; p < 0:001) compared to the wet season, and factors such as the region of residences like central region, mother employment, and domestic cooking energy like wood were also potential risk factors of ARIs disease among under-five children in Uganda. In addition, three methods such as Decision Tree (Accuracy = 61.2%; AUC=0.610), Elastic Net Logistic Regression (Accuracy=61.7%; AUC=0.627), and Binary Logistic Regression (Accuracy=62.1%; AUC=0.638) showed approximately similar performances in predicting and classifying ARIs disease outcomes. However, the Random Forest (Accuracy=88.7%; AUC=0.951) showed superior difference in performance comparatively.
Conclusion: Government and healthcare stakeholders need to make effective programs to improve public health policy against childhood infectious diseases by targeting the proper provision of maternal and child health-related education to household heads and mothers to adopt and prioritize breastfeeding practices, childcare, and ensure proper dwelling places for families and young children particularly in crowded regions and geographic places where ARIs prevalence is high.