Understanding drivers of neonatal mortality in Zimbabwe: A machine learning approach using survey data

Area of Science:

• Public Health
• Machine Learning
• Demography

Background:

• Neonatal mortality rate (NMR) in Zimbabwe exceeds the regional average and is increasing, hindering progress towards Sustainable Development Goal 3.
• Despite improvements in other child mortality indicators, NMR remains a critical challenge in Zimbabwe.

Purpose of the Study:

• To utilize machine learning to identify the primary predictors of neonatal mortality in Zimbabwe.
• To inform targeted interventions for reducing neonatal deaths in the country.

Main Methods:

• Pooled secondary data analysis of three Zimbabwe Demographic Health Surveys (2005-2015) involving 16,941 live births.
• Development and evaluation of multiple supervised binary classification machine learning models to predict neonatal death.
• Selection of the best model using sensitivity and Area Under the Receiver Operating Curve (AUC ROC), followed by logistic regression for variable importance and effect assessment.

Main Results:

• The eXtreme Gradient Boosting Model was the best performing model, achieving a sensitivity of 0.74.
• Key predictors identified include early breastfeeding initiation, birth weight, household size, and newborn postnatal care (PNC).
• Low birth weight was positively associated with neonatal mortality, while larger household size, early breastfeeding, and PNC were negatively associated.

Conclusions:

• Machine learning effectively identifies critical factors influencing neonatal mortality in Zimbabwe.
• Interventions should prioritize preventing and managing low birth weight infants.
• Promoting early breastfeeding initiation and ensuring newborn postnatal care are crucial for improving neonatal survival.