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

TL;DR

This study uses machine learning to identify key factors affecting neonatal mortality in Zimbabwe, highlighting the importance of early breastfeeding and postnatal care.

Contribution

The study introduces a machine learning approach to uncover novel predictors of neonatal mortality in Zimbabwe.

Findings

Early breastfeeding initiation and postnatal care significantly reduce neonatal mortality.

Lower birth weight is positively associated with neonatal mortality.

Household size negatively correlates with neonatal mortality odds.

Abstract

In Zimbabwe, the neonatal mortality rate (NMR) is higher than the regional average, and the country is not on track to reach the Sustainable Development Goal of reducing the NMR by 2030. While other child mortality indicators have improved, NMR has increased. Using machine learning, we aimed to identify the key predictors of neonatal mortality in Zimbabwe. Pooled secondary data analysis of three rounds of the Zimbabwe Demographic Health Survey (ZDHS) from 2005 to 2015 was done. The study population was all the live births born to women aged 15–49 years within the 5 years prior to each round of the survey (n = 16,941). Multiple supervised binary classification machine learning models were built to predict neonatal death based on socio-economic, mother’s demographic, prenatal, delivery, and neonatal characteristics available in ZDHS. Sensitivity and area under the receiver operating curve (AUC ROC) were used to select the best model for the prediction of neonatal mortality. The best model was used to identify relatively important variables, and logistic regression was used to assess the magnitude and direction of effect. The eXtreme Gradient Boosting Model outperformed other models with a sensitivity of 0.74. Early breastfeeding initiation, birth weight, household size, and newborn post-natal care (PNC) were identified as the top predictors of neonatal mortality. Logistic regression revealed that lower birth weight (aOR [95%CI]: 0.9997 [0.9995 – 0.9999]) was positively associated with odds of neonatal mortality, while household size (aOR [95%CI]: 0.84 [0.80 – 0.89]), early breastfeeding initiation, aOR (0.28 [95%CI]: [0.21 – 0.37]) and newborn postnatal care, (aOR [95%CI]: 0.08 [0.06 – 0.11]) were negatively associated with the odds of neonatal mortality. This study demonstrates the potential of machine learning in identifying key predictors of neonatal mortality in Zimbabwe. To accelerate the reduction in neonatal mortality, interventions should focus on preventing and specially managing low birth weight babies to improve survival. Furthermore, health facilities and community-level support for early initiation of breastfeeding and PNC checks should be promoted for all eligible newborns.

Figures