Perinatal mortality, encompassing stillbirths and neonatal deaths, remains a significant public health challenge, particularly in regions like Africa, where mortality rates are the highest globally. Despite improvements over the past two decades, the newborn phase remains the most perilous period of life. This article summarises findings from a study published in the peer-reviewed Journal of Artificial Intelligence where the researcher applied the advanced machine learning algorithms on the primary data and predicted the women with risk factors backed by robust model evaluation with higher accuracy. The researcher has aimed to use his expertise for both social and commercial good.
The primary goal of the study was to distinguish pregnant women at risk of delivering vulnerable babies in Malawi, a country in southeastern Africa with high perinatal mortality rates. The study sought to identify early predictors of adverse newborn outcomes by leveraging sophisticated computational techniques.
The research utilized a dataset comprising various features, ranging from medical records to the socioeconomic conditions of the women. The output variable was binary, indicating either a normal or adverse pregnancy outcome. The data was initially unbalanced, with more normal deliveries than adverse ones, necessitating the use of the Synthetic Minority Oversampling Technique (SMOTE) to address this imbalance.
Two ensemble models, Random Forest and Gradient Boosting, were trained to predict perinatal mortality. These models demonstrated higher accuracy and precision in identifying adverse pregnancy outcomes. Additionally, feature selection methods and Shapley Additive Explanations (SHAP) were used to pinpoint the most significant risk factors influencing mortality.
The study identified 56 features from medical, economic, and social factors, as critical determinants of perinatal mortality:
- Medical Factors:
- Respiratory rate, weight of mothers, and blood pressure levels were crucial in distinguishing normal births from adverse outcomes.
- Women who had received iron infusion treatments (either orally or via ferric carboxymaltose) had better pregnancy outcomes, highlighting the importance of addressing iron deficiency.
- Economic Factors:
- Economic hardships significantly influenced birth outcomes. Women who spent less on hospital care, whose families lacked land ownership, and who did not have access to safe drinking water were more likely to give birth to unhealthy babies.
- Daily wage earners were particularly vulnerable.
- Social Factors:
- Illiteracy, early pregnancies, tribal and religious affiliations also played a role in determining the birth outcomes. These factors underscore the multifaceted nature of perinatal mortality, where both health and socio-cultural dimensions intersect.
The research has highlighted the effectiveness of machine learning models in predicting perinatal mortality and identifying high-risk factors among pregnant women in Malawi. By providing early warnings and identifying critical risk factors, these models can help healthcare providers implement targeted interventions to improve maternal and newborn health outcomes.
With their ability to process and analyse complex datasets, machine learning techniques offer valuable insights into the various determinants of perinatal mortality. This study demonstrates their predictive power and emphasizes the need for comprehensive strategies that address medical, economic, and social determinants to reduce perinatal mortality in high-risk regions.
