Predicting adverse pregnancy outcome in Rwanda using machine learning techniques

Abstract

Adverse pregnancy outcomes have significant short-term and long-term effects on both the mother and the infant, these outcomes can lead to neonatal and maternal morbidity and mortality, with long-term consequences such as developmental delays, chronic health conditions, and increased healthcare costs. This study was cross-sectional and utilized the data from Rwanda demographic and health survey carried on 14634 women (RDHS, 2019-2020). K-Fold cross validation (k=10) was used to split the dataset, SMOTE solved class imbalance. Descriptive analysis determined the observed adverse pregnancy outcomes, Multivariate analysis employed to identify the factors associated with adverse pregnancy outcomes. Seven ML algorithms were employed and model performance metrics (accuracy, precision, recall, F1 Score and AUC) were evaluated to identify the best performance algorithms. The findings revealed that 93.4% of pregnancies resulted in live births, 4.5% ended in miscarriage or stillbirths (2.1%). The multiple logistic regression analysis indicated that Advanced maternal age (AOR: 3.452, 95% CI: 1.946-6.680), a higher age at first sexual intercourse (AOR: 1.421, 95% CI: 1.300 – 1.611), and many unions (AOR: 1.320, 95% CI: 1.104 – 1.573) were risk factors for adverse pregnancy outcomes. However, the risk was lower among married women (AOR: 0.894, 95% CI: 0.787 – 0.966) and women who attended antenatal care (ANC) visits (AOR: 0.801, 95% CI: 0.664 – 0.9615). The K-nearest neighbors (KNN) model was the most effective model for predicting adverse pregnancy outcomes with 86% accuracy, precision (89%), recall score (97%), F1 score (93%) and AUC (0.842). Predicting adverse pregnancy outcomes is an essential step in ensuring the health and well-being of both the mother and the infant. It allows for early identification of potential risks, enabling appropriate interventions and treatments to be implemented to prevent or manage adverse outcomes..