Predicting adverse pregnancy outcome in Rwanda using machine learning techniques

Abstract

Background Adverse pregnancy outcomes pose significant risk to maternal and neonatal health, contributing to morbidity, mortality, and long-term developmental challenges. This study aimed to predict these outcomes in Rwanda using supervised machine learning algorithms. Methods This cross-sectional study utilized data from the Rwanda Demographic and Health Survey (RDHS, 2019–2020) involving 14,634 women. K-fold cross-validation (k = 10) and synthetic minority oversampling technique (SMOTE) were used to manage dataset partitioning and class imbalance. Descriptive and multivariate analyses were conducted to identify the prevalence and risk factors for adverse pregnancy outcomes. Seven machine learning algorithms were assessed for their accuracy, precision, recall, F1 score, and area under the curve (AUC). Results Of the pregnancies analyzed, 93.4% resulted in live births, while 4.5% ended in miscarriage, and 2.1% in stillbirth. Advanced maternal age(>30 years),women aged 30–34 years (adjusted odds ratio [AOR] = 5.755; 95% confidence interval [CI] = 3.085–10.074; p < 0.001), 35–39 years (AOR = 8.458; 95% CI = 4.507–10.571; p < 0.001), 40–44 years (AOR = 11.86; 95% CI = 6.250–21.842; p < 0.001), and 45–49 years (AOR = 14.233; 95% CI = 7.359–25.922; p < 0.001), compared to those aged 15–19 years, and multiple unions (polyandry) (AOR = 1.320; 95% CI = 1.104–1.573, p = 0.002), and women not visited by healthcare provider during pregnancy (AOR = 1.421; 95%CI = 1.300–1.611, p<0.001) were factors associated with an increased risk of adverse pregnancy outcomes. In contrast, being married (AOR = 0.894; 95% CI = 0.787–0.966) and attending at least two antenatal care (ANC) visits (AOR = 0.801; 95% CI = 0.664–0.961) were linked to reduced risk. The K-nearest neighbors (KNN) model outperformed other ML Models in predicting adverse pregnancy outcomes, achieving 86% accuracy, 89% precision, 97% recall, 93% F1 score, and an area under the curve (AUC) of 0.842. The ML models constantly highlighted that woman with advanced maternal age, those in multiple unions, and inadequate ANC were more susceptible to adverse pregnancy outcomes. Conclusions Machine learning algorithms, particularly KNN, are effective in predicting adverse pregnancy outcomes, facilitating early intervention and improved maternal and neonatal care.