Predicting Adverse Perinatal Outcomes among Women with Hypertensive Disorders of Pregnancy in Addis Ababa
No Thumbnail Available
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Background - Hypertensive disorders of pregnancy (HDP) are among the leading causes of
adverse perinatal outcomes (APOs) worldwide, with a particularly high burden in low- and middle-
income countries (LMICs) like Ethiopia. Despite advancements in maternal healthcare, traditional
risk assessment methods often fail to account for the complex nature of HDP, resulting in delayed
interventions and preventable maternal and neonatal complications. The lack of efficient and
accurate risk prediction tools in resource-limited settings contributes to high morbidity and
mortality rates among pregnant women and newborns.
Objective - This study aimed to develop a machine learning (ML) predictive model using
electronic medical record (EMR) data to identify women with HDP who are at high risk for APOs.
Methodology - This study applied a retrospective analysis of EMRs collected from selected public
hospitals in Addis Ababa (AA), focusing on predicting APOs among pregnant women diagnosed
with HDP. A total of 1042 instances was collected from four hospitals in AA. After performing
data cleaning and preprocessing, including handling missing values using three imputation
techniques (Median, KNN, and MICE), six supervised ML algorithms—Logistic Regression,
Naïve Bayes, Support Vector Machine, Decision Tree, Random Forest, and XGBoost—were
developed. Model performance was evaluated using, internal cross-validation, two hyperparameter
optimization strategies (Grid Search and Bayesian Optimization), and with metrics including
accuracy, sensitivity, precision, F1-score, AUC-PR and ROC-AUC.
Result - Random Forest and XGBoost models had the highest predictive performance, with
XGBoost achieving the best overall results. However, even the best-performing model reached a
maximum recall of only 69% for predicting the positive class, which indicates limitations in
identifying APOs.
Conclusion - The findings indicate that ML models show promise in predicting APOs among
women with HDP using routinely collected clinical data. Most models achieved high specificity
and ROC-AUC value, showing their potential for use in risk stratification. However, the recall
performances were suboptimal for clinical application, which suggests the current models are not
yet suitable for deployment. Further research with larger datasets, more diverse features, and
broader validation is necessary.
Description
Keywords
Predicting Adverse Perinatal Outcomes among Women