Prediction of risk of pre-eclampsia among pregnant women at the China-Uganda Friendship Hospital, Naguru

Abstract

Background: Preeclampsia presents a highly prevalent burden on pregnant women with a global estimated incidence of 4.6%. Preeclampsia increases the maternal risk of death and is one of the leading causes of perinatal morbidity and mortality. In Uganda, 336 women die per 100,000 live births due to pregnancy-related causes. The Department of Obstetrics and Gynecology of China-Uganda Friendship Hospital in Naguru, receives about 520 mothers with severe preeclampsia, 132 with eclampsia, 14 with postpartum eclampsia, and 12 maternal deaths in 2017. The continued occurrence of maternal death and other adverse outcomes due to hypertensive disorders of pregnancy in Uganda is an indication of failure to effectively identify the disease early enough. Objective: The aim of the study was to develop, train, and test a prediction model for the risk of pre-eclampsia among pregnant women and to develop a web-based application that uses machine-learning algorithms to predict women’s risk of preeclampsia, to be used by healthcare workers. Method: A retrospective study utilizing data from 2,193 antenatal care records and key predictors such as maternal age, symptoms (headaches, visual disturbances), and clinical signs (blood pressure, proteinuria), was analyzed using advanced machine learning algorithms including Logistic Regression, Decision Trees, and Gradient Boosting Classifier. The Gradient Boosting Classifier emerged as the optimal model, achieving an Area Under the ROC Curve (AUC) of 0.85, indicative of robust discriminative ability. The study highlights age and proteinuria as critical factors influencing preeclampsia risk assessment. Ethical considerations were upheld throughout, ensuring confidentiality and data protection. Results: Of the five models developed, the Gradient Boosting Classifier performed the best with an accuracy of 87.42% and an AUC of 0.85, indicating excellent discriminative ability. The key factors identified as predictors of preeclampsia included maternal age, proteinuria, systolic blood pressure, and the presence of headaches or visual disturbances. These factors contributed significantly to the model’s predictive power, highlighting the potential for improving early detection in clinical settings. Conclusion: Machine learning, particularly the Gradient Boosting Classifier, shows strong potential for predicting preeclampsia and aiding in early interventions. The web-based tool developed in this study can support healthcare workers in timely diagnosis, potentially improving maternal health outcomes in resource-limited settings.