A machine learning model to predict preeclampsia in pregnant women

Abstract

Preeclampsia is the new onset of hypertension and is one of the leading causes of maternal mortality in South Africa and the world. Preeclampsia is usually diagnosed after 20 weeks’ gestation. Due to South Africa’s poor level of antenatal care, the prediction of pregnant women at risk of developing preeclampsia could be an essential component of improving the level of antenatal care. This study used an antenatal care dataset from a South African obstetrician. A literature review identified eight risk factors: systolic blood pressure, diastolic blood pressure, maternal age, body mass index, diabetes status, hypertension history, nulliparity, and maternal disease. Two models were developed that could accurately predict the development of preeclampsia: one before 16 weeks’ gestation, and the other within three check-ups. These models were evaluated using five metrics: classification accuracy, area under the curve, precision, recall and F-Score. The results of this study show a promising future for using machine learning models in healthcare. To the authors’ knowledge, these models are the first machine-learning models for predicting preeclampsia using a South African dataset. Future work will focus on validating the models on data collected from field studies in hospitals and clinics around South Africa.