Machine Learning-based Prediction Model for Adverse Pregnancy Outcomes: A Systematic Literature Review

Most of Adverse Pregnancy Outcomes (APO) are preventable particularly if the health personnel can early detect the risk. This study aimed to review articles on how the machine learning model can predict APO for early detection to prevent neonatal mortality. We conducted a systematic literature review by analyzing seven articles which published between 1 January 2013 and 31 October 2022. The search strategy was the populations are pregnant women, intervention using machine learning for APO prediction, and the outcomes of APO are Low Birth Weight, preterm birth, and stillbirth. We found that the predictors of LBW were demographic, maternal, environmental, fetus characteristics, and obstetric factors. The predictors of preterm birth were demographics and lifestyle. Meanwhile, the predictors of stillbirth were demographic, lifestyle, maternal, obstetric, and fetus characteristics. It was indicated that Random Forest (Accuracy: 91.60; AUC-ROC: 96.80), Extreme Gradient Boosting (Accuracy: 90.80; AUC-ROC: 95.90), logistic regression (accuracy 90.24% and precision 87.6%) can be used to predict the risk of APO. By using a machine learning algorithm, the best APO prediction models that can be used are logistic regression, random forest, and extreme gradient boosting with sensitivity values and AUC of almost 100%. Demographic factors are the main risk factors for APO.