Improving Prediction and Risk Factor Analysis of Low Birthweight Baby Outcomes in a U.S. Hospital System

Yang Ren, Dezhi Wu, Yan Tong, Ana Lòpez-De Fede, Sarah Gareau

Abstract: Despite considerable progress in reducing neonatal mortality in recent decades, Low Birthweight (LBW) remains a prominent cause of neonatal death and is associated with long-term adverse health effects. The birth outcome is affected by negative inpatient care signs (e.g., severe maternal morbidity) during the delivery and is impacted by different indicators related to the pregnant women's health records, demographic, and social determinants. Thus, it is critical to build accurate LBW prediction models and to find the LBW-related risk factors. In this study, we proposed to develop a prediction model to accurately identify LBW outcomes using machine learning and deep learning methods. Specifically, we applied six popular machine learning and deep learning algorithms to build our baseline models, including Naïve Bayes, support vector machine (SVM), Random Forest, XGBoost, Multilayer Perceptron, and Keras neural network. We found that the XGBoost model achieved the highest performance with an F1-score of 0.82. We also identified and quantified risk factors from the Random Forest and XGBoost models' input variables. The preliminary findings provide informative guidelines for health practitioners to understand the cause of LBW for individually targeted early prevention programs.