Prediction of mortality among neonates with sepsis in the neonatal intensive care unit: A machine learning approach

Abstract

Introduction: The use of machine learning (ML) methods can help clinicians predict neonatal sepsis better. Predicting mortality due to sepsis is essential for benchmarking and assessing NICU healthcare services. Methodology: The newborn records of those diagnosed with neonatal bacterial sepsis were reviewed retrospectively over five years. For feature selection and model development, the WEKA v-3.8.6 tool was employed. Numerous ML models, including Naive Bayes, Random Forest, Bagging, Logistic Regression, and J48 models, were created after identifying significant risk factors for newborn sepsis. Based on these models' reliability, we used them to predict sepsis and mortality in the NICU. Result: Records of 388 sepsis patients were used to build the model using training and test data sets. Mortality was best predicted using the feature selection method, OneR attribute evaluation + Ranker method, and logistic regression performed better (A = 88.4; ROC = 0.906) than others. Conclusion: These effective ML models can assist clinicians in forecasting mortality in neonates admitted to NICUs with sepsis.

Keywords

• Neonatal sepsis
• Neonatal mortality
• Random forest
• Logistic regression
• Receiver operating curve