A neonatal sepsis prediction algorithm using electronic medical record data from Mbarara Regional Referral Hospital

Neonatal sepsis is a global challenge that contributes significantly to neonatal morbidity and mortality. The current diagnostic methods depend on conventional culture methods, a procedure that takes time and leads to delays in making timely treatment decisions. This study proposes a machine learning algorithm utilizing electronic medical record (EMR) data from Mbarara Regional Referral Hospital (MRRH) to enhance early detection and treatment of neonatal sepsis. We performed a retrospective study on a dataset of neonates hospitalized for at least 48 hours in the Neonatal Intensive Care Unit (NICU) at MRRH between October 2015 to September 2019 who received at least one sepsis evaluation. 482 records of neonates met the inclusion criteria and the dataset comprises 38 neonatal sepsis screening parameters. The study considered two outcomes for sepsis evaluations: culture-positive if a blood culture was positive, and clinically positive if cultures were negative but antibiotics were administered for at least 120 hours. We implemented k-fold cross-validation with k set to 10 to guarantee robust training and testing of the models. Seven machine learning models were trained to classify inputs as sepsis positive or negative, and their performance was compared with physician diagnoses. The results of this study show that the proposed algorithm, combining maternal risk factors, neonatal clinical signs, and laboratory tests (the algorithm demonstrated a sensitivity and specificity of at least 95%) outperformed the physician diagnosis (Sensitivity = 89%, Specificity = 11%). SVM model with radial basis function, polynomial kernels, and DT model (with the highest AUROC values of 98%) performed better than the other models. The study shows that the combination of maternal risk factors, neonatal clinical signs, and laboratory tests can help improve the prediction neonatal sepsis. Further research is warranted to assess the potential performance improvements and clinical efficacy in a prospective trial. • The combination of maternal risk factors, neonatal clinical signs, and laboratory tests can effectively diagnose neonatal sepsis. • The proposed algorithm can help identify neonatal sepsis cases as it exceeded clinicians' sensitivity and specificity. • Machine Learning (ML) algorithms can identify neonatal sepsis cases within a large and complex database. • Support Vector Machine (SVM) model with radial basis function, polynomial kernels, and Decision Tree (DT) model performed better in predicting neonatal sepsis.