Abstract:

Objective To investigate high-risk factors for neonatal hyperbilirubinemia(NHB) and establish and evaluate a NHB risk prediction model using multiple machine learning methods, providing evidence for NHB prevention and treatment research. Methods The neonates from the pediatric outpatient and inpatient departments of Shandong University of Traditional Chinese Medicine Affiliated Hospital were enrolled. Data on neonatal hyperbilirubinemia were collected via questionnaire surveys. Statistical analysis was performed using the Scikitlearn machine learning software in Python3, with the Logistic regression algorithm employed to screen for hyperbilirubinemia-related indicators. Simulated data with varying sample sizes were generated from longitudinal physical examination cohort samples. Six machine learning algorithms were applied to establish neonatal hyperbilirubinemia risk prediction models. The ability and accuracy of these models to distinguish neonatal hyperbilirubinemia were validated using the area under the receiver operating characteristic curve(ROC). Results Among the six machine learning algorithms, the Random Forest model demonstrated the best overall performance. Key features ranking highly in this risk prediction model included pregnancy diseases, neonatal hemolytic disease, abnormal postnatal symptoms(infection), and preterm birth status. Conclusion The machine learning-based risk prediction model for neonatal hyperbilirubinemia holds clinical diagnostic value for its prevention and management. The resulting neonatal jaundice management system enables enhanced monitoring and management of high-risk infants, thereby reducing or preventing complications.

Key words: Neonatal hyperbilirubinemia, Prediction model, Machine learning, Risk assessment