Development of Microbiota Database in the Neonatal Intensive Care Unit and Apply Artificial Intelligence to Analyze Big Data for Early Prediction of Neonatal Sepsis

Neonatal sepsis is associated with a high mortality and morbidity rate in the neonatal intensive care unit (NICU). The estimated incidence of neonatal sepsis is 0.78-2.2 per 1000 live births, and approximately one-third of all neonates who are admitted in the NICU will experience at least one episode of sepsis during hospitalization. The mortality rate of neonatal sepsis is around 6.8-19.5%. Severe neonatal sepsis is usually caused by multi-drug resistant (MDR) pathogens, has a fulminant course and is associated with the worst and devastating outcomes. Neonates with severe sepsis usually die within 48 hours after onset of neonatal sepsis. Early initiation of appropriate antibiotics can improve the outcome; however, it requires early prediction and accurate diagnosis of neonatal sepsis, especially the most important is to precisely identify that is severe neonatal sepsis. The artificial intelligence (AI) based model, using computerized integration of various parameters and deep machine learning, has been successfully applied in the adult ICU to identify physiomarkers that can be used to distinguish real sepsis onset from other non-septic conditions, identify early predict severe sepsis and guide therapeutic strategies. However, this AI based predictive model has not been explored in the NICU. Therefore, we aim to drive and validate an AI program that analyzes the routinely generated data in the NICU for early prediction of sepsis onset and severe neonatal sepsis. Our research team has successfully used machine learning MALDI-TOF approach to detect heterogenous vancomycin intermediate Staphylococcus aureus. We will extend this method to rapidly identify antibiotic resistant gram negative pathogens, especially MDR gram negative pathogens. The human microbiota, defined as the sum of all microbial communities living in or on the human body, plays an important role in host defenses, immunological response and pathological mechanism and sustains good health. Our previous study found microbiomes of neonates are associated with neonatal sepsis, but the case numbers are inadequate and lack of database to characterize other factors that might affect the microbiomes. We hypothesize that a microbiome-AI model can help clinicians to accurately identify severe neonatal sepsis and identify those with high risk of recurrent sepsis. After combining microbiome database, AI predictive model for neonatal sepsis and AI MALDI-TOF model, we aim to investigate whether these high-tech new methods can help clinicians in the decision making of administration of antibiotics, choices of antibiotics and further therapeutic strategies. Furthermore, we will develop microbiota database in the NICU, and investigate the various factors that may affect the microbiomes of neonates. Finally, we will apply these AI models clinically in the NICU to investigate their help in the treatment decision making and whether these AI models can save medical resources. Because MDR pathogens and severe sepsis usually occur in the recurrent episode of sepsis, we will prospectively perform the microbiota analysis from stool sample of neonates who survive the first episode of neonatal sepsis. We will investigate microbiota signature associated with severe and recurrent sepsis, which can inspire the development of similar microbiome-based therapeutic method and future decrease of neonatal sepsis.

Project ID:PC10901-0350
External Project ID:MOST108-2314-B182-064-MY3