Vital sign-based detection of sepsis in neonates using machine learning

Area of Science:

• Neonatal Medicine
• Computational Biology
• Biomedical Engineering

Background:

• Sepsis is a major cause of illness and death in newborns.
• Diagnosing neonatal sepsis early is challenging due to vague symptoms.

Purpose of the Study:

• To assess the predictive capability of machine learning using non-invasive monitoring and demographic data for neonatal sepsis detection.
• To improve early identification of sepsis in neonates.

Main Methods:

• A single-center study analyzed data from 325 infants.
• Time-domain features from heart rate, respiratory rate, and oxygen saturation were extracted.
• A Naïve Bayes algorithm was employed for sepsis prediction up to 24 hours prior to clinical suspicion.

Main Results:

• The algorithm achieved an area under the ROC curve of 0.82 for sepsis prediction.
• Combining multiple vital signs enhanced predictive performance over heart rate data alone.
• Sepsis risk increased 150-fold 10 hours before clinical suspicion.

Conclusions:

• Machine learning algorithms utilizing non-invasive data show significant predictive value for neonatal sepsis.
• These methods offer a promising approach to personalize care and decrease neonatal morbidity and mortality.