Background: Heart rate characteristics aid early detection of late-onset sepsis (LOS), but respiratory data contain additional signatures of illness due to infection. Predictive models using machine learning and cardiorespiratory data may improve early sepsis detection.
Objective: Test the hypothesis that heart rate (HR) and oxygenation (SpO2) data contain signatures that improve continuous sepsis risk prediction over HR or demographics alone.
Methods: We analyzed cardiorespiratory data from very low birth weight (VLBW, <1500g) infants admitted to three NICUs from 2012-2021 with annotated blood cultures. We developed and externally validated four machine learning models to predict imminent LOS using features calculated every 10m: mean, standard deviation, skewness, kurtosis of HR and SpO2, and their cross-correlation. We compared feature importance, discrimination, calibration, and dynamic risk prediction across modeling methods and cohorts. We built logistic regression models of demographics and HR or SpO2 features alone for comparison with HR-SpO2 models.
Results: Data were available for 2,494 VLBW infants who had 302 LOS events. Performance, feature importance, and calibration were similar among modeling methods. SD of HR, skewness of HR, and kurtosis of SpO2 ranked as important features most consistently across models. All models had favorable external validation performance. The HR-SpO2 model performed better than models using either HR or SpO2 alone. Demographics improved the discrimination of all physiologic data models but dampened their dynamic performance.
Conclusion: Cardiorespiratory signatures detect LOS in VLBW infants at 3 NICUs. Demographics risk-stratify, but predictive modeling with both HR and SpO2 features provides the best dynamic risk prediction.