Development of a Non-Invasive Procedure to Early Detect Neonatal Sepsis using HRV Monitoring and Machine Learning Algorithms

Heart rate variability (HRV) monitoring has shown to be promising to early diagnose neonatal sepsis and therefore the objective is to develop a minimally invasive and cost-effective tool, based on HRV monitoring and machine learning (ML) algorithms, to predict sepsis risk in neonates within the first 48 hours of life. Seventy-nine new-borns, with less than 48 hours of life and with a gestational age between 36 and 41 weeks, borned in the Consorci Hospital General Universitari of Valencia were enrolled after the tutor's authorization. Fifteen of them were diagnosed with sepsis. Electrocardiogram signal was monitored and recorded for 90 minutes and HRV parameters were calculated. Clinical data was extracted from the electronic medical record and sepsis was confirmed by central laboratory analyses. Supervised ML algorithms were evaluated based on sensitivity, specificity, positive predictive value, negative predictive value and area under the receiver operating characteristic curve (AUC). Significant differences were observed in the power spectrum density at very low and low frequency bands and in long-term non-linear components. The AUC revealed that Adaptive boosting was the ML model with greater sensitivity and specificity (AUC=0.94) followed by Bagged Trees (AUC=0.88) and Random Forest (AUC=0.84). In conclusion, HRV and Adaptive Boosting algorithm can be used to identify new-borns with higher risk of suffering neonatal sepsis during their first 48 hours.