HRV-Based Regression Analysis in Newborns With Sepsis: Forecasting BAYLEY-III Scores at 6 and 12 Months

Abstract

The early detection of neurodevelopmental disorders in newborns is of utmost importance in clinical practice. Recently, to predict the neurodevelopment scores in preterms, Artificial Intelligence (AI) methods have been proposed mainly based on Electroencephalographic (EEG) or heart rate variability (HRV) analysis. In this work, HRV measures of preterm newborns with and without Sepsis are computed and used as input features of AI regression models. The study assesses the reliability of such features in predicting BAYLEY-III scores obtained during the clinical follow-up at 6- and 12-months. Forty-eight preterms (gestational age $27.8{\pm }1.8$ weeks) were involved, 27 of which were diagnosed with Sepsis. HRV analysis was performed on ECG signals recorded at the corrected term age. BAYLEY-III score prediction was implemented, considering HRV features as input predictors of ensemble regression models. Models were validated using the Leave-One-Subject-Out (LOSO) framework. Encouraging results were achieved, with a Mean Absolute Error (MAE) < 5 points for the Sepsis group in the BAYLEY-III cognitive and language scales at 6- and 12-months. Preliminary results suggested that the autonomic nervous system development may be linked to central nervous system maturation. HRV features, and AI regression models could predict alterations that affect the correct neurodevelopment of newborns.

Keywords

• BAYLEY-III
• entropy
• HRV
• neonatal sepsis
• neurodevelopment
• preterm