Logistic regression models for predicting intraventricular haemorrhage in preterm infants using respiratory and blood pressure signals

Despite the decline in mortality rates for extremely preterm infants, intraventricular haemorrhage (IVH) remains a threat to their survival. In this study, we sought to explore logistic regression models for predicting IVH as they would be applied in a clinical setting, using features derived from respiratory and blood pressure signals. Calculated predictors included mean (μ) and the short- and long-term scaling exponents (α 1 , α 2 ) from detrended fluctuation analysis. The model fitted with short-term scaling exponent (α 1 ) of the beat-to-beat diastolic blood pressure (DBP) exhibited an area under receiver-operator characteristic curve (AUC) of 0.788 (0.62, 0.96), with a sensitivity of approximately 0.875 at a specificity of 0.75. Of the multivariable models explored, the highest AUC was 0.831 (0.66, 1.00), combining μ DBP with α 1 of the beat-to-beat systolic blood pressure (SBP).