Developing a Multivariable Prediction Model to Identify a High-Volume Patent Ductus Arteriosus in Neonates Born Extremely Preterm.

Abstract

To create a clinical model to predict presence of a high-volume, hemodynamically significant patent ductus arteriosus (PDA) in premature infants. A single-center retrospective cohort analysis was conducted in infants < 27 weeks gestational age who received targeted neonatal echocardiography (TNE) screening for hemodynamically significant PDA. TNEs were categorized as high-volume shunt (score ≥ 9) or no PDA/low-volume shunt (score ≤ 2) using the modified Iowa PDA score. Continuous vital sign data were evaluated in 5-minute increments for the hour preceding each TNE. A Fourier transform of the arterial blood pressure waveform was used to extract features quantifying the waveform shape. Predictor variables were used to create a logistic regression model to classify each 5-minute window as high-volume PDA or no PDA/low-volume shunt. The model included 173 TNEs (37 high-volume shunt, 136 no PDA/low-volume shunt) from 90 patients. Mean airway pressure was higher in the high-volume shunt group (P < .01), but other variables were similar. Once the best model was identified, further training 1000 times resulted in a mean testing area under the curve of 0.8891 (95% CI 0.8858, 0.8924). Identification of a high-volume PDA shunt is possible with high accuracy using logistic regression modeling. Further work should focus on identification of moderate-volume shunts and correlation with clinical outcomes.