Machine learning-based tool to assess risk of hemodynamically significant PDA in extremely premature infants

A patent ductus arteriosus (PDA) is associated with complications in extremely preterm infants and its assessment requires trained personnel and equipment not always available. A prediction tool might guide the urgency and clinical decision making to allocate resources to patients with the highest risk. The aim of this study was to generate a clinical tool to assess the probability of a hemodynamically significant PDA in extremely preterm infants.

An integrative review was performed, and potential risk factors were identified based on published research, pathophysiological hypotheses and clinical experience. Variables were selected based on stepwise regression with backward elimination and forward selection and then used to generate a nomogram for risk assessment in a single center retrospective study in 677 extremely premature infants.

A model comprising six variables was derived that achieved a sensitivity of 74.8% and a specificity of 53.4%, with an area under the ROC curve of 0.685 on an independent test set. This model was subsequently used for the creation of a nomogram.

This is the first study to report a machine learning-based prediction tool for the risk assessment of hemodynamically significant patent ductus arteriosus using real-world data.

We report the first machine learning-based risk assessment tool for the prediction of a hemodynamically significant PDA. Several characteristics differ significantly between patients with spontaneous closure of the arterial duct and those with PDA, but a comprehensive and accurate assessment to predict hemodynamic significance is missing. The simple pen-and-paper score enables us to allocate risk for hsPDA using an easy-to-use pen-and-paper score with high accuracy. Risk assessment at an early stage might help guide targeted surveillance, prophylactic strategies, or individualized treatment decisions in extremely preterm infants.