Clinical Fellow Boston Children's Hospital Boston, Massachusetts, United States
Background: The patent ductus arteriosus (PDA) is associated with significant morbidity and mortality in the preterm infant. Many clinicians elect for pharmacologic closure to abate this risk profile. However, patients often fail to respond to pharmacotherapy and ultimately require surgical or catheter interventions. Currently, there is no way for clinicians to predict which patients will respond to pharmacotherapy which can lead to delays in care. Objective: To develop a novel machine learning model based on the pre-treatment echocardiogram that predicts the likelihood of successful pharmacologic closure of the PDA in preterm infants. Design/Methods: The requirement for informed consent was waived by the IRBs at Boston Children’s Hospital, Beth Israel Deaconess Medical Center (BIDMC), and Brigham and Women’s Hospital (BWH). Our inclusion criteria were preterm infants ( < 37 weeks) admitted to the NICUs at BIDMC and BWH between January 2016 and December 2021 who received pharmacologic treatment for their PDA. These infants were identified from the Vermont Oxford Network database. Infants were then excluded if they received prophylactic indomethacin, had early termination of therapy, did not have an echo prior to therapy, or had congenital heart disease. This yielded a cohort of 134 infants from BIDMC and 81 infants from BWH. Pharmacologic closure was deemed successful if there was no evidence of repeat courses of pharmacotherapy, catheterized occlusion, or surgical ligation. Currently, cardiologists, blinded to success of treatment, are downloading the echos obtained just prior to initiation of treatment and are labeling the PDA-specific views. We anticipate this process will be complete within a month. Next, we will train a model on the echos of the BIDMC infants and externally validate it on the echos of the BWH infants. We will also train and internally validate a model on the combined cohort which has been randomized into training and test sets in a 70:30 split respectively. We anticipate the training and validation to take 2 months.
