Predictive model of ibuprofen treatment failure in very preterm infants with patent ductus arteriosus using machine learning techniques

Abstract

Background: The approach to patent ductus arteriosus (PDA) remains controversial. We aim to develop an algorithm to predict ibuprofen treatment failure (TF) using machine learning (ML) techniques. Methods: Secondary analysis of a trial of very preterm infants receiving intravenous ibuprofen to treat PDA. A predictive model on TF was developed with ML. The impact of TF on outcomes was analyzed. Results: One hundred forty-six infants were included. ML techniques showed that a logistic regression model predicted TF with an AUC 0.65. A multiple regression model found that bronchopulmonary dysplasia (BPD) was associated with TF, p = 0.03. Other neonatal outcomes did not differ between the study groups. Conclusions: It is feasible to build a predictive model of ibuprofen TF with ML that could assist clinicians during the PDA treatment decision-making process. The identification of responders prior to intervention would mitigate adverse effects in non-responders, providing them with an alternative approach.