Introduction
Respiratory distress syndrome (RDS) is a common cause of respiratory morbidity in premature infants due structural and functional immaturity of the lungs together with deficiency of alveolar surfactant, resulting in microatelectasis and low lung volumes 1,2. The mainstay in the management of RDS is appropriate respiratory support and surfactant replacement as needed to achieve and maintain functional residual capacity (FRC) for better gaseous exchange.
Bronchopulmonary dysplasia (BPD) is the most common complication of prematurity, leading to many long-term morbidities and adverse neurodevelopmental outcomes 3. The Neonatal Research Network collected data on over 34,000 infants born at 22–28 weeks’ gestation between 1993–2012 and demonstrated significant increases in survival among infants born at 23- 25 weeks gestational age4. These infants are at a high risk of developing BPD – with an incidence of 60–80% 5. Despite the increased use of antenatal corticosteroids, surfactant, and improved ventilation techniques, the incidence of BPD remains unchanged and may be increasing 6,7. Several treatment strategies remain in place to prevent BPD, with one such strategy being avoidance of mechanical ventilation altogether to prevent ventilator-induced lung injury and use of early noninvasive continuous positive airway pressure (CPAP). Over the past decade, trials have demonstrated the feasibility and apparent benefits of providing early CPAP, including a reduced need for intubation and surfactant therapy 6,8,9. The clinical outcomes for infants who succeed on CPAP are excellent, with low rates of BPD, mortality, intraventricular hemorrhage, retinopathy of prematurity, and lower risk for adverse neurodevelopmental sequelae in school-age 10.
However, despite best efforts, CPAP failure rates remain high and is associated with mortality and significant morbidity such as pneumothorax and BPD 5,11. Identification of these infants in attempt to predict failure will allow early proactive intervention in the course of the disease, in turn reducing the complications of RDS and CPAP failure.
Previous research studies have evaluated variables that might predict CPAP failure. Several factors to predict treatment failure have been studied including demographics, perinatal and neonatal care variables, and was shown that increased fraction of inspired oxygen (FiO2) requirement in the first few hours of life (HOL) was most consistent to predict CPAP failure 11–13. These studies focused on perinatal and neonatal variables, known to be risk factors of severe RDS.
In this study, we assess whether the use of machine learning and electronic decision support systems may prove valuable to predict CPAP failure in preterm infants with RDS and allow early proactive intervention to minimize CPAP failure burden.