Welcome to the 9th EAPS Congress Programme Scheduling
The congress will officially run on Barcelona Time (GMT+2)
To convert the congress times to your local time Click Here
MACHINE LEARNING OF RESPIRATORY OUTCOME PREDICTIONS IN PREMATURE INFANTS BORN AT < 30 WEEKS GESTATION BASED ON DEVELOPMENTAL ORIGINS OF BRONCHOPULMONARY DYSPLASIA
- Srinandini S. Rao (United States of America)
Abstract
Background and Aims
Bronchopulmonary dysplasia (BPD) remains a significant challenge. As a multifactorial disease, BPD has been shown to be associated with various maternal and antenatal factors, which are then reflected as the severity of illness in the preterm infants’ immediate postnatal life. Current understanding of BPD pathophysiology points towards a developmental origin of BPD, which may be proportionate to the degree of irreversible insult during fetal lung development. We developed random forest prediction models based on this concept by combining perinatal and respiratory support data in the first 14 days of life.
Methods
This is a single-center retrospective study from 2013 to 2019 with subjects born at less than 30 weeks’ gestation randomly split into training (80%) and testing (20%) datasets. Perinatal features and respiratory mode at each 24-hour interval for the first 14 intervals were extracted from electronic medical records. A random forest (RF) algorithm was used to train perinatal and respiratory data separately, followed by developing a final ensemble model. Model performance was assessed by receiver’s operating characteristics (ROC).
Results
Table 1 shows demographic summarization. Figure 1 shows model performance for each RF model developed using perinatal data and respiratory data. Figure 2 demonstrates the final ensemble models with ROC area under the curve between 0.87-0.91 for the training dataset, and between 0.83-0.867 for the testing dataset, respectively.
Table 1
Figures 1
Figure 2
Conclusions
Model performance was adequate, with perinatal data playing a significant role, reaffirming the developmental origins of BPD. This model may be further developed into a prediction tool for BPD.