acePrediction of bronchopulmonary dysplasia seven days after birth using respiratory and oxygenation timeseries with machine learning
Accurate prediction of bronchopulmonary dysplasia (BPD) development would allow targeted early treatment. This study aims to develop a machine learning (ML) model incorporating respiratory and oxygenation timeseries to predict BPD development within 1 week after birth. Data was collected retrospectively from a neonatal intensive care (2009-2015). Readily available clinical data and respiratory and oxygenation timeseries (mode of respiratory support, FiO2, SpO2) were gathered. Descriptive features were extracted…
Recording noise, speech and music inside an incubator by DrTrumpet. CC BY-SA 4.0 · https://creativecommons.org/licenses/by-sa/4.0
In a retrospective study of 513 preterm infants in a neonatal intensive care unit between 2009-2015, investigators developed machine learning models to predict bronchopulmonary dysplasia within one week after birth using clinical data and respiratory and oxygenation timeseries. The best model combined clinical data with advanced timeseries features processed by a long short-term memory network, achieving AUC 0.83.
Early accurate BPD prediction could enable targeted early treatment for at-risk infants, and the study indicates that temporal patterns in FiO2, SpO2 and support mode add value beyond static clinical variables. As of the July 2026 publication date, results reflect retrospective performance, not prospective deployment, and generalizability beyond the single-center historical cohort remains to be tested.
- Study used retrospective data from 2009-2015 including 513 preterm infants, 102 of whom developed BPD at 36 weeks postmenstrual age.
- Timeseries inputs included mode of respiratory support, FiO2, SpO2, processed via descriptive features and via compression with LSTM combined with clinical data network.
- Comparison models included logistic regression, support vector machine and XGBoost trained on clinical and descriptive data.
- Best clinical-only model was logistic regression with AUC 0.80, while clinical plus descriptive timeseries reached AUC 0.81.
Machine learning models that incorporate compressed respiratory and oxygenation timeseries improved early prediction of BPD at 7 days after birth compared to clinical-data-only models.
The rundown
Researchers gathered readily available clinical data plus respiratory and oxygenation timeseries (mode of respiratory support, FiO2, SpO2) and extracted descriptive features. They compressed timeseries for an LSTM neural network combined with a clinical-data network, and compared against LR, SVM and XGBoost models.
Among 513 infants, 19.8% developed BPD at 36 weeks postmenstrual age. The advanced timeseries model achieved AUC 0.83 (95% CI 0.81-0.84), significantly outperforming clinical-only LR at AUC 0.80 and clinical plus descriptive features at AUC 0.81, suggesting added predictive value from temporal patterns.
Sources
- Peer-reviewedPediatric Research2026-07-17
How should this claim be treated?
The debate