AI Tool Predicts Bronchopulmonary Dysplasia Risk in Preterm Infants

Bronchopulmonary dysplasia is a chronic lung disease of prematurity marked by impaired alveolar development and prolonged oxygen dependence. It can hinder growth and neurodevelopment and can be fatal, making accurate early risk stratification a priority in neonatal intensive care. Clinicians often lack reliable tools to identify which infants will deteriorate. A new study shows a time‑series machine learning approach that forecasts risk more precisely and could enable earlier, individualized interventions.

Researchers at UC Davis Health (Sacramento, CA, USA), in collaboration with the University of Rochester Medicine (Rochester, NY, USA), developed a dynamic risk prediction model for bronchopulmonary dysplasia (BPD), a serious lung condition affecting extremely preterm infants. Published in The Journal of Pediatrics under the title “Time-Series Machine Learning for Prediction of Bronchopulmonary Dysplasia,” the study introduces a machine-learning approach that continuously updates risk estimates over time, addressing limitations of existing assessment tools in this highly vulnerable population.

The innovation applies a time‑series framework that ingests repeated clinical data points across hospitalization rather than relying on a single snapshot. It contrasts with the Neonatal Research Network’s online BPD calculator, which estimates risk from static information such as birth weight and the level of respiratory support at defined time points. By modeling how physiological and treatment variables evolve, the new method aims to reflect each infant’s changing trajectory in the neonatal intensive care unit (NICU).

Using a database assembled at University of Rochester Medicine, the team built three successive computational models to predict BPD. The most advanced model employed long short‑term memory, a recurrent neural network technique, and showed stronger predictive performance to inform care decisions. The authors also analyzed clinical and demographic variables and reported that the first temperature recorded after birth closely correlated with later BPD risk, underscoring the importance of maintaining thermal stability during and immediately after delivery.

The researchers intend to incorporate the analysis tool into electronic health records so that risk estimates are available during daily NICU rounds. Plans also include establishing a deidentified infant database at UC Davis to expand research on diverse patient populations and refine the model for broader clinical use.

"The ability to identify children who will develop severe BPD would help us target those kids earlier. Eventually, we would like to incorporate this into the electronic health record to provide more immediate insights at the point of care," said Divya Chhabra, associate professor of pediatric pulmonology at UC Davis Health and first author on the study.

"The more data we added to the model, the better it got. In the future, we hope these predictions are available to us when we are rounding in the Neonatal Intensive Care Unit (NICU). As a result, our approach to each patient's condition would change. Also, sharing this information is good for families. Having a baby in the NICU is scary, and better data can help reduce people's fears," said Chhabra.

Related Links
UC Davis Health
University of Rochester Medicine