Machine learning model may help spot lung disease risk in preterm babies
Researchers reported a time-series model that improved predictions for bronchopulmonary dysplasia in premature infants.
By Tom Brennan · Health & Medicine Correspondent
3 min read
A new computational approach may help neonatal teams identify premature infants at risk of bronchopulmonary dysplasia earlier, according to a study published in The Journal of Pediatrics. UC Davis Health said the work could support more individualized care for fragile newborns whose risk is difficult to judge in the neonatal intensive care unit.
Bronchopulmonary dysplasia, or BPD, is a chronic breathing disorder seen in premature infants with underdeveloped lungs, according to UC Davis. The condition can affect growth and neurodevelopment and can be fatal, UC Davis said.
The study was led by Divya Chhabra, an associate professor of pediatric pulmonology at UC Davis Health and the paper’s first author. Chhabra said identifying children likely to develop severe BPD could help clinicians focus treatment earlier, and she said the team hopes the tool can eventually be built into electronic health records for use at the bedside.
According to UC Davis, clinicians already have access to an online BPD calculator developed through the Neonatal Research Network, a large group of clinical sites. That calculator estimates risk using patient information such as birth weight and respiratory support, but the study authors said it relies on data from single points in time rather than a fuller stream of information as a baby’s condition changes.
Chhabra and colleagues at University of Rochester Medicine, where she worked before joining UC Davis Health in September 2025, created a database of sick infants to study BPD and other health problems, according to UC Davis. The researchers gathered chart data that included vital signs, birth weights, gestational ages, medications and oxygen needs, Chhabra said.
The team used those records to build a more dynamic prediction system, according to UC Davis. Instead of basing the calculation on one snapshot of a child’s condition, the researchers analyzed information from a series of points over time.
The investigators tested three computational models, according to UC Davis. Each added level of complexity improved the results, and the third model used a machine learning method known as long short-term memory, which UC Davis said produced much stronger predictive performance for guiding care.
Chhabra said the model improved as the researchers added more data. She also said having better predictions during NICU rounds could change how clinicians respond to each patient and could help families understand what doctors are seeing, according to UC Davis.
The study also found that an infant’s first recorded temperature was closely associated with later BPD risk, according to UC Davis. Chhabra said that finding pointed to the need to keep premature babies warm during and immediately after delivery because they may not be able to maintain body temperature on their own.
UC Davis said the researchers hope the BPD analysis tool will eventually be added to electronic health records so clinicians can receive more direct guidance while caring for infants. Chhabra also said she wants to create a deidentified infant database at the UC Davis NICU, similar to the one developed at Rochester, to study a different patient population and support more precise care.
The paper, “Time-Series Machine Learning for Prediction of Bronchopulmonary Dysplasia,” was published in The Journal of Pediatrics with the DOI 10.1016/j.jpeds.2026.115003.
This story draws on original reporting from Medical Xpress.