The first temperature reading of a premature infant at UC Davis Health in Sacramento—just after birth—could now be one of the most powerful predictors of a life-altering condition. For years, bronchopulmonary dysplasia (BPD), a chronic lung disease affecting premature babies, has been difficult to anticipate early enough for precise intervention. But a breakthrough machine learning model developed by Dr. Divya Chhabra and her team is changing that reality, offering neonatal teams across the U.S. a dynamic, data-driven way to identify at-risk infants much earlier. This isn’t just predictive analytics—it’s a lifeline for fragile newborns whose underdeveloped lungs make every breath a challenge.
BPD affects thousands of premature infants annually, with potential consequences ranging from delayed growth to neurodevelopmental issues—and in some cases, it can be fatal. While the Neonatal Research Network’s existing BPD calculator has been a helpful tool, it relies on static data points, limiting its ability to adapt as a baby’s condition evolves. Chhabra’s new model, born from her time at the University of Rochester Medicine and refined at UC Davis Health, transforms this approach by using longitudinal data collected from multiple time points in the NICU. By analyzing vitals, oxygen levels, medications, birth weight, and gestational age over time, the team built a far more responsive predictor.
The most advanced version of their model uses long short-term memory (LSTM), a machine learning technique especially effective for sequential data. It outperformed earlier versions, improving prediction accuracy with each additional data input. One of the most striking findings? The infant’s very first temperature reading showed a strong correlation with BPD risk—a simple yet profound insight underscoring the critical importance of keeping newborns warm immediately after birth. "The more data we added to the model, the better it got," Chhabra said. "In the future, we hope these predictions are available to us when we are rounding in the Neonatal Intensive Care Unit. As a result, our approach to each patient's condition would change."
The vision is clear: integrate this tool directly into electronic health records so clinicians receive real-time risk assessments at the bedside. That means not only faster, personalized care but also greater reassurance for families navigating the emotional rollercoaster of the NICU. Chhabra is now working to build a de-identified infant database at UC Davis Health, mirroring the Rochester cohort, to unlock even more research opportunities in precision neonatal medicine. With every data point, the dream of individualized, proactive care for the tiniest patients grows closer to reality.