Researchers Demonstrate Machine Learning Approach to NEC Prediction

A recent publication in BMC Bioinformatics describes a machine learning system that could potentially be used to predict NEC before the disease advances. The technology, developed by researchers at the University of Pittsburgh and Columbia University, uses a neural network to determine a premature infant’s risk of NEC from the stool microbiome and basic clinical information.

One of the major problems with NEC is that we don’t have any reliable way of distinguishing infants who are at high risk from those with lower risk,” said Dr. Thomas Hooven, the study’s senior author who works as a neonatologist and researcher in the University of Pittsburgh Department of Pediatrics. “With this system, we hope to change that by calculating a baby’s risk in real-time using nothing more than some stools collected from otherwise discarded diapers.”

The other members of the study team, Dr. Ansaf Salleb-Aouissi and graduate student Yun Chao Lin, are in Columbia University’s Department of Computer Science. They engineered the software behind the prediction tool, which uses the same kind of technology that allows computers to identify faces or translate from one language to another.

To test and validate their system, the team used publicly available data from two previously published studies of NEC. These data consisted of microbiome profiles from over 3,500 stool samples collected from 261 preterm babies at multiple NICUs across the United States. Using these historical samples to simulate the NEC prediction model, the team demonstrated that they could predict NEC cases an average of 8 days before diagnosis with a sensitivity of 86% and specificity of 90%.

This technology raises a new medical question: what would be the best way to care for an infant whose NEC risk was increasing? “That’s an important thing to consider,” said Dr. Hooven. “The truth is that neonatologists have never had an opportunity to think about this. NEC has always been an emergency condition that doctors and nurses had to respond to once it was already out of control.” Dr. Hooven hopes that once medical teams can calculate a baby’s NEC risk in advance, they can test simple interventions to reduce the risk and avoid a life-threatening crisis.

With the insight needed to intervene much earlier, preventing NEC in a baby with a very high risk of developing the disease may only require something as simple as giving I.V. nutrition for a few days, or a probiotic,” added Hooven.

The group presented their research at the 2022 Pediatric Academic Societies meeting. The research team is currently planning experiments that will test their NEC prediction algorithm with new patient samples from at-risk infants in Pittsburgh. They are also interested in collaborating with other clinicians and scientists studying NEC. Dr. Hooven invites members of the NEC community to contact him at thomas.hooven@chp.edu.

Subscribe to our blog & stay in touch with the NEC Society here.

Leave a Reply

%d bloggers like this: