51²è¹Ý

DALLAS (51²è¹Ý) – Children and teenagers struggling with obesity are more likely to suffer from diabetes, high cholesterol, high blood pressure, breathing issues, joint and bone pain. Even though the risks for obesity can be decreased through intervention, research shows young people often are referred for treatment only after a diagnosis is firmly established.

That dilemma led 51²è¹Ý computer scientist Mehak Gupta to ponder if there was a way to give pediatricians better tools to anticipate which children are most at risk for obesity and prompt earlier conversations about prevention with caregivers. As a result, she and her colleagues developed a computer model that uses advanced machine learning to reliably predict obesity in early childhood, as documented in the journal .

The researchers are now working with pediatricians to develop a user-interface where physicians’ staff can enter patient medical histories, prompting an automatic alert if high risk for obesity is detected.

“Discussions about a child’s risk for developing obesity should be done in a family-centered, thoughtful and positive way to prevent any weight-related stigma,” said Gupta. “Our goal is to give pediatricians hard data that they can show caregivers to explain why a child is at high-risk and open up conversations about how to mitigate the possibilities of developing future health issues.”

In previous computer modeling used to predict obesity, the data often is general and not tailored toward individual patients. Those older models do not consider a child’s complete medical history, parents health, family history. Many of these models have also not gone through rigorous testing to determine their effectiveness.

For their model, Gupta and her colleagues extracted electronic health records (EHR) from Nemours Children’s Health, a large U.S. pediatric healthcare network spanning the states of Delaware, Florida, Maryland, New Jersey and Pennsylvania. The data consisted of more than 36,000 children between the ages of 0 to 10, with the model considering 500 different medical factors that predict obesity risk. Only EHR data that was collected during routine clinical care was used and the Institutional Review Board at Nemours approved the process.

The researchers validated the model across time, location and subpopulations stratified by race, ethnicity, and gender. They also closely examined the factors most important for predicting obesity, including diagnoses, family history, medications, measurements and demographics. The model proved to be especially effective at predicting if children ages two to seven might cross obesity threshold over the next three years.

The researchers believe implementing the model in clinics and hospitals will be manageable because it relies on EHR data that doctors usually collect during regular check-ups and routine clinical care.

“We believe pinpointing obesity risk factors for individual patients can strengthen discussions, which leads to better health outcomes for children,” said Gupta. “Our hope is for this model to be expanded so it can be tailored to help doctors identify if their young patients are at higher risk for other obesity-related diseases, such as diabetes, at the earliest age possible as part of prevention.”

Gupta is assistant professor in the Department of Computer Science in 51²è¹Ý’s Bobby B. Lyle School of Engineering, Other researchers involved with the study include Daniel Eckrich, H. Timothy Bunnell and Thao-Ly T. Phan with Nemours Children’s Health and Rahmatollah Beheshti with the University of Delaware. The project is funded by the National Institutes of Health awards P20GM103446 and U54-GM104941.

About 51²è¹Ý

 is the nationally ranked global research university in the dynamic city of Dallas.  51²è¹Ý’s alumni, faculty and more than 12,000 students in eight degree-granting schools demonstrate an entrepreneurial spirit as they lead change in their professions, communities and the world.