19/11/2025
Ottawa, Ontario — Wednesday November 19, 2025
Predicting which young people with eating disorders will face a more complex treatment journey is a major challenge – and one that is seen often, with one in every three adolescents experiencing a relapse after initial treatment.
A new study led by the CHEO Research Institute’s Stephanie Ryall, Nicole Obeid, and Khaled El Emam is the first ever to apply machine learning to predict clinical eating disorder outcomes in youth.
“There has been limited progress to date in personalized treatment and predicting clinical course outcomes for young people with eating disorders.” said PhD student and study first author Stephanie Ryall. “Our study shows that machine learning can help identify early on which individuals are at risk of a complex clinical course, potentially enabling targeted intensive treatment and monitoring to improve outcomes.”
Using data from over 300 adolescents treated at CHEO’s Eating Disorders Program, researchers compared seven machine learning models trained on clinical characteristics from the first treatment episode to traditional statistical methods to predict which youth were likely to need readmission or receive more intensive care. They found that the best-performing model, called random forest, was significantly more accurate than standard logistic regression in predicting future complex clinical needs – and the model performed even better when more clinical characteristic data points were added to training.
“We’re excited to continue refining the predictive model by integrating biological and psychological data to advance precision medicine in this field.” said Stephanie Ryall.
This research shows the potential of machine learning predictive models to support personalized care through optimizing early intervention for youth with eating disorders. Future research will expand the sample size and look to validate these findings for use as a tool for clinicians.