Study Provides New Insights on Anorexia Nervosa Improvement

Anorexia nervosa is a psychiatric illness characterized by food avoidance, fear of weight gain and a perception of being overweight despite being extremely low weight. The mortality rate for this disease is 12 times higher than any other illness facing people assigned female at birth aged 15 to 24; however, healthcare professionals struggle to treat anorexia because it is highly psychological yet manifests as physiological. For instance, some adolescents find a sense of self-worth and control in their ability to resist food — a thought pattern that requires intensive mental health treatment.

To address the gaps in clinical understanding around anorexia, Children’s Hospital Colorado child and adolescent psychiatrist Joel Stoddard, MD, joined a team led by Guido Frank, MD, formerly an eating disorder researcher and University of Colorado School of Medicine adjoint professor in child psychiatry. Together, they conducted the largest study on anorexia using machine learning.

Their goal was to predict which variables during treatment, such as age, anxiety levels or eating habits, played the most important role in patients’ long-term improvement. Dr. Stoddard, the machine-learning lead in the study, gathered data from 160 individuals (who when combined had more than 100 variables to consider) before they participated in a partial hospitalization program and six months following their treatment.

In alignment with previous understanding about anorexia, the researchers hypothesized that the most significant factors determining a patient’s success in the program would be psychological. To reduce bias — a major detriment to traditional machine learning models — Dr. Stoddard used ensemble modeling. “Ensemble modeling means instead of relying on one kind of machine learning algorithm, you combine a bunch of them using information from each algorithm to find the best predictor,” Dr. Stoddard says.

To the researchers’ surprise, each model in this diverse collection identified a single top predictor: The more patients increased their body mass index (BMI) during the partial hospitalization program, the more likely they were to have a healthier BMI six months after treatment. This showed that psychological factors might actually play a less significant role in healing from anorexia than previously believed.

“Many of the variables we looked at are cognitive because there’s these ideas about psychological processes that might improve outcomes,” Dr. Stoddard says. “The thing that came out as being the most important was weight gain during treatment which was, in turn, predicted by broad clinical decision-making on the individual treatment course."