Applying Machine Learning to Predict Complex Clinical Course in Youth With Eating Disorders

Comprehensive Summary

This study aimed to review the predictive performance of machine learning (ML) to predict the clinical course of youth with eating disorders. Clinical data from 327 children treated at the Eastern Ontario Eating Disorders Program from 2018-2024 was used. Complex clinical course was defined as readmission or a trajectory deviating from the expected step-down in level of intensity. 34 intake and discharge variables were used to train 7 ML models and logistic regression. In the results, the random forest models had the best performance, achieving an AUC of 0.723, which was a lot better compared to the logistic regression. The most important predictor was found to be weight change during the treatment. In the discussion, the authors talk about the potential for the random forest model to be used for future predictions.

Outcomes and Implications

This research is important as developing more predictive models can help ease the workload on clinicians, and it can also help spot early signs of a complex clinical course that otherwise would have gone unnoticed. Such a tool could also be used to develop more personalized treatment plans for patients. In future studies, the authors recommend for the use of larger data sets and populations with a more diverse set of eating disorders as this population of patients primarily had anorexia nervosa (84%).