Machine-Learning Applications in Eating-Disorder-Outcome Prediction: A Systematic Scoping Review

Abstract

Eating disorders (EDs) are complex and debilitating conditions. Prior efforts to predict outcomes (onset, prognosis, treatment response) have yielded inconsistent findings. Machine-learning (ML) techniques have shown promise to improve outcome prediction, but a systematic literature synthesis is missing. We conducted a systematic scoping review to summarize extant literature on ML applications in ED-outcome-prediction research, identifying 75 studies. ML has mostly been used to predict ED diagnostic status ( k = 45); other studies have predicted escalation of ED risk and symptoms ( k = 13), treatment outcomes ( k = 12), and ED onset ( k = 6). Decision trees, random forest, and support-vector machines were the most common models used. Although many studies reported moderate to high predictive performance, the benefits of ML over traditional statistical techniques remains unclear in light of inconsistent findings. We make several recommendations for future research (i.e., integrating multiple data types, external validation) to encourage continued progress in this developing field.