Personalized Treatment Selection for Digital Eating Disorder Interventions: A Proof-of-Concept for the Personalized...

Although a range of evidence-based treatments for eating disorders exist, treatment response varies substantially. The ability to match individuals to a treatment which they are most likely to benefit from may help improve treatment efficiency and therapeutic outcomes. The present study applies a treatment selection approach called the personalized advantage index (PAI) and evaluates its utility for matching individuals to a broad versus focused digital program for eating disorder symptoms.

Data were used from a randomized non-inferiority trial comparing the two interventions (n = 214). Machine learning models (elastic net and random forest) were trained to predict post-intervention symptom severity for a broad and focused digital intervention using 40 self-reported baseline predictors. The PAI was calculated to identify the predicted optimal treatment for each participant.

Elastic net performed marginally better than the random forest at predicting intervention outcomes (Elastic net R² = 0.29; Random Forest R² = 0.26). Independent samples t-tests indicated no significant differences in post-intervention outcomes between participants who received their PAI-indicated treatment and those who did not in both the full sample and a subsample with larger predicted differential responses.

PAI-based treatment matching did not improve post-treatment outcomes in this context. Greater utility of the PAI approach may emerge when applied to different treatment orientations or delivery formats, enabling greater opportunity for differential effects to emerge.