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

Area of Science:

• Psychiatry
• Clinical Psychology
• Digital Health

Background:

• Evidence-based treatments for eating disorders show variable response rates.
• Personalized treatment selection may enhance therapeutic efficiency and outcomes.
• Digital interventions offer scalable options for eating disorder care.

Purpose of the Study:

• To evaluate the utility of the Personalized Advantage Index (PAI) for matching individuals to either a broad or focused digital intervention for eating disorder symptoms.
• To assess if PAI-guided treatment selection improves post-intervention outcomes compared to non-guided treatment.

Main Methods:

• Utilized data from a randomized non-inferiority trial (n=214) comparing broad and focused digital interventions.
• Trained machine learning models (elastic net, random forest) using 40 baseline predictors to forecast symptom severity for each intervention.
• Calculated the PAI to determine the predicted optimal treatment for each participant.

Main Results:

• Elastic net models showed slightly better prediction accuracy (R²=0.29) than random forest (R²=0.26).
• No significant differences in post-intervention outcomes were found between PAI-indicated and non-indicated treatment groups.
• This lack of difference persisted in both the full sample and a subsample with larger predicted differential treatment effects.

Conclusions:

• PAI-based treatment matching did not enhance outcomes for digital eating disorder interventions in this study.
• The PAI approach may demonstrate greater utility with different treatment modalities or delivery formats.
• Further investigation is needed to explore the PAI's potential in contexts allowing for more pronounced differential treatment effects.