A recent randomized trial tested whether a personalized approach to assigning digital treatments could help people with eating disorder symptoms. The study included 214 participants and compared two ways of assigning treatment: one using a machine learning tool called the Personalized Advantage Index (PAI) to match each person to either a broad or a focused digital program, and the other without this matching.
The researchers found that the PAI-based matching did not lead to better post-intervention outcomes compared to non-matched assignment. In other words, people who received their PAI-indicated treatment did not show significantly different symptom improvement than those who did not. The study also examined the accuracy of two machine learning models in predicting outcomes: the Elastic net model performed slightly better (R = 0.29) than the Random forest model (R = 0.26), but both were modest.
This is a randomized trial, which is a strong design, but the results are clear: in this context, using the PAI to choose between broad and focused digital programs did not improve outcomes. The study did not report any safety concerns, but it also did not provide details on side effects or dropouts.
The main limitation is that the PAI did not show utility for this specific choice. It is possible that the tool might work better in other treatment settings or with different types of programs. For now, these findings suggest that personalized matching may not be the key to improving digital treatment outcomes for eating disorders.
