Leveraging artificial intelligence to personalize treatment for eating disorders: A proof-of-concept study

Area of Science:

• Psychiatry and Behavioral Sciences
• Computational Psychology
• Artificial Intelligence in Healthcare

Background:

• Eating disorders (EDs) have high relapse and mortality rates, with limited treatment efficacy for many adults.
• Personalized monitoring using idiographic, data-driven methods shows promise but struggles with high-dimensional data and real-time adaptation.
• Artificial intelligence (AI) offers a novel approach to identify complex patterns in data for precise, adaptive interventions.

Purpose of the Study:

• To investigate the utility of AI-driven analysis of ecological momentary assessment (EMA) data for personalized monitoring in eating disorder treatment.
• To develop interpretable symptom phenotypes and quantify individualized treatment response using advanced machine learning techniques.

Main Methods:

• Collected EMA data (behaviors, cognitions, affect, co-occurring symptoms) from 35 participants with ED diagnoses over 14 weeks.
• Employed fuzzy encoding for EMA responses, dimensionality reduction via a self-supervised Auto-Encoder, and fuzzy c-means clustering.
• Quantified treatment response using Mahalanobis distance on successive data embeddings to track psychological shifts.

Main Results:

• The Auto-Encoder model achieved strong performance (RMSE=0.25, R²=0.8).
• Identified four distinct, interpretable symptom phenotypes anchored in eating anxiety.
• Mahalanobis distance effectively captured individualized psychological changes, with specific modules showing significant impact on eating-related anxiety.

Conclusions:

• AI-driven analysis of EMA data provides a powerful tool for responsive, personalized monitoring in eating disorder treatment.
• This approach facilitates the identification of unique symptom profiles and tracks individualized treatment progress.
• Findings support the integration of AI and EMA for adaptive, precision-based mental healthcare.