"Developing machine learning models of self-reported and register-based data to predict eating disorders in...

Area of Science:

• Psychiatry
• Public Health
• Data Science

Background:

• Early detection and prevention of eating disorders (EDs) in adolescence are critical public health challenges.
• Adolescent EDs require effective predictive tools for timely intervention.

Purpose of the Study:

• To develop and validate diagnostic and prognostic models for predicting EDs in adolescents.
• To identify key predictors for ED risk assessment in early and late adolescence.

Main Methods:

• Utilized data from 44,357 participants in the Danish National Birth Cohort.
• Trained machine learning and logistic regression models using ~100 predictors from self-reported and registry data.
• Evaluated model performance using Area Under the receiver operating characteristic Curve (AUC) for diagnostic and prognostic tasks.

Main Results:

• Machine learning model achieved diagnostic AUC of 81.3% and prognostic AUC of 76.9%.
• Logistic regression with top 10 predictors showed comparable performance.
• Key predictors identified include sex, emotional symptoms, peer/conduct problems, stress, parental BMI, body dissatisfaction, and early BMI.

Conclusions:

• Developed models show potential for supporting clinical risk assessment in adolescent EDs.
• Models may aid in identifying individuals for low-risk preventive interventions.
• Further validation studies are necessary to assess real-world clinical effectiveness.