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

Overview

This repository contains the code and resources for predicting eating disorders among adolescents using both register-based and self-reported data. The study encompasses data from over 40,000 adolescents, aiming to develop robust diagnostic and prognostic models.

Publication

Katsiferis A, Joensen A, Petersen LV, Ekstrøm CT, Olsen EM, Bhatt S, Nguyen TL, Strandberg Larsen K. Developing machine learning models of self-reported and register-based data to predict eating disorders in adolescence. npj Mental Health Research. 2025;4:65.

Key Findings

• Diagnostic model (identifying EDs by DNBC-11): AUC = 81.3 [95% CI: 78.0, 84.6]
• Prognostic model (predicting EDs by DNBC-18): AUC = 76.9 [95% CI: 74.3, 79.5]
• A simplified 10-predictor logistic regression achieved comparable performance to the full ML model
• Top predictors: sex, emotional symptoms, body dissatisfaction, peer problems, stress, conduct problems, parental BMI, and childhood BMI

Data Source

The Danish National Birth Cohort (DNBC) following 96,822 children from before birth through young adulthood.

Metric	Value
Diagnostic sample	44,357 participants
Prognostic sample	26,127 participants
Predictors evaluated	~100
Follow-up period	18 years

Interactive Risk Calculator

Try the online risk calculator: https://alkat19.github.io/ED_Pred/

This interactive tool implements a simplified logistic regression model using the top 10 predictors from our study, allowing users to estimate eating disorder risk based on factors measured around age 11.

Disclaimer: This calculator is for educational and research purposes only. It is NOT a diagnostic tool and should NOT replace professional clinical assessment. The model was developed using Danish data and may not generalize to other populations.

Repository Contents

File	Description
Pre_Processing.R	Data cleaning, transformation, and preparation
Modelling_Diagnostic_Main.R	Primary diagnostic model development
Modelling_Diagnostic_Secondary.R	Secondary (extended) diagnostic models
Modelling_Prognostic_Main.R	Primary prognostic model development
Modelling_Prognostic_Secondary.R	Secondary (extended) prognostic models
Figures.R	Generation of visualizations and plots
Tables.R	Summary tables and descriptive statistics
Risk_Calculator_App.R	Source code for the interactive Shiny calculator
docs/	Shinylive deployment files for the web-based calculator

Citation

If you use this code or the risk calculator, please cite:

@article{katsiferis2025eating,
  title={Developing machine learning models of self-reported and register-based data to predict eating disorders in adolescence},
  author={Katsiferis, Alexandros and Joensen, Andrea and Petersen, Liselotte Vogdrup and Ekstr{\o}m, Claus Thorn and Olsen, Else Marie and Bhatt, Samir and Nguyen, Tri-Long and Strandberg Larsen, Katrine},
  journal={npj Mental Health Research},
  volume={4},
  pages={65},
  year={2025},
  publisher={Nature Publishing Group}
}

Contact

For questions about the research or code: alexandros.katsiferis@sund.ku.dk

License

Please refer to the publication for data availability information.