A roadmap for applying machine learning when working with privacy-sensitive data: predicting non-response to...

Area of Science:

• Psychiatry
• Computer Science
• Data Science

Background:

• Machine learning (ML) offers potential for predicting patient outcomes in psychiatric disorders.
• Patient data privacy is a significant challenge in developing predictive models.

Purpose of the Study:

• To develop and validate a machine learning model for predicting treatment response in patients with eating disorders (EDs).
• To demonstrate the feasibility of using ML for outcome prediction without compromising patient privacy.

Main Methods:

• Applied Random Forest (RF) and least absolute shrinkage and selection operator (LASSO) algorithms to routine outcome monitoring data.
• Utilized data from 593 patients with EDs, collected at baseline, 3 months, and 6 months.
• Predicted absence of reliable improvement at 12 months post-treatment.

Main Results:

• The RF model trained on baseline and 3-month data reduced prediction errors by 31.3% compared to chance.
• Adding 6-month follow-up data yielded only minor improvements in predictive accuracy.

Conclusions:

• A validated ML model can assist clinicians and researchers in predicting treatment response for ED patients.
• ML provides a viable method for creating accurate prediction models for psychiatric disorders like EDs while maintaining privacy.