Applying ensemble machine learning models to predict individual response to a digitally delivered worry postponement...

Area of Science:

• Mental Health
• Digital Interventions
• Machine Learning in Healthcare

Background:

• Generalized anxiety disorder (GAD) is a common mental health condition often left untreated.
• Worry is a central feature of GAD, linked to adverse health outcomes, necessitating accessible treatments.
• Digital interventions offer a scalable approach to managing GAD symptoms like worry.

Purpose of the Study:

• To investigate the predictability of personalized treatment response to a digital intervention for GAD.
• To leverage pretreatment individual differences for tailoring digital interventions.
• To identify key factors influencing treatment outcomes in a digital worry postponement intervention.

Main Methods:

• 163 participants completed a six-day digital worry postponement intervention.
• Linear mixed-effect models analyzed changes in daytime and nighttime worry.
• Ensemble machine learning models, with SHAP for feature importance, predicted worry changes.

Main Results:

• Moderate prediction accuracy was achieved for daytime worry duration (r²=0.221) and nighttime worry frequency (AUC=0.72).
• Baseline worry levels and subjective health complaints were the most significant predictors.
• Predictive performance was poor for nighttime worry duration and daytime worry frequency.

Conclusions:

• Baseline characteristics can accurately predict treatment response to digital GAD interventions.
• The worry postponement intervention may be most effective for individuals with high baseline worry and low subjective health complaints.
• Further research is needed to understand daily worry dynamics and optimize personalized digital mental health solutions.