Machine Learning Model for Response to Internet-Delivered CBT vs Antidepressant Medication

Area of Science:

• Psychiatry
• Digital Health
• Machine Learning

Background:

• Personalized treatment selection is crucial for depression, as no single therapy is universally effective.
• Multivariable predictive models can aid in tailoring depression treatments to individual patients.

Purpose of the Study:

• To develop and validate a machine learning model for predicting response to internet-delivered cognitive behavioral therapy (iCBT).
• To assess the treatment specificity of the predictive model against antidepressant medications.

Main Methods:

• A prognostic study involving 883 participants (776 iCBT, 107 antidepressants) aged 18-70.
• Machine learning (elastic net regression) models were trained on baseline self-report and cognitive data to predict depression severity changes at 4 weeks.
• Models were tested on holdout data and retrained on single-treatment cohorts to evaluate specificity.

Main Results:

• The best model, using 27 predictors including baseline depression and treatment expectations, explained 14% of the variance in depression change (R²=14%).
• The model demonstrated good performance on holdout iCBT (R²=18.8%) and antidepressant (R²=17.9%) data.
• Training models on single-treatment groups improved prediction specificity.

Conclusions:

• Baseline self-reported data are valuable predictors of iCBT response, outperforming cognitive data.
• The developed model shows generalizability across different depression treatments, including antidepressants.
• Training predictive models on distinct treatment cohorts enhances their specificity for personalized treatment selection.