Predicting remission after internet-delivered psychotherapy in patients with depression using machine learning and...

Area of Science:

• Computational psychiatry
• Machine learning in healthcare
• Digital therapeutics

Background:

• Predicting treatment response in Major Depressive Disorder (MDD) is crucial for effective patient management.
• Internet-based Cognitive Behaviour Therapy (ICBT) is a widely used, accessible treatment for mild-to-moderate MDD.
• Identifying factors that predict remission can optimize ICBT delivery and patient outcomes.

Purpose of the Study:

• To develop and validate a supervised machine learning model for predicting MDD remission post-ICBT.
• To assess the utility of multi-modal data, including genetic information, in predicting treatment outcomes.
• To compare the performance of a random forest model against other predictive models.

Main Methods:

• Utilized a dataset of 894 genotyped adult patients with MDD treated with guided ICBT.
• Integrated demographic, clinical, process, and genetic (polygenic risk scores) data as predictors.
• Employed recursive feature elimination for predictor selection and cross-validation for internal validation; external validation against null, logit, XGBoost, and meta-ensemble models.

Main Results:

• A random forest model, using 45 selected predictors from all data types, achieved an accuracy of 0.656 (AUC 0.687) in predicting remission on unseen data.
• The model demonstrated reasonable predictive accuracy compared to a null model (P = 0.004) and slightly outperformed a logistic regression model.
• Transparency analysis confirmed the model's reliance on all predictor types for both group and individual predictions.

Conclusions:

• A novel, empirically validated multi-modal classifier can predict MDD remission status following ICBT in routine care.
• The findings suggest that integrating diverse data sources can enhance prediction accuracy for treatment outcomes.
• This multi-modal predictive approach holds potential for informing tailored treatment strategies and warrants further investigation for clinical utility.