Related Experiment Video
Updated: Aug 9, 2026
Implementation of a Real-Time Psychosis Risk Detection and Alerting System Based on Electronic Health Records using CogStack
Published on: May 15, 2020
Prediction of Clinically Meaningful Improvement After Internet-Delivered Cognitive Behavioral Therapy for Depression
Olly Kravchenko1, Matthew Halvorsen1,2, Julia Bäckman1
1Centre for Psychiatry Research, Department of Clinical Neuroscience, Karolinska Institutet, & Stockholm Health Care Services, M48, Karolinska Universitetssjukhuset Huddinge, Region Stockholm, Stockholm, 14186, Sweden, 46 709604983.
Predicting treatment success for internet-delivered cognitive behavioral therapy (ICBT) is crucial. Machine learning models integrating clinical and sociodemographic data can identify patients likely to benefit from ICBT for depression and anxiety.
Area of Science:
- Psychiatry
- Digital Health
- Machine Learning
Background:
- Up to 50% of patients do not achieve significant symptom reduction with internet-delivered cognitive behavioral therapy (ICBT).
- Identifying non-responders before treatment initiation is key for effective treatment planning.
- Current prediction methods for ICBT outcomes require enhancement.
Purpose of the Study:
- To improve baseline prediction of clinically meaningful improvement in patients undergoing ICBT for common psychiatric disorders.
- To inform treatment planning and patient stratification at intake.
- To develop and evaluate multimodal predictive models.
Main Methods:
- Developed multimodal predictive models using clinical, sociodemographic, and genetic data from 1790 patients.
- Applied machine learning algorithms including random forest (RF), logistic regression, and ensemble methods.
- Validated models using nested cross-validation and a 20% holdout test set, measuring performance via AUC.
Main Results:
- Phenotypic models achieved comparable performance (AUCtest 0.732-0.749), with RF showing the best discrimination (AUCtest 0.749).
- RF and ensemble models incorporating register data outperformed the benchmark screening model (P=.04 to P=.02).
- Polygenic scores did not add independent predictive value in this study.
Conclusions:
- Baseline prognostic prediction of clinically meaningful improvement after ICBT is feasible.
- These findings support the prospective validation of model-informed risk stratification for ICBT.
- Multimodal data and machine learning enhance prediction of treatment response.