Predicting early dropout in online versus face-to-face guided self-help: A machine learning approach

Area of Science:

• Psychiatry and Mental Health
• Health Informatics
• Machine Learning in Healthcare

Background:

• Early dropout from psychological interventions negatively impacts treatment effectiveness and outcomes.
• Identifying factors to improve patient engagement in mental health services is crucial.
• Brief psychological interventions, including guided self-help, are widely adopted.

Purpose of the Study:

• To investigate if matching patients to face-to-face or computerized low-intensity psychological interventions improves attendance and depression treatment outcomes.
• To evaluate the effectiveness of a machine learning model in predicting optimal treatment modality for patients.

Main Methods:

• Analysis of archival clinical records for 85,664 patients accessing guided self-help (GSH).
• Supervised machine learning applied to a training sample (n=55,529) and cross-validated on a test sample (n=30,135).
• Primary outcome: early dropout (≤3 sessions); clinical utility assessed via logistic regression and chi-square tests.

Main Results:

• Patients receiving their model-indicated treatment modality were 12% more likely to receive adequate treatment (OR=1.12, p=.02).
• This effect was stronger in a balanced subsample (OR=2.10, p<.001), indicating better attendance.
• Improved treatment outcomes were observed when patients were matched to their indicated treatment modality.

Conclusions:

• Machine learning algorithms can effectively predict optimal treatment modalities for patients.
• Optimizing patient-treatment matching through machine learning can enhance engagement and improve outcomes in psychological interventions.
• This approach holds promise for improving the delivery and effectiveness of mental health services.