Improved dropout prediction in group cognitive behavior therapy (CBT) using classification trees

Area of Science:

• Psychiatry
• Clinical Psychology
• Data Science in Healthcare

Background:

• Psychotherapy dropout significantly reduces treatment effectiveness.
• Predicting patient dropout remains a challenge in clinical practice.
• Classification trees offer a practical approach to identify at-risk patients using intake data.

Purpose of the Study:

• To develop and test classification tree models for predicting dropout in weekly and intensive Cognitive Behavior Therapy (CBT) group programs.
• To identify key intake variables associated with patient dropout.

Main Methods:

• Collected intake data from day-patients in weekly and intensive CBT programs (2015-2019).
• Trained and tested two classification tree models to predict dropout.
• Analyzed the predictive power of various intake variables, focusing on comorbidity.

Main Results:

• Dropout rates were 21.9% for weekly CBT and 13.2% for intensive CBT.
• The number of comorbid diagnoses was the most significant predictor of dropout in both programs.
• Classification tree models achieved moderate predictive accuracy (around 62-63%) for identifying dropouts.

Conclusions:

• Comorbidity is a critical factor in assessing dropout risk for CBT patients.
• Simple classification tree models can predict dropout with moderate accuracy early in therapy.
• Intensive, condensed treatment formats may improve patient retention.