(PS2-B38) The Predicting Model of Dropout in Cognitive-behavioral Therapy for Panic Disorder: A Machine Learning Approach

Background: Cognitive-behavioral therapy (CBT) is empirically supported treatment for panic disorder (PD). Attrition is an important issue in the practice and research of CBT, however, there are no consistent predictors of dropout during CBT for PD. Machine learning is a novel and proper approach to analyze complex data and has been widely used in several fields. However, machine learning approach has rarely tested in research of CBT.

Objective: The present study aimed to examine the predictors of dropout in CBT for PD using machine learning approach from our naturalistic and longitudinal clinical activities.

Methods: Two hundred eight patients with PD were treated with manualized group CBT from 2001 to 2017 in our institute. The program consisted of ten weekly sessions. Each session was scheduled to last 120 minutes. The program included (i) psychoeducation concerning anxiety and PD cognitive-behavioral model, (ii) breath retraining, (iii) cognitive restructuring to identify and dispute maladaptive thoughts, and (iv) in vivo graded exposures to anxiety-provoking situations.

Using this data set, our study implemented machine learning approach to test if it is possible to predict attrition from CBT for PD.

The independent variables were five personality dimensions in NEO Five Factor Index (neuroticism, extraversion, openness, agreeableness, conscientiousness), depression subscale of Symptom Checklist-90 Revised, age, baseline score of Panic Disorder Severity Scale (PDSS). The dependent variable was the presence or absence of dropout.

The prediction analysis was carried out using four machine learning approaches; Logistic Regression, Decision Tree, Random Forest and Light Gradient Boosting Machine.

The study’s protocol was approved by the Ethics Committee of Nagoya City University Graduate School of Humanities and Social Sciences.

Results: Random Forest and Light Gradient Boosting Machine identified more dropout during CBT for PD than other algorithms. These approaches also showed the accuracy of prediction of attrition was about 90%.

Conclusion: The machine learning algorithms could detect many of participants as dropout after CBT for PD. For the purpose of clinical decision-making, we could use this machine learning method. Furthermore, this study was conducted as a naturalistic study in a routine clinical setting. Our results in machine learning approach could be generalized to regular clinical settings.