Predicting first session working alliances using deep learning algorithms: A proof-of-concept study for personalized...

Area of Science:

• Psychotherapy Research
• Artificial Intelligence in Healthcare
• Clinical Decision Support

Background:

• The working alliance is crucial for psychotherapy outcomes.
• Accurate prediction of working alliances after the first session can inform clinical practice.
• Developing data-driven predictive models is an emerging area in mental health.

Purpose of the Study:

• To develop a deep learning model for predicting working alliances post-first therapy session.
• To assess the impact of patient versus combined patient and therapist indicators on predictive accuracy.
• To provide a foundation for data-informed clinical decisions in psychotherapy.

Main Methods:

• Utilized fully connected neural networks (FCNNs), a deep learning algorithm.
• Trained models on data from 325 patients and 32 therapists at university counseling centers.
• Compared models with patient indicators only versus those with both patient and therapist indicators.
• Validated the optimal model on an independent sample of 85 patients and 8 therapists.

Main Results:

• The model incorporating both patient and therapist indicators (R²: 0.30 ± 0.02) outperformed the patient-only model (R²: 0.11 ± 0.02).
• The model retained predictive value (R² = 0.11) when applied to an independent general hospital sample.
• Therapist-level indicators significantly enhanced predictive performance.

Conclusions:

• Inclusion of therapist indicators improves the accuracy of predictive models for working alliances.
• The developed model can assist in clinical decision-making for therapist-patient matching.
• This approach opens new avenues for predictive analytics in psychotherapy research.