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

The aim of this proof-of-concept study is to develop a predictive model based on deep learning algorithms to predict working alliances after the first therapeutic session and to provide a basis for clinical decisions.

Using a sample of 325 patients and 32 psychotherapists from three university counseling centers, a deep learning algorithm known as fully connected neural networks (FCNNs) was adopted to construct data-driven predictive models. The performance differences between the model including only patient indicators and the model including both patient and therapist indicators were compared. The optimal model was further tested in a general hospital sample of 85 patients and 8 therapists.

The model incorporating both patient indicators and therapist-level indicators (R²: 0.30 ± 0.02) performed better than the model incorporating only patient indicators (R²: 0.11 ± 0.02). The performance of this model decreased when being transferred to the independent general hospital sample, but still retained some predictive value (R² = 0.11).

This study showed that the inclusion of therapist-level indicators can improve the performance of a predictive model in predicting working alliances. This model could assist clinical decisions on choosing psychotherapists for patients and may also initiate new possibilities for future research.