Towards precision psychotherapy: Leveraging session-level data to predict session-by-session treatment outcomes
Abstract
Theoretical Background: Clinical support tools can potentially enhance cognitive behavioral therapy (CBT) by providing session-by-session treatment recommendations. To enable such tools, prediction models are needed that can accurately track therapy outcomes based on assessable session data. This study developed models that predict session-by-session mental health levels and identified variables that influence these predictions. Methods: We analyzed data from a monocentric, naturalistic therapy-process study of outpatients with affective, anxiety, and psychotic disorders (n=325) participating in up to 24 weekly sessions of CBT in routine care in Germany. Within a nested cross-validation approach, we used mixed-effect random forest models to predict session-by-session scores on a brief self-report measure of mental health (i.e., symptoms, mood, distress, functioning) assessed in the next session. Included predictors were baseline psychopathology, type of CBT interventions conducted in the prior session, and patient- and therapist-reported mechanisms of change (e.g., cognitive change, self-understanding, coping) assessed in the prior session. Variable importance was examined using SHAP values. Results: The mixed-effects random forest predicting next-session outcome achieved an R²=0.25 across the test folds. The most influential predictors were baseline psychopathology and prior session self-reported behavioral mechanisms of change, i.e., engagement in meaningful/positive activities and coping. Conclusion: Session-by-session prediction of mental health outcomes using routine outcome monitoring data yielded promising predictive accuracy. Self-reported mechanisms of change are particularly valuable for predicting session-by-session mental health levels. These findings may suggest that activating behavioral mechanisms of change in therapy may have a stronger impact on short-term mental health than activating cognitive mechanisms of change.
Related
- Association of a structured nutrition-centred multidisciplinary supportive-care program with treatment tolerance and clinical outcomes in patients undergoing precision radiotherapy for head and neck cancer: a prospective cohort study
- A clinician-informed framework for precision psychotherapy in psychosis
- Supporting Video-Based Psychotherapy While Asking Who Benefits and Who Remains in Treatment.
- What Predicts Psychotherapy Outcomes Across Disorders and Treatments: an Umbrella Review
- One Dance, Two Partners: Session-to-Session Therapist–Client Rating Configurations and Outcomes
BibTeX
@misc{strakeljahn2026towards,
title = {{Towards precision psychotherapy: Leveraging session-level data to predict session-by-session treatment outcomes}},
author = {Felix Strakeljahn and Tania Marie Lincoln and Niklas Jacobs and Celine Brandt and Kevin Hilbert and Katrin Bahlinger and Inga Frantz and Björn Schlier},
howpublished = {PsyArXiv},
year = {2026},
month = oct,
doi = {10.31234/osf.io/ehjc6_v2},
url = {https://osf.io/preprints/psyarxiv/ehjc6_v2/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Towards precision psychotherapy: Leveraging session-level data to predict session-by-session treatment outcomes
AU - Felix Strakeljahn
AU - Tania Marie Lincoln
AU - Niklas Jacobs
AU - Celine Brandt
AU - Kevin Hilbert
AU - Katrin Bahlinger
AU - Inga Frantz
AU - Björn Schlier
PY - 2026
DA - 2026/10/20
PB - PsyArXiv
DO - 10.31234/osf.io/ehjc6_v2
UR - https://osf.io/preprints/psyarxiv/ehjc6_v2/
AB - Theoretical Background: Clinical support tools can potentially enhance cognitive behavioral therapy (CBT) by providing session-by-session treatment recommendations. To enable such tools, prediction models are needed that can accurately track therapy outcomes based on assessable session data. This study developed models that predict session-by-session mental health levels and identified variables that influence these predictions. Methods: We analyzed data from a monocentric, naturalistic therapy-process study of outpatients with affective, anxiety, and psychotic disorders (n=325) participating in up to 24 weekly sessions of CBT in routine care in Germany. Within a nested cross-validation approach, we used mixed-effect random forest models to predict session-by-session scores on a brief self-report measure of mental health (i.e., symptoms, mood, distress, functioning) assessed in the next session. Included predictors were baseline psychopathology, type of CBT interventions conducted in the prior session, and patient- and therapist-reported mechanisms of change (e.g., cognitive change, self-understanding, coping) assessed in the prior session. Variable importance was examined using SHAP values. Results: The mixed-effects random forest predicting next-session outcome achieved an R²=0.25 across the test folds. The most influential predictors were baseline psychopathology and prior session self-reported behavioral mechanisms of change, i.e., engagement in meaningful/positive activities and coping. Conclusion: Session-by-session prediction of mental health outcomes using routine outcome monitoring data yielded promising predictive accuracy. Self-reported mechanisms of change are particularly valuable for predicting session-by-session mental health levels. These findings may suggest that activating behavioral mechanisms of change in therapy may have a stronger impact on short-term mental health than activating cognitive mechanisms of change.
ER -