Preregistration for "Multicenter Speech Analysis of Psychotherapeutic Treatment in a Transdiagnostic Context (SAVER)"

Background: Mental disorders impose a substantial burden of disease worldwide; However, it remains largely unclear why some psychotherapeutic processes are more successful than others and what the key predictors of success are. Language is the central medium of psychotherapeutic change and contains a wealth of information about symptoms and therapeutic mechanisms. AI-based language analysis opens up new possibilities for diagnostic characterization, the automated identification of active therapeutic elements, and the prediction of key factors influencing treatment outcomes. Objectives: (1) Prediction of categorical and dimensional diagnostics from language (H1); (2) Identification of active therapeutic elements (H2a); (3) Prediction of therapeutic process variables (H2b); (4) Robustness of prediction models against speech anonymization (H3); (5) Evaluation of LLMs in simulated therapeutic interactions (H4). Participants: N = 500 patients aged 18–65 years who are undergoing psychotherapy in German-language, primarily conducted in person and as individual therapy, at six DZPG sites. Naturalistic cohort without randomization. Study methods: A multicenter, prospective observational study with a 36-month recruitment period. Audio-recorded sessions, supplemented by standardized self-report and raters’ report instruments. Analysis using modern transformer models.