AI Ambient Scribe in Psychiatry

Mean minutes spent writing clinical notes (progress notes, initial psychiatric assessment reports, and discharge letters), extracted automatically from EHR (ORBIS) audit logs as form-level editing durations, compared pre-intervention vs. during intervention between groups.

Quality of initial psychiatric assessment reports rated with the Physician Documentation Quality Instrument, 9-item version (PDQI-9). Each of the 9 items is scored on a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree) and summed to a total score ranging from 9 (minimum) to 45 (maximum); higher scores indicate better documentation quality. Each report is independently rated by 2 of 8 experienced mental health specialists in a balanced rotating-pair design, blinded to provider identity, group, time point, and site.

Patient-rated working alliance measured with the Working Alliance Inventory-Short Revised, German patient version (WAI-SR). The 12 items are each scored on a 5-point scale (1 to 5) and summed to a total score ranging from 12 (minimum) to 60 (maximum); higher scores indicate a stronger (better) working alliance. Reported anonymously by patients after outpatient consultations. In addition to the two-sided superiority test, non-inferiority is prespecified: a lower bound of the two-sided 95% CI of the difference-in-differences estimate above -3.5 WAI-SR sum-scale points will be interpreted as evidence against a clinically relevant deterioration of the alliance.

Calendar time from the administrative creation of the initial psychiatric assessment note in the EHR to its completion, capturing documentation backlog. This outcome is separate from total active time spent writing the note.

Coverage and density of guideline-relevant psychiatric assessment content, scored by a large language model (SwissGPT) against a 71-item checklist derived from the American Psychiatric Association practice guidelines for the psychiatric evaluation of adults. Each item is scored as covered (1), partially covered (0.5), or not covered (0); report length in characters is used to compute information density.

Patient-perceived empathy measured with the Consultation and Relational Empathy (CARE) measure. The 10 items are each scored on a 5-point scale (1 = poor to 5 = excellent) and summed to a total score ranging from 10 (minimum) to 50 (maximum); higher scores indicate greater perceived empathy (better). Completed anonymously by patients alongside the WAI-SR. A study-specific German translation is used; its psychometric properties will be examined and reported.

AI-generated initial psychiatric assessment reports, exactly as produced by the ambient scribe and before any human correction, rated by the generating providers (intervention group only) with the Provider Documentation Summarization Quality Instrument (PDSQI-9), an LLM-adapted 9-attribute instrument screening for hallucinations, omissions, redundancy, and biased language. Each attribute is scored on a 5-point scale (1 to 5) and summed to a total score ranging from 9 (minimum) to 45 (maximum); higher scores indicate better quality of the AI-generated summary.

Work-related burnout of participating providers in both groups, measured with the German version of the Burnout Assessment Tool (BAT). The 23 core-symptom items (subscales: exhaustion, mental distance, cognitive impairment, emotional impairment) are each scored on a 5-point frequency scale (1 = never to 5 = always) and averaged to a core score ranging from 1.00 (minimum) to 5.00 (maximum); higher scores indicate more burnout symptoms (worse outcome). The 10 secondary-symptom items are scored and averaged on the same 1.00-5.00 metric, with higher scores likewise indicating a worse outcome.