Patients after pulmonary vein isolation (PVI), transcatheter aortic valve implantation (TAVI) or acute percutaneous coronary intervention (PCI) for myocardial infarction
Conditions
Interventions
Group 1: Intervention arm (LLM assistance): After informed consent and 1:1 randomization, a preliminary discharge letter draft is generated by an LLM pipeline (GPT-OSS-120B, Apache-2.0) operated local
Sponsors
Universitätsmedizin Göttingen
Eligibility
Sex/Gender
All
Age
18 Years to No maximum
Inclusion criteria
Inclusion criteria: - Adult patients (= 18 years), capable of consent, hospitalized at one of the cardiology wards (2022, 5021, 5022, 1025) of the University Medical Center Göttingen. - Pulmonary vein isolation (PVI), transcatheter aortic valve implantation (TAVI) or acute percutaneous coronary intervention (PCI) for myocardial infarction performed during the current inpatient stay. - Written informed consent.
Exclusion criteria
Exclusion criteria: - Minors (< 18 years). - Patients lacking capacity to consent without an available legal representative or authorized proxy. - Patients who refuse or withdraw their consent. - Patients for whom adequate informed consent cannot be obtained despite supportive measures (e.g., interpreters, relatives, digital tools).
Design outcomes
Primary
| Measure | Time frame |
|---|---|
| WHAT: Physician documentation time per discharge letter (in minutes). WHEN: For every case included in the study at the time of discharge letter preparation. HOW: Automated measurement via system time stamps in the clinical documentation system (IXserv) when opening and closing or via the processing time of the letter module. | — |
Secondary
| Measure | Time frame |
|---|---|
| 1) Formal and content-related error frequency per discharge letter WHAT: Number and severity of formal and content-related errors per letter, classified using a traffic-light system (red = clinically relevant/potentially harmful; yellow = formally/content-wise relevant; green = minor). WHEN: After finalization and signature of the respective discharge letter (downstream, without patient contact). HOW: Blinded peer review by two independent consultant-level reviewers using a standardized assessment form. 2) Numerical correctness of objectively verifiable data points WHAT: Concordance of defined numerical values between routine data and the final letter (e.g., left ventricular ejection fraction, laboratory values, medication dosages). WHEN: After letter completion, case-by-case for every included patient. HOW: Automated and/or manual comparison of values contained in the discharge letter with the routine data documented in the clinical information system (MEDIC/Meona). 3) Subjective user satisfaction and perceived work relief of medical staff WHAT: Self-reported user satisfaction, perceived work relief, and cognitive workload of the documenting physicians. WHEN: After study use (regularly during the study period and at the end of the study). HOW: Standardized, validated questionnaires (System Usability Scale, SUS; NASA Task Load Index, NASA-TLX) in pseudonymized form. 4) Administrative turnaround time WHAT: a) Time interval between medical discharge decision and final letter signature (in minutes); b) time of actual patient discharge; c) physician documentation effort in terms of separate documentation instances (e.g., ward-round notes vs. discharge letter). WHEN: For every case included in the study, during the inpatient stay and at discharge. HOW: Automated system-side recording via time stamps and logging in the clinical documentation system (IXserv/Meona). 5) Technical robustness of the LLM pipeline WHAT: Frequency of technical failures, requi | — |
Countries
Germany
Contacts
Public ContactGabriel Riedemann
Universitätsmedizin Göttinegn
Outcome results
None listed