Ambient AI Documentation in Oncology Consultations (NVC-AI-DOC)

BACKGROUND Electronic health record (EHR) documentation imposes a substantial administrative burden on physicians and is a major contributor to professional burnout. Studies consistently show that clinicians spend up to half their working time on computer-based documentation, leaving less than 30% of the working day for direct patient contact. Ambient AI documentation systems have shown reductions in documentation time and cognitive load across multiple health systems in the United States and Europe, with AI-generated note quality approximating that of physician-authored notes as measured by a modified Physician Documentation Quality Instrument (PDQI-9). No studies have evaluated ambient AI documentation in the Lithuanian language or in the Lithuanian oncology setting.

STUDY DESIGN Prospective, single-group, single-center feasibility study conducted at the National Cancer Center (NVC), Vilnius, Lithuania. No randomization or control arm.

AI DOCUMENTATION SYSTEM The VOCALI ambient AI system is activated at the start of each eligible consultation. Before the consultation, the physician loads pseudonymized clinical data from the EHR (diagnosis, MDT decision, treatment plan) into the system. The system audio-records the consultation in real time, integrates the preloaded clinical data, and generates a structured draft of the E025 outpatient clinical note in Lithuanian. The physician reviews, edits if necessary, and approves the final document in the EHR. The system does not make autonomous clinical decisions.

PROCEDURES

• Informed consent obtained before consultation, including consent to audio recording.
• PSQ-18 patient satisfaction questionnaire completed by participants on the same day post-consultation (~5 min).
• PDQI-10 documentation quality instrument rated by two independent raters per AI-generated note.
• NASA-TLX physician workload questionnaire administered at baseline (T0), 1 month (T1), and end of enrollment (T2).
• SUS system usability scale completed by physicians at T2.
• Documentation time and draft editing extent (% characters changed) logged automatically by the system.
• Safety incidents reported by physicians within 3 working days; monthly safety review meeting held; immediate notification of the NVC Deputy Director for Medicine and the ethics committee required for level-5 (critical) incidents.

SAMPLE SIZE 250 participants will be enrolled over a 12-month period. As a feasibility study, sample size was determined by practical feasibility and literature precedent rather than formal power calculation. With n = 250, proportions can be estimated with a 95% CI of approximately ±3-4%; if no critical incident is observed, the upper 95% confidence limit is 1.2% (rule of three). Approximately 20% data loss is anticipated.

STATISTICAL ANALYSIS R ≥4.0 or SAS ≥9.4. Descriptive statistics for all outcomes. PDQI-10 mean score with 95% CI and domain-level analysis; documentation time as mean with 95% CI and median with IQR; SUS mean with 95% CI; PSQ-18 subscale means. Inter-rater reliability assessed by rWG coefficient (target ≥0.7). Critical incident rate reported with 95% CI using the Clopper-Pearson method. NASA-TLX change (T2-T0) analyzed by paired t-test or Wilcoxon signed-rank test depending on distribution. All inferential tests are descriptive in intent; results interpreted via confidence intervals.

DATA PROTECTION All participant data are pseudonymized immediately upon enrollment. Audio recordings and transcripts are stored in the VOCALI provider's secure infrastructure under a data processing agreement per GDPR Article 28; the provider has no access to identifiable participant data. Data retained for 5 years after study completion.