AI-assisted Documentation
Conditions
Brief summary
Background: No randomized controlled trial evidence currently exists on the effectiveness of ambient artificial intelligence (AI)-assisted clinical documentation. Objective: To evaluate the effect of AI-assisted documentation on nurse productivity, professional experience, and documentation quality in nurse-led telehealth (telephone and chat) contacts in Finnish primary care (Wellbeing Services County of Kanta-Häme, OmaHäme). Methods: In this randomized, open-label, repeated crossover trial, approximately 64 nurses are allocated 1:1 to an ABAB or BABA sequence of four two-week periods (A = AI-assisted documentation, B = standard manual documentation) over eight weeks. The primary outcome is the number of patient contacts handled per nurse, analyzed with a generalized linear mixed-effects model for count data with nurse as a random effect; the treatment effect is expressed as an incidence rate ratio (IRR). A Monte Carlo simulation-based power analysis indicated 85% power to detect an IRR of 1.15 at a two-sided alpha of 0.05. Secondary outcomes include self-reported work-time savings, nurse experience and satisfaction, and patient satisfaction. Discrepancies between AI-generated draft notes, and final signed notes are analyzed to characterize the frequency, type, and clinical criticality of AI errors and omissions. The trial is investigator-initiated (OmaHäme, HUS Helsinki University Hospital, University of Helsinki) and funded by the Strategic Research Council (GAINS project). The technology provider (Tandem Health) supplies the technical solution and participates in study design and manuscript preparation; responsibility for the study design, data analysis, and conclusions rests with the academic study group.
Interventions
OTHERAI-assisted clinical documentation (ambient AI scribe, Tandem Health)
During AI-assisted periods, nurses use an ambient AI scribe that transcribes the patient contact and generates a draft clinical note, which the nurse reviews, edits, and approves. During control periods, nurses document contacts manually according to standard practice. All other aspects of care follow normal clinical routines.
Sponsors
Hospital District of Helsinki and Uusimaa
Tandem Health AB (technology provider)
University of Helsinki
Wellbeing Services County of Kanta-Häme (OmaHäme)
Study design
Allocation
RANDOMIZED
Intervention model
CROSSOVER
Primary purpose
HEALTH_SERVICES_RESEARCH
Masking
NONE
Intervention model description
Repeated (ABAB/BABA) crossover design. Participating nurses are randomized 1:1 to one of two sequences: ABAB or BABA, where A denotes AI-assisted documentation and B standard manual documentation. Each sequence consists of four consecutive two-week periods (10 working days per period) over eight weeks, without washout periods. Each nurse serves as their own control.
Eligibility
Sex/Gender
ALL
Healthy volunteers
Yes
Inclusion criteria
* Nurses at OmaHäme (Wellbeing Services County of Kanta-Häme) handling telephone or chat patient contacts
Exclusion criteria
* None
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Number of patient contacts handled per nurse per working day/hour. | Daily/hourly, over four 2-week periods (8 weeks). | Count of telehealth (telephone/chat) contacts handled per nurse per working day/hour during time of excess demand extracted from routine service data. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Self-reported work-time savings. | Week 8 (end of study). | Nurses' estimate of the effect of AI-assisted documentation on time spent on documentation, assessed with a study-specific end-of-study questionnaire (5-point Likert scale). |
| Nurse-reported experience and satisfaction. | Baseline (week 0) and week 8 (end of study). | Perceived workflow, efficiency, workload, and job satisfaction, assessed with study-specific baseline and end-of-study questionnaires (5-point Likert scales). |
Countries
Finland
Contacts
CONTACTVille Vartiainen, MD, PhD, MSc
Outcome results
None listed