Protocol for an EHR-embedded pragmatic randomized control trial of Ambient AI to Reduce Nursing Staff Documentation Time
Open accessThe DOI matches its Crossref record; 11 of the 28 references it cites were found in a registry or an index; 11 of the 21 sources its claims cite were checked for support.
Abstract
Background: Documentation burden significantly impacts nursing workload and well-being, with nurses spending an estimated 20-40% of their time on documentation. Ambient AI technologies offer potential to reduce documentation time by mapping real-time nurse-patient conversations to structured EHR data entries with human-in-the-loop verification. Methods: This protocol describes a pragmatic, EHR-embedded randomized controlled trial evaluating the effectiveness of an Ambient AI tool in reducing nursing documentation time across three inpatient medical/surgical units. The study employs a closed-cohort, stepped-wedge, unit-randomized design, integrating the intervention into routine clinical workflows. The primary outcome is documentation time per shift hour, derived from EHR audit logs. Secondary outcomes include documentation burden, professional well-being, and perceived usability. Results: The trial is being implemented within a shared governance model that integrates executive oversight, operational feasibility, and research rigor. Multidisciplinary workgroups coordinate technical integration, user experience, and analytics, ensuring alignment between operational priorities and pragmatic trial objectives. Early implementation has highlighted the importance of adapting training and analytic strategies to address differential intervention exposure, as well as the need for rapid operational responses to late-emerging technical issues. Discussion: This protocol demonstrates the feasibility of embedding a randomized pragmatic trial within a health system-led operational deployment of Ambient AI for inpatient nursing documentation, highlighting the necessity of adapting existing outpatient provider-focused AI implementation strategies for inpatient nursing.
Overview
CoChat Summary
TL;DR
This protocol describes a pragmatic, EHR-embedded, closed-cohort stepped-wedge randomized trial of the Abridge Ambient AI tool on three inpatient medical/surgical units, testing whether AI-drafted flowsheet documentation reduces nursing documentation time. It reports the governance, design, and early implementation experience rather than trial outcomes, which are not yet available.
Key findings
- No effectiveness results are reported; at publication, nursing champions on the first unit had begun using the tool, and 16 participants were enrolled in survey data collection, which the authors describe as demonstrating "challenges of recruiting direct care nursing staff in study participation."
- The study is powered to detect a 24-second reduction in flowsheet time per shift hour (84% power, assuming SD of 3 minutes); if SD is as large as 4 minutes, power is 82% to detect a 31.2-second reduction.
- Early implementation surfaced operational issues: unit-wide broadcast announcements intermittently interrupted recordings, and a vendor-initiated model update during rollout led to a recommended delay and adjustment of the study calendar.
- Training required more intensive support than outpatient provider playbooks assume, because nurses were less likely than physicians/APPs to have experience verbalizing assessment findings; nurse champions gaining early access created six implementation subwaves within three randomized parent waves, requiring analytic adjustment for differential exposure.
Methods
- Design: pragmatic randomized controlled trial with a closed-cohort, stepped-wedge, unit-randomized design; three medical/surgical units (27–28 beds each; staffing 4–6 patients/RN, 8–10 patients/NA) at a Midwest academic medical center, randomized 1:1:1 to staggered start times after a 4-week standard-of-care period, transitioning at 3-week intervals over 22 weeks. Blinding is not feasible.
- Intervention and outcomes: Abridge™ Ambient AI (ASR + LLMs) embedded in Epic® via the Rover mobile app, generating discrete flowsheet values requiring nurse confirmation before filing. Primary outcome is active time spent in flowsheets per shift hour from EHR audit logs; secondary outcomes include flowsheet time per patient per shift, clicks/taps, overtime charting, Mini-Z 3.0 well-being/EMR-stress subscales, and System Usability Scale scores.
- Analysis: intent-to-treat, with linear mixed effects models at both individual and unit levels (random effects for unit, staff member, and unit-by-period interaction; adjustment for secular trend and nurse champion status); secondary survey outcomes dichotomized and controlled via Benjamini-Hochberg (two-sided type I error 0.1); sensitivity analyses under MCAR, MAR, and MNAR. Approximately 127 RNs and 72 nursing assistants potentially meet eligibility criteria.
Limitations
- Stated by the authors: the primary flowsheet-time measure captures only one component of documentation workload, does not differentiate documentation from data review (which may temporarily increase during early adoption), and does not assess documentation quality or total documentation volume; null or modest effects may reflect a learning curve rather than lack of impact.
- Stated by the authors: recruitment of direct-care nursing staff into study participation has been challenging, and late-emerging technical issues required rapid operational responses.
- Evident from the text: this is a protocol with no outcome data yet, so all effectiveness claims remain untested; the trial is single-health-system and limited to three medical/surgical units, and the vendor's mid-rollout model update means the intervention itself changed during the study period.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
11 Valid · 0 Not valid · 8 Not checked · 61 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Nurses consistently perceive their documentation workload as high [4]”
What the sources say
- Partly supportedcites [4] 4. De Groot, K., et al., Nursing documentation and its relationship with perceived nursing workload: a mixed-methods study among community nurses. BMC Nurs, 2022. 21 (1): p. 34. sourceExperimental — accuracy not yet measured
The Conclusions report that a majority of community nurses regularly perceived a high workload due to documentation. The claim broadens this finding to nurses generally and says they perceive it consistently, whereas the study reports responses from community nurses and not all respondents reported high workload regularly. (LLM, conf=0.87)
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Traditional efficacy trials are often poorly suited to evaluating clinical AI tools once deployed in practice, particularly those intended to operate continuously within complex sociotechnical environments [21].”
What the sources say
- Partly supportedcites [21] 21. Capili, B. and J.K. Anastasi, Pragmatic Clinical Trials: A Study Design for Real-World Evidence. Am J Nurs, 2025. 125 (2): p. 56-58. sourceExperimental — accuracy not yet measured
The Introduction says traditional RCTs test efficacy under controlled conditions but often fall short in informing everyday clinical practice. That supports the general concern about evaluating interventions after deployment, but the paper does not address clinical AI tools, continuous operation, or complex sociotechnical environments. (LLM, conf=0.88)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Nursing documentation is a time-intensive activity that occupies up to 40% of a nursing shift [1].”
- Also stated in: § Abstract · p. 2 · hedged
“with nurses spending an estimated 20-40% of their time on documentation”
What the sources say
- Partly supportedcites [1] 1. Momenipur, A. and P.R. Pennathur, Balancing Documentation and Direct Patient Care Activities: A Study of a Mature Electronic Health Record System. Int J Ind Ergon, 2019. 72 : p. 338-346. sourceExperimental — accuracy not yet measured
The Results (Time distributions between documentation, patient care and other activities; page unavailable) say documentation took nearly 40% of time across all healthcare workers and was a major activity for nurses. They do not give a 20–40% estimate for nurses specifically; the 40% figure combines nurses, physicians, and residents. (LLM, conf=0.86)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § Abstract · p. 2 · hedged
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Ambient Listening technology, or Ambient AI, may reduce documentation burden by leveraging automated speech recognition (ASR) and large language models (LLMs) to transform real-time clinical conversations into drafted EHR documentation records [11].”
What the sources say
- Partly supportedcites [11] 11. Bracken, A., et al., Artificial Intelligence (AI) - Powered Documentation Systems in Healthcare: A Systematic Review. J Med Syst, 2025. 49 (1): p. 28. sourceExperimental — accuracy not yet measured
The Introduction says ambient AI scribes translate spoken conversations into written documentation in real time, and the Efficiency section reports documentation time savings. The paper describes ambient AI as a hybrid of generative AI and dictation, but does not specifically say it uses ASR and LLMs together or produces drafted EHR records. (LLM, conf=0.85)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“this burden has been associated with emotional exhaustion and depersonalization [5]”
What the sources say
- Supportedcites [5] 5. Gesner, E., et al., Documentation Burden in Nursing and Its Role in Clinician Burnout Syndrome. Appl Clin Inform, 2022. 13 (5): p. 983-990. source
The Discussion reports statistically significant correlations between documentation burden and two burnout domains: emotional exhaustion and depersonalization. This directly supports the cited association in nurses. (LLM, conf=0.94)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
16 more claims · 2 with their cited sources checked
Create a free account to see all 80 claims and to verify the 3 references still open.
- It may state two separate claims as one. 23
- Probably not a substantial claim. 53
- The quote only partly states this claim. 2
Citations
Cites
Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing
Benjamini, Y., & Hochberg, Y. (1995) · Journal of the Royal Statistical Society Series B: Statistical Methodology · doi:10.1111/j.2517-6161.1995.tb02031.x · Cited by 109,482
Harris, P. A., Taylor, R., Thielke, R., Payne, J., Gonzalez, N., & Conde, J. G. (2009) · Journal of Biomedical Informatics · doi:10.1016/j.jbi.2008.08.010 · Cited by 51,384
Cruz Rivera, S., Liu, X., Chan, A.-W., Denniston, A. K., Calvert, M. J., The SPIRIT-AI and CONSORT-AI Working Group, SPIRIT-AI and CONSORT-AI Steering Group, Darzi, A., Holmes, C., Yau, C., et al. (2020) · Nature Medicine · doi:10.1038/s41591-020-1037-7 · Cited by 933
Association of Electronic Health Record Design and Use Factors With Clinician Stress and Burnout
Kroth, P. J., Morioka-Douglas, N., Veres, S., Babbott, S., Poplau, S., Qeadan, F., Parshall, C., Corrigan, K., & Linzer, M. (2019) · JAMA Network Open · doi:10.1001/jamanetworkopen.2019.9609 · Cited by 333
Alobayli, F. Y., O’Connor, S., Holloway, A. S., & Cresswell, K. M. (2023) · Digital Health · doi:10.1177/20552076231220241 · Cited by 134
Documentation burden in nursing and its role in clinician burnout syndrome
Gesner, E., Dykes, P. C., Zhang, L., & Gazarian, P. (2022) · Applied Clinical Informatics · doi:10.1055/s-0042-1757157 · Cited by 118
Artificial Intelligence (AI) – Powered Documentation Systems in Healthcare: A Systematic Review
Bracken, A., Reilly, C., Feeley, A., Sheehan, E., Merghani, K., & Feeley, I. (2025) · Journal of Medical Systems · doi:10.1007/s10916-025-02157-4 · Cited by 113
The Mini Z Worklife and Burnout Reduction Instrument: Psychometrics and Clinical Implications
Linzer, M., McLoughlin, C., Poplau, S., Goelz, E., Brown, R., Sinsky, C. A., & for the AMA-Hennepin Health System (HHS) burnout reduction writing team (2022) · Journal of General Internal Medicine · doi:10.1007/s11606-021-07278-3 · Cited by 108
De Groot, K., De Veer, A. J. E., Munster, A. M., Francke, A. L., & Paans, W. (2022) · BMC Nursing · doi:10.1186/s12912-022-00811-7 · Cited by 105
Lindsay, M. R., & Lytle, K. (2022) · Applied Clinical Informatics · doi:10.1055/a-1868-6431 · Cited by 69
Metadata
- Authors
- Wieben, A., Pfaff, J., Baumann, M. R., Resnik, F., Brzozowski, S., Langer, C., Stine, K., Gillis, C., Sullivan, A. G., Voegele, C., Mrotek, L. A., Afshar, M., Burnside, E. S., Hankwitz, J., Rasmussen, S., Jackson, R., & Kohler, B.
- Affiliations
- University of Wisconsin–Madison (US)
- University of Wisconsin Health (US)
- Corresponding author
- Wieben, Ann
- ORCID
- Venue
- medRxiv
- Publisher
- openRxiv
- Published
- 2026-07-13
- OpenAlex
- W7168123157
- Cited by
- 0
- References
- 23
- Type
- posted content
- Language
- en
- Open access
- green
- doi.org
- Licence
- Locations
- medRxiv (acceptedVersion, cc-by)
- Funders
- National Institute of Health Clinical and Translational Science Award (NIH/NCATS UL1TR002373)
- Topics
- Health Sciences › Nursing › Research and Theory › Nursing Diagnosis and Documentation
- Health Sciences › Health Professions › Health Information Management › Electronic Health Records Systems
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Keywords
- ambient AI, nursing documentation, documentation burden, inpatient nursing, nursing workload, human-in-the-loop verification, clinical workflow integration, shared governance, stepped wedge design, usability
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.