JMIR Publications

ABSTRACT

Background:

Ambient artificial intelligence (AI) scribes are being adopted rapidly, but evidence comes mostly from physician-led ambulatory care settings and relies on timestamps that may not capture how documentation flow occurs. Nurse-led telephone services, in which the clinician is not co-present with the patient, have been largely underexamined.

Objective:

To evaluate how an ambient AI scribe changed documentation time and workflow structure in a hospital-based aged care telephone support service, and to explain those changes through the accounts of the senior nurses who used it.

Methods:

We conducted an explanatory sequential mixed-methods study in a public hospital-based facility that advises residential aged care facilities through an Emergency Support Line (ESL). The quantitative strand used electronic medical record event logs from a baseline month (May 2026) and a one-month AI scribe trial (June 2026). ESL form completion times were compared within each nurse using Wilcoxon rank-sum and permutation tests, and across nurses with an exact sign-flip test. Process maps were discovered for each period. The qualitative strand comprised post-trial semi-structured interviews with senior nurses, conducted by a researcher independent of line management and analysed with reflexive thematic analysis. We integrated the strands by placing each event-log finding beside the related interview theme and judging whether the two agreed, disagreed or added new explanation, and how any disagreement could be accounted for.

Results:

We analysed 983 ESL forms from 5 nurses. Median ESL form completion time fell for every nurse, by 8.3 to 23.9 minutes (trial-to-baseline ratios 0.13 to 0.33; all P<.001); the mean of nurse-level median changes was −15.9 minutes (exact sign-flip P=.06). Process mining showed that this reduction coincided with a restructured workflow: a new recording activity (mean 14.4 minutes) preceded most forms, followed by a mean 86-minute interval before form completion, and a separate approval step largely disappeared (514 to 11 events). However, interviewees [n=6] described the opposite experience: a perception of a longer documentation process; unpredictable draft quality; omissions and verbosity requiring full re-listening; loss of the clinician's own voice; and continued manual backup notes. Integration indicated that the measured form-level saving largely reflected relocation of documentation work outside the timed interval rather than net time saved.

Conclusions:

In nurse-led telephone triage, an ambient AI scribe shortened the time-stamped documentation step while nurses experienced greater overall burden. Single-interval time stamps can overstate AI scribe benefits; evaluations should measure end-to-end episode time, verification work and draft quality, and allow longer lead-in periods before judging effect.