JMIR Publications

ABSTRACT

Background:

Ambient Artificial Intelligence (AI) tools are rapidly being adopted across healthcare systems to reduce documentation burden, cognitive load, and clinician burnout. By reducing attention devoted to documentation during clinical encounters, ambient AI may also improve clinician-patient interactions and patient experience. However, evidence on the impact of ambient AI on patient experience remains limited, particularly in large, diverse, safety-net systems.

Objective:

The current evaluation examined the impact of Nabla, an ambient AI documentation tool, on patient experience by assessing: 1) clinician perspectives on how Nabla influenced engagement with patients during clinical encounters and 2) changes in patient satisfaction following Nabla implementation.

Methods:

We conducted an observational evaluation at Denver Health, a large, integrated, academic safety-net healthcare system in Denver, Colorado, where Nabla was implemented across ambulatory care settings beginning in December 2024. Descriptive statistics summarized clinician perceptions of Nabla’s impact on documentation, patient engagement, and clinical focus. Patient experience was assessed using top-box scores (5/5 ratings) for five Press Ganey Care Provider items: clinician concern, explanations, involvement in decision-making, discussion of treatment, and likelihood of recommending the clinician. Pre- and post-Nabla top-box ratings were compared using mixed-effects logistic regression, with survey period as a fixed effect and clinician as a random intercept to account for clustering. Odds ratios (ORs), 95% confidence intervals (CIs), and model-predicted probabilities were estimated; absolute differences were calculated as percentage-point changes in predicted probabilities.

Results:

During the first year of implementation, 848 ambulatory care clinicians used Nabla across 241,699 patient encounters. Of 293 clinicians who responded to the survey (estimated 19.5% response rate), 191 reported using Nabla within the preceding three months and were included in the analysis. Most reported spending less time documenting during patient encounters (77%) and improved ability to give patients undivided attention (83%), focus on clinical decision-making (62%), and build patient relationships (58%); 47% reported spending more time discussing care plans. A total of 16,410 patient experience surveys across 397 clinicians were included (10,985 pre-Nabla and 5,425 post-Nabla). In mixed-effects models, the odds of a top-box rating were significantly higher post-Nabla across all items (OR range, 1.22–1.26; all P<.001). Adjusted top-box scores increased by 3.2–3.7%, including clinician concern (78.1 to 81.4%), clear explanations (77.4 to 80.6%), involvement in care decisions (77.6% to 81.2%), discussion of treatment options (76.4 to 79.7%), and likelihood of recommending the clinician (78.4 to 81.5%) (all P<.001).

Conclusions:

Our findings in a large, diverse safety-net health system demonstrated that Nabla Ambient AI use was associated with improvements in both clinician-reported patient engagement and patient-reported experience. These findings suggest that ambient AI may have benefits extending beyond reducing documentation burden to improving the patient experience.