HCA Healthcare's Nurse Handoff app puts the EHR and AI output side by side

The split-screen is the product decision

The core of Nurse Handoff isn't a chatbot. It's a two-panel screen: the familiar electronic health record on one side, and the AI-generated shift summary on the other. Nurses review, cross-check, and add notes throughout the shift on a hospital-provided mobile device.

That layout is a deliberate frontend choice. Rather than asking nurses to trust a summary in isolation, the app keeps the source record visible alongside the generated output. The interface never hides where a claim came from — it lets a nurse look left to verify what appears on the right.

For a task HCA says happens roughly 60,000 times a day across 190 hospitals, the design brief was to make the tool 'as simple and straightforward as its name.' The restraint is the feature.

Citations built into the generated view

Product owner K.C. DeShetler, a registered nurse on HCA's Digital Transformation and Innovation team, describes the mechanics behind what nurses see: the team used retrieval augmented generation to attach citations to generated content, plus templates that organize the output the way nurses expect.

We fed our model the information nurses want to know, prompted the model in a multitude of ways, used retrieval augmented generation to identify citations for the generated content, provided templates for organizing information the way we want and so forth.Montana Labs

Citations and templates are frontend commitments as much as backend ones. A summary that traces back to notes, orders, and tests gives a nurse a way to check the machine before staking patient safety on it — and templates mean the output arrives in a shape a nurse can scan quickly at shift change, not a wall of prose.

Nurses as the editing loop

The interface was shaped through direct testing with frontline nurses. Samantha Hall, an RN at TriStar Hendersonville Medical Center, was an early tester who reviewed output to flag what was unnecessary, repetitive, inaccurate, or missing.

We went through that process three or four different times. And each time, it became a little bit more accurate, a little less filled with fluff that we don't need.Montana Labs

Cutting 'fluff' is a UX outcome, not just a model-accuracy one. What a nurse needs to see at 7 a.m. handoff is a curation problem, and the people doing the handoff were the ones deciding what stayed on screen. At one facility, nurses rated the tool 86% factual and 90% helpful — two distinct measures, one about correctness and one about whether the interface actually earns a place in the workflow.

Why a five-hospital pilot precedes 99,000 users

HCA is piloting Nurse Handoff in five hospitals before any plan to reach all 99,000 nurses in the system. That gap between pilot and full rollout is where a frontend either survives or fails.

The specific implication here: in a safety-critical setting, the interface has to carry the burden of trust that the model cannot. HCA's answer was to keep the source visible, cite the generated claims, template the layout, and let nurses rewrite what they see across several rounds. The AI ingests notes, orders, and tests; the design determines whether a nurse believes what comes back — and whether any of those estimated 10 million annual hours actually return to patient care.