In the largest standardised study of its kind, US physicians scored their electronic health records at 45.9 on the System Usability Scale. Benchmarked against more than 1,300 usability studies from other industries, that lands in the bottom 9 percent, inside the range labelled not acceptable. A grade of F.
The same study found a dose response relationship with burnout. Every single point of improvement on that scale was associated with 3 percent lower odds of physician burnout. Not a correlation with satisfaction. With burnout.
Numbers like that get quoted often and understood rarely, because usability failures do not look dramatic. They look like eleven seconds here and a dismissed pop up there. So here is one shift, annotated. Each entry is a real pattern from the research, marked with what it costs and the rule it implies.
A note before we start. The clinician below is a composite used to make the evidence legible. The failures are documented. She is not.
07:12 — Login, and the eleven screens before the first patient
She logs in and works through session start, unit selection, and three notifications carried over from the previous shift. None of it is clinical. All of it is required.
What it costs. Start of shift friction is invisible in any feature list and unavoidable every single day. Multiplied across a department and a year, it is one of the largest recoverable costs in healthcare UX design, and nobody owns it because it belongs to no single feature.
The rule. Measure time to first clinical action, not time to load. A screen that renders in 400 milliseconds and still takes ninety seconds to get through has not been optimised, it has been benchmarked against the wrong thing.
09:40 — The alert she has seen four hundred times
Ordering a routine medication, she gets an interaction warning. She overrides it. It is the same warning she saw yesterday for the same patient.
What it costs. A systematic review of 23 studies found average alert override rates ranging from 46 to 96 percent. In one three year study at a large teaching hospital, clinicians overrode 73.3 percent of reviewed medication alerts, and roughly 40 percent of those overrides were inappropriate. In another analysis, only 7.3 percent of alert cases examined were clinically appropriate in the first place.
Repetition drives it. Research on ambulatory clinicians found that around a quarter of drug alerts and a third of clinical reminders were repeats for the same patient within the same year.
The rule. An alert that fires for a decision already made is not a safety feature, it is training in dismissal. Suppress within an episode, tailor by clinical role, and reserve interruptive design for the small set of cases that actually warrant stopping someone.
11:05 — Forty nine minutes she cannot see on any schedule
Between patients she works through her inbox. Results, refills, messages.
What it costs. Research in primary care settings found clinicians receiving over 56 alerts a day and spending around 49 minutes responding to asynchronous ones. That time is real, and it appears on no rota anywhere.
The rule. Design the queue, not just the notification. Most healthcare software treats the inbox as a dumping ground because nobody owns it as a workflow. Grouping, bulk actions, and defensible defaults are worth more here than anywhere else in the product.
14:30 — The workaround that became the process
To record something the system models badly, she uses a free text field the vendor intended for something else. Everyone on the unit does. There is a laminated card at the station explaining how.
What it costs. Every laminated card is a specification nobody wrote. The workaround is the real workflow, and it is invisible to analytics because on paper the feature is being used as designed.
The rule. Go and look. No amount of remote research surfaces the laminated card. Woltrio's UI and UX design work in clinical settings starts with observation for exactly this reason, because the gap between the documented workflow and the actual one is where most of the value sits.
16:50 — Three hundred reminders, one prevented event
A preventative care reminder fires. She dismisses it, because this patient discussed it last month and it is documented elsewhere.
What it costs. One frequently cited analysis found that nearly 300 reminders were required to prevent a single adverse drug event. The Joint Commission has estimated that between 85 and 99 percent of physiological alarm signals require no action at all. The signal to noise ratio is the design problem, in both the software and the hardware.
The rule. Every alert needs a measured yield, and alerts below threshold get retired. The reason this rarely happens is not technical. Nobody wants to be the person who removed a warning. That makes alert retirement a governance decision that must be designed for at the start, with named ownership and a review cadence, not left to whoever inherits the system.
19:20 — Pyjama time
The shift ended. The notes did not. She finishes them at home.
What it costs. This is the endpoint of every failure above, and it is the one clinicians describe when they talk about why they are leaving. Documentation burden is where poor clinical UX converts into workforce loss.
The rule. This is the strongest argument for ambient and assistive tooling, and the reason AI development in healthcare earns its place fastest here rather than in diagnosis. But it only works if the output lands inside the workflow. A summary the clinician has to go and fetch is another queue.
Why healthcare UX design is a safety discipline, not a styling one
Read the shift again and notice what never appeared. Colour. Typography. Layout polish. None of the costs above are aesthetic.
Clinical UX is about interruption cost, cognitive load, error recovery, and the sequencing of work under time pressure. A clinician works in short bursts, interrupted constantly, often with a patient present and consequences attached. That is the most demanding context of use in commercial software, and it is routinely designed for as if it were an office application.
This is also why demo quality misleads so badly. A demo runs the happy path, unhurried, with one task. A shift is nine interruptions and a laminated card.
What to do about it
Four things, in order.
Observe before designing. Sit with the users. The laminated card only exists in the room.
Instrument the boring metrics. Time to first clinical action, alerts per encounter, override rate per alert type, and after hours documentation time. These are the numbers that move burnout, and almost nobody tracks them at the product level.
Set a usability baseline. Run the System Usability Scale against your own product with real clinical users. You now have a benchmark and, given the dose response finding above, a defensible business case for the work.
Design the failure paths. What happens when the upstream system is down, the data is missing, or the user is wrong. In clinical software these are not edge cases, they are Tuesday.
Woltrio builds this into delivery rather than reviewing for it at the end, pairing design with frontend development and, where the work touches the record system, with custom EMR and EHR development. Where the workflow is genuinely unclear, a scoped MVP tested with real clinicians settles it faster than another round of wireframes.
The point is not that clinical software should be beautiful. It is that every dismissed alert is a design decision someone made, and the person paying for it is at home at 19:20 finishing notes.
Start with a scoped discovery from Woltrio.
