Clinician workload in EHR design: inbox, documentation and alerts

30 September 2026

Patient portal & EHR UX Patterns for Health Trust, Ethics & Safety

In one 2017 log study of 471 primary care physicians, desktop tasks took about as long as seeing patients. This article looks at what research measures in documentation, the inbox and alerts, and what design can and cannot change.

Not clinical or legal advice This article summarises published research for design education. It is not clinical advice, not a compliance verdict and not a judgement on any named product. Most studies below come from one health system or one record vendor, so treat the figures as examples, not as universal rates.

Clinician workload in a health record is the time and attention that documentation, patient messages and alerts take. In a 2017 Health Affairs study of 471 primary care physicians, the logs showed 3.17 hours a day on desktop tasks against 3.08 hours on face-to-face visits (Tai-Seale et al.). That is one health system in 2011-14, so it is a snapshot. It still shows why interface designers belong in a workload conversation, not only a usability one.

Three sources of load

Most of what a clinician does inside a record falls into three groups. Documentation is writing the note, the order and the referral. The inbox is the stream of patient messages, prescription requests and results waiting for a reply. Alerts are the prompts that interrupt the first two. Each group has its own evidence, and each responds differently to design.

A 2024 narrative review in JMIR Medical Informatics (Asgari et al.) gives a useful frame. It describes cognitive load as the mental effort needed to hold and process information in working memory. It also separates load that comes from the task itself from load that comes from how information is presented. Design mostly acts on the second kind. It cannot make a clinical decision simpler, but it can stop the screen adding effort of its own.

Documentation: where the time goes

The Tai-Seale study used time stamps in the record to estimate work. Of all logged time, 49 percent was face-to-face visits and 51 percent was desktop medicine. Within the desktop share, progress notes were the largest piece, at 34 percent of the total. The authors list limits. Phone time was not captured, time when the physician did not touch the record was not counted, and the health system was an early adopter of the record, so results may not carry to other settings.

The review by Asgari and colleagues names several contributors to documentation burden: many templates, copy-and-paste that can carry wrong data forward, and documentation required for reasons beyond direct care. Some of those are policy choices, not screens. A designer can still ask which fields a note really needs from the person typing, and which the system already knows.

The inbox: what one study measured

A 2024 study in the Annals of Family Medicine (Arndt et al.) followed 141 academic primary care physicians at a single health system from May 2019 to March 2023. Times were measured per 8 hours of scheduled appointments. Between the year before the pandemic and the last study year, time in the record rose by 28.4 minutes (7.8 percent). Inbox time rose by 14.0 minutes (24.4 percent) and time on orders by 23.1 minutes (58.9 percent). Patient medical-advice messages rose by 5.4 per 8 hours (55.5 percent).

The authors state their own limits: one record vendor, one academic centre, and an inbox measure that understates the total time inbox work needs. The study counts time and messages. It does not show which design changes would reduce either. Do not read it as evidence that any particular inbox layout works.

The design question sits on the patient side too. A portal that makes it easy to send a message is likely to receive more messages. That is often a good outcome for patients, so the aim is not to hide the message box. The aim is to route each request to the right person and to give the clinician enough context to answer once. The Clinical vs Patient Interfaces Foundation piece and the article Clinical dashboard vs patient portal look at the same data from both sides.

Alerts: high override rates

A 2021 scoping review of drug-drug interaction alerts (Villa Zapata et al., MEDINFO 2021) covered 34 studies and reported override rates from 55 percent to 98 percent. A high override rate is not proof that clinicians are careless. The review concludes that decision support needs to take user, drug and institutional factors into account. Asgari and colleagues link excess notification to alert fatigue and recommend measuring alert burden and usefulness, which they call alert stewardship.

The Showcase card Alert fatigue shows one design response: a flat list of alerts set against a list sorted into tiers. The card illustrates a pattern. It is not a measured result. Which alerts should exist at all is a clinical governance decision that designers inform but do not own.

Interruptions cost attention even when the alert is right. The UX Patterns for Health Foundation piece covers interruptive prompts and cognitive load in more detail.

What design can and cannot change

Design can change how information is presented, how many steps a task takes and whether an alert interrupts or waits. The reviews above treat these as part of the picture.
Design cannot change how many messages patients send, what regulators require to be documented, or staffing. Those sit with the organisation.
Design cannot on its own show that a change saves time. That takes measurement in the real setting, which the studies above did with logs, not opinions.
A design that removes work in one place can move it to another. The Medication reconciliation card is a task where several sources have to be gathered in one view, so the clinician does not hunt across screens.

Questions a designer can ask

Which tasks do clinicians repeat most often, and how many screens does each take today?
Which fields does the person type that the system could already fill in?
For each alert, who acts on it, what do they do next, and how often is it dismissed? Is anyone measuring that?
Which patient messages could go to another role, and does the clinician see enough context to reply once?
What work happens after hours, and on which screen?
Have clinicians taken part in testing, in the real workflow and not only in a meeting?

Asgari and colleagues also recommend involving clinicians in EHR design. The Process & Research piece covers how to test with them. For the wider context of portal and record redesign, see Patient portal & EHR redesign.