The Interrupted Shift

The shift starts quietly. Twelve patients on the ward list, mostly stable post-ops, two new admissions overnight. By 9 a.m. the resident is on the third bed when her phone vibrates. New admission in casualty — query meningitis, junior wants senior review. She puts the round on pause.

She resumes the round at 9:23. She is now thinking about the meningitis case as she stands at Bed 4. She misses that the diuretic dose was doubled overnight and that urine output dropped 200 ml. She moves on.

At 11:14, the lab rings. K+ on Bed 4 is 5.9.

She did not forget it

That is not a clinical error in the textbook sense. The resident saw the patient. Notes were complete. But the interruption between Bed 3 and Bed 4 cost her the working memory needed to register the trajectory before it became urgent.

The distinction that matters: she did not forget the diuretic change. Forgetting implies it was once held. It was never encoded at all. Both facts were on the chart, in front of her, legible — and reading a number is not the same operation as connecting it to another number and noticing that the pair means something. The connecting step is what working memory is for, and hers was still holding a query-meningitis case in casualty.

This is why the error is invisible from the inside. There is no moment of "I should check that." There is a resident who looked at a chart, saw nothing alarming, and was correct about every individual thing she saw.

The published cost

The published data on this is older than most realise. Tipping and colleagues (J Hosp Med, 2010) shadowed US hospitalists and found they were interrupted once every nine minutes on average. The cognitive recovery cost — the time before reasoning fully re-engages with the original task — runs from 23 seconds for trivial work to several minutes for complex differentials.

Westbrook and colleagues (BMJ Qual Saf, 2010) measured the dose-response. Each interruption during medication administration was associated with a 12.7% increase in clinical error rate. Four interruptions roughly doubled the rate of major errors.

A dose-response relationship is a strong finding. It means this is not a story about occasional unlucky interruptions landing at bad moments — it is a systematic, countable relationship between how fragmented a shift is and how often it goes wrong. Interruption density is a measurable property of a ward, like its bed occupancy, and unlike bed occupancy nobody measures it.

These numbers were measured in 2010. The structure of ward work has only gotten denser. Indian government hospital admission days mean 20–30 patients per resident, and the interruption count per shift — measured informally by the residents I've spoken with — sits comfortably above forty.

Why this is worse in medicine than in other interrupt-heavy work

We talk about resident burnout as if it were primarily about hours. Hours matter. But hours alone don't explain why a resident finishing an 80-hour week feels different from a junior software engineer finishing the same hours. The difference is the cost of every interruption, multiplied by the density.

There is a second difference, and it is the one that turns out to be actionable. When the engineer is interrupted, their context is on the screen. The open file, the failing test, the stack trace, the half-written function — all of it is still sitting there twenty minutes later, exactly as they left it. Resuming is a matter of reading. The externalised state does most of the remembering.

When the resident is interrupted, the context was in her head. The chart is still on the ward, but the chart does not contain the thing she lost, which was not a fact but a state of attention: that she was mid-round, that she had done Beds 1 to 3, that the thing she was tracking this morning was fluid balance. Nothing in the building holds that. She rebuilds it from scratch, badly, while standing at Bed 4.

The engineer's tools are built around the assumption that they will be interrupted. The doctor's are built around the assumption that they will not be.

What preserving context would actually require

You cannot restore someone's working memory. What you can do is make re-encoding cheap enough that a 23-second recovery does not turn into a missed trend. Concretely, that means a few things that no current ward tool does:

• The list remembers where you were. Which beds are done, which are not, on a round that started forty minutes ago.
• "Changed since you last looked" — not changed since yesterday, not changed since admission. Since this resident's last view of this patient. That is the only diff that maps onto what her head is missing.
• The interrupted task survives the lock screen. A phone that locks in the corridor and returns to a home screen has thrown away the state; returning to exactly where you were is not a convenience feature.
• What arrived while you were away, in one place. The diuretic change made overnight and the urine output drop are precisely the kind of thing that should be waiting in a single line rather than distributed across two chart sections.

None of these are clever. They are all the same idea — externalise the state that currently lives only in the resident's head, so an interruption costs seconds instead of a missed potassium.

What I keep coming back to is this: every ward tool we have — paper, EHR, WhatsApp, search engines — assumes the doctor's working memory is intact. None of them are designed to preserve context across an interruption. The patient list disappears when the screen locks. The handover sheet is paper, susceptible to smudge and pocket damage. The EHR has no notion of "where you were."

When we started designing Medows, this was the hidden problem we kept finding. Not "save the doctor time." Not "make the documentation faster." Preserve the context across the interruption.

A workspace, not a record.