The Twenty-Eight Patient List Problem

Twenty-eight patients on the list. That is the number a PGT in a busy general medicine unit in India can carry on a single shift. It is also, demonstrably, more than working memory can reliably hold.

Miller's classic (Psychological Review, 1956) put the working memory span at 7 ± 2 items. Later work (Cowan, Behavioral and Brain Sciences, 2001) refined this downward to four when items are unlinked. Twenty-eight is approximately seven times that capacity.

What residents actually do

What residents actually do is chunk. They group patients by clinical category — three sepses, two CCFs, four post-ops, one stroke. They use bed numbers as anchors. They use the corner of the paper list as a working notation, sometimes a tally, sometimes initials, sometimes a colour code with one pen.

It is a genuinely impressive adaptation and it should be said plainly: a resident who can hold twenty-eight patients in four chunks and retrieve any one of them on demand is doing something cognitively remarkable. The chunk is not a shortcut. It is a compressed model — "Bed 12 is the septic one who is improving" — and it works because most of what you need about most patients most of the time is the category.

Where the chunk breaks

The chunking works for retrieval. It does not work for changes. When patient eleven's potassium comes back at 5.9 between rounds, the chunk doesn't update — the resident has to manually re-encode the patient into a new category. This is when things drop. Not at the patient's bedside. In the gap between the lab returning and the resident's next look at the list.

The mechanism is worth stating precisely, because it is not forgetfulness. The compressed model is stale, not absent. The resident still confidently knows something about Bed 11 — they know the version of Bed 11 that was true at 9 a.m. A stale model does not feel like a gap in knowledge; it feels like knowledge. There is no experience of not-knowing to prompt a check.

That is why "remember to review the labs" fails as an instruction. You cannot set a reminder for the thing you already believe you know.

The 47-minute gap

We measured this informally for a single resident at a tertiary government hospital for a week. Over the seven shifts, the lag between a critical lab returning and the resident actively reviewing it ranged from four minutes (when the lab phoned) to 47 minutes (when it sat in the system unread). Of nine critical-range values, four were noticed by the lab call, three by the next routine round, one by the nurse, and one by the resident reviewing the patient's chart for an unrelated reason.

Read that breakdown again with an eye to who did the noticing. Four by the lab. One by the nurse. One by accident. Three by a routine that happens on a fixed schedule regardless of the result.

Zero were noticed by the resident going to look because they were expecting something. The resident's system for catching critical results is, in practice, other people's processes and luck — which works, until the night the lab is short-staffed and does not phone.

Why an alert is the wrong instrument

The obvious fix is to make the phone buzz. It does not survive contact with the ward.

Twenty-eight patients generate a continuous stream of results, most of them unremarkable. Any alerting threshold loose enough to catch the potassium at 5.9 will also fire for a dozen things that do not matter, and the resident — who is examining a patient, or talking to a family, or in the middle of a procedure — cannot be interrupted a dozen times an hour. What they will do instead, quickly and rationally, is learn the dismissal gesture. Within a week the buzz is being cleared before it is read.

An alert also demands an acknowledgement, which converts noticing into a task. Tasks accumulate. A resident with nineteen unacknowledged alerts has a worse instrument than one with none, because now the queue itself is the thing being avoided.

Peripheral attention as a design material

A workspace should not make the resident remember to look. It should show the change. In Medows, the colour of the patient's card on the ward list updates the moment a new lab value arrives. K+ at 5.9 turns the card amber. ECG findings amend the same card. A new vital out of range shifts the trend strip.

This is not an alert. Alerts get muted. It is a passive change in the surface itself — the kind of change your peripheral attention picks up while you're walking past on the way to another bed.

Designing for that channel has its own rules, and they are unusually strict:

• It must read in one second, at arm's length, without focusing. That means colour and position, not text.
• It must cost nothing to ignore. No acknowledgement, no dismissal, no badge count. The resident who glances and moves on has lost nothing.
• It must not move under the reader. A card that re-sorts while the resident is reading it destroys the spatial memory — "the diabetic is third from the bottom" — that makes the list fast in the first place.
• It must decay. An amber that never resolves becomes wallpaper within two days.

The honest limit

None of this helps if nobody opens the list. A passive surface is only as good as the frequency with which it is glanced at, and there are stretches of a shift — a long procedure, a resuscitation, three hours in casualty — where the list is not looked at at all.

So the same change signal has to survive into the places the resident will look: the top of the list when they next open it, and the handover at the shift boundary, where a result that arrived unread during a code becomes a "Pending" line rather than a fact nobody has yet met. Ambient display is the first line, not the only one.

The 28-patient list problem is real, and it is not solved by trying to remember harder. It is solved by a list that does some of the remembering for you.