Sepsis That Walks In Talking

She is 71. Post-op day 2 after laparoscopic cholecystectomy. Vitals on every individual check look fine.

Time	HR	BP	SpO₂	Temp
9 a.m.	84	124/76	97%	37.2°C
1 p.m.	91	118/72	96%	37.7°C
3 p.m.	96	110/70	94%	38.1°C

No individual number triggers a paper-based alarm. Each reading sits well within "normal." But the slope is unmistakable when you see all three together. Heart rate drifting up. Saturation drifting down. Temperature climbing.

This is what we mean by sepsis that walks in talking. Patients deteriorating slowly enough that no single check is alarming, fast enough to be a different patient by morning.

Why nothing fired

The obs chart on the wall is a threshold instrument. It has a normal band, and it asks one question of each reading: is this value inside the band? Every value in that table answers yes.

The thresholds are set where they are for a reason. Tighten them and the ward drowns in alarms that mean nothing, staff learn to silence them without reading, and the one that matters is silenced in the same reflex. Loosen them and you get this table. Every threshold system lives somewhere on that trade-off, and wherever you put it, the patient who moves steadily within the band walks straight through.

Which is the thing worth noticing: this is not a failure of a badly-set threshold. It is a failure of the question. "Is this value normal?" cannot detect a patient who is becoming abnormal. Only "how does this value compare with that patient's last three?" can, and no paper obs chart has ever asked it.

qSOFA is a snapshot too

The Sepsis-3 redefinition (Singer et al., JAMA, 2016) made qSOFA a screening tool for exactly this kind of patient — three trivial measurements (RR ≥ 22, altered mental state, SBP ≤ 100) flagging a deteriorating trajectory. But qSOFA is still a point-in-time check. The published validation studies (Freund et al. and Raith et al., both JAMA, 2017) all use single measurements.

Run it on this patient at 3 p.m. and see what happens. Systolic 110 — no point. Alert and oriented — no point. Respiratory rate — not charted at all that afternoon, which is the single most common gap on any obs sheet in any hospital. She scores zero. She has an E. coli bacteraemia and a bile leak.

A score of zero is not a false reassurance the tool intended to give. It is a correct answer to a question nobody should have been asking on its own.

What the trend strip has to show

What's missing in the standard workflow is the trend strip. The view that shows you the patient's curve over the last six hours, on the same screen as their card, in a colour that shifts when the slope is wrong.

Four things have to be true of it or it does not work:

1. Direction, not value. The visual encodes which way the line is going. A number, however well formatted, requires the reader to remember the previous number.
2. A window long enough to have a slope, short enough to be about now. Six hours catches the afternoon drift. Twenty-four hours flattens it into noise.
3. On the card, not behind a tap. Anything requiring a tap gets looked at when you are already suspicious, which is the moment you no longer need it.
4. Colour derived from the slope, not from the reading. Otherwise you have rebuilt the threshold system in a nicer font.

In Medows, the trend strip is the front layout for every bed card. HR, BP, SpO₂, and temp are inline mini-bars — six hours of history, colour-coded by direction. When the curve points the wrong way, the card shifts amber, then red, regardless of whether any individual value is critical.

The false-positive problem, honestly

A slope-sensitive card fires more often than a threshold does. That is arithmetically unavoidable — there are more ways to drift than to cross a line — and any claim otherwise is marketing.

Post-operative patients drift constantly and mostly innocently. Pain makes the heart rate climb. Atelectasis drops the saturation on day two in a patient who will be fine. A ward where every one of those turned amber and demanded acknowledgement would be a ward where amber means nothing by the end of the first week.

So the design constraint is that the signal must be cheap to be wrong about. The card changes colour and changes its position in the list. It does not interrupt, it does not require dismissal, it does not page anyone, and there is no button to click to make it go away. A resident glancing at an amber card and thinking "that's the atelectasis" has spent one second and lost nothing. That is what buys the right to be wrong four times out of five — and being wrong four times out of five is entirely acceptable for the fifth. It is the same argument as the bed card in decompensated heart failure, arriving from the opposite direction.

What happened

For the patient above, the card was already amber at 1 p.m. and a soft red at 3 p.m. The PGT noticed it on the way to lunch, looked at the trend, asked Medows about the differential for a post-op day 2 patient with this trajectory. The response included blood cultures from two sites, lactate, urinalysis, a fluid challenge, and a flag to call surgery about possible bile leak.

Cultures grew E. coli. The bile leak was confirmed at re-look surgery the next morning. She went home on day 9. She was not an ICU admission. She walked out talking.

That is the case for a workspace. The trend was there in the data. It just needed a surface that showed it.