Ask a clinician what happens when they are running behind and they will tell you they go faster. That is not quite what the evidence shows. What actually happens is more specific, more predictable, and considerably more useful to anyone designing education.
They do not accelerate uniformly. They stop looking for the thing that probably is not there.
I.Start with the arithmetic
Before any of the cognitive science, there is a constraint that most educational design simply ignores. In 2022 a team led by Justin Porter modelled what it would take for a primary care physician to deliver all guideline-recommended care to a real patient panel — real data, real multimorbidity, including documentation.
II.What compression actually does to a decision
The cleanest experiment we know of on this is small, and it deserves to be much better known. Tsiga and colleagues took 34 general practitioners and put the same clinical scenarios in front of them twice — once with no time constraint, once with a countdown clock running on the screen. Same doctors. Same cases. The only variable was time.
Here is what did not change: the diagnosis. In both conditions, 98.53% reached the correct answer. Identical, to two decimal places.
Here is what did.
Two features of that result matter more than the headline.
The first is what survived. Lung auscultation was performed by 94% of GPs unhurried and 98.5% hurried — it went up. The prototypical, high-yield, expected examination held firm. What collapsed was the check that exists precisely because the thing it looks for is rare.
The second is that they knew. Confidence in the diagnosis fell from 85.6% to 62.7%. These were not doctors sailing through in blissful ignorance. They were aware they had cut something, and they cut it anyway, because something had to give.
Compression does not degrade the answer. It removes the safety net — and it removes it invisibly, because the answer still looks right.
III.Why the search narrows in that particular way
This is not a failure of diligence. It is how expert cognition works, and the mechanism is well described.
Heuristics — intuitive shortcuts for estimating probability — are not a defect in clinical reasoning. They are what makes it possible. The alternative is to compute a base-rate-adjusted pre-test probability for every item on the differential, weigh the performance characteristics of every candidate test, and derive a post-test probability before acting. No one has ever practised that way, and nobody could.
Two of these shortcuts do most of the work, and both have a known failure mode.
- Representativeness — judging likelihood by how well the patient matches the prototype of a condition. It ignores prevalence. And crucially, prototypes of rare diseases are built on a smaller observed sample, so they are less accurate. The result is an underestimated pre-test probability, a testing threshold that is never reached, or the wrong focused history being taken in the first place.
- Availability — judging likelihood by how easily an example comes to mind. Common diagnoses are recalled readily and considered promptly. Uncommon ones are not.
Under time pressure these sharpen. The search contracts around the prototype, which is usually right, and the tail falls off the edge. Tsiga's authors reached for exactly this explanation: under time pressure, GPs were more likely to overlook the less-frequent condition.
IV.This is how every expert under time pressure works
It would be convenient to treat this as a medical problem. It is not. Gary Klein's studies of fireground commanders, trauma nurses and other experts produced the recognition-primed decision model, and it describes the same thing: experts under time pressure do not compare options. They recognise the situation as an instance of a type, generate the first course of action that fits, run it forward mentally to see whether it makes sense, and act. If it survives the simulation, they do it. They do not generate a second option to weigh it against.
This is fast, and it is usually right, and it is the reason expertise is worth having. It also has one documented precondition. Recognition-primed decision-making requires extensive prior experience of that situation type, because the whole mechanism depends on there being a stored pattern to match against.
Which means it fails precisely where there is no pattern — the rare presentation, the therapy approved last year, the guideline that changed after the clinician trained.
V.What other high-stakes fields did about it
Aviation, surgery and anaesthesia all confronted this. None of them responded by teaching harder.
They put the knowledge in the room
Aviation's step-change in safety came from checklists, standardised callouts and cockpit redesign — from changing what happens in the moment — far more than from additional pilot instruction. Medicine borrowed it: a 19-item surgical safety checklist, introduced across eight hospitals on four continents, cut death and inpatient complications each by more than a third. Nobody had been taught new surgery.
But the study that isolates the variable most cleanly is a simulation trial of surgical crisis checklists.
of life-saving steps missed by expert operating-room teams managing simulated surgical crises.
Adjusted relative risk 0.28. Every single team performed better with the checklist than without it.
of participants said that if one of these crises happened while they were the patient, they would want the checklist used.
Seventeen teams, 106 simulated crises, across an academic centre and two community hospitals. These were trained, experienced clinicians. Knowledge was not the variable. They knew the material. What changed was whether it was present in the room at the moment it was needed.
And the correction that keeps everyone honest
When Ontario later mandated the surgical safety checklist across its hospitals and measured the result, there was no significant reduction in mortality or complications. The tool had become a form signed on the way past. A cognitive aid works when it genuinely changes what happens; it does nothing when it becomes paperwork. Anyone arguing that education belongs at the point of care — as we do — has to hold both findings at once.
VI.What this ought to imply for education design
If the clinical decision is made by pattern-recognition under compression, then education has three honest options, and only three.
- Build the pattern. Deliberate practice across varied cases, spaced over time, with the learner made to commit to an answer. This is the only route that changes what recognition retrieves — and it is slow, and it works only where enough cases exist to build a pattern from.
- Supply the pattern at the moment. Where no clinician will ever accumulate enough exposure — which is the definition of rare disease — the only intervention that can work is one present at the decision.
- Target the check that gets dropped. Not the diagnosis. The diagnosis was fine. The nervous system examination was not.
VII.What education design does instead
Almost none of that. The standard programme is delivered in a single block, in advance of any particular patient, largely passively, and is measured on recall.
And there is a deeper mismatch, which is worth stating plainly because it is rarely said. Educational content overwhelmingly teaches the typical presentation — the classic case, the prototype, the thing that appears in the case vignette. But the typical presentation is exactly what representativeness and availability already handle well. It is the part clinicians get right at 98.5% with a clock running.
So the field spends heavily on reinforcing the part of reasoning that works, measures the faculty that is not failing, and delivers all of it at a moment when the clinician is not deciding anything.
VIII.What would follow from taking this seriously
- Design for the moment of compression, not the hour of study. Ask where the clinician will be and what they will have when the decision arrives.
- Pick the dropped check as the target. In the Tsiga data the intervention worth building is not "teach viral RTI better". It is "make the nervous system examination happen when the clock is running."
- Build patterns where patterns can be built — varied cases, spaced, with retrieval — and stop pretending that is what a single webinar does.
- Supply the pattern where it cannot be built. In a disease a clinician meets once in a career, recognition has nothing to retrieve. No amount of prior teaching fixes that; only presence does.
- Measure whether the dropped check got done. Not whether the learner enjoyed the module.
IX.The honest caveats
- The Tsiga study is small
- 34 GPs, simulated scenarios, one country, and viral respiratory infection — a condition with limited management options. The authors say so themselves, and note the effect of time pressure might be larger in more ambiguous cases, not smaller. It is an experiment, not an audit.
- Recognition-primed decision-making is descriptive
- It tells you how experts decide, not how they ought to. It is a strong account of the mechanism and a weak basis for prescribing an intervention on its own.
- Cognitive-aid evidence is strongest in crises
- Surgical crises have defined protocols and a clear trigger. Most outpatient decisions have neither. Transferring the finding requires care, and the Ontario result shows what happens when it is transferred carelessly.
- And we are not neutral
- We design education delivered at the point of care, so we came to this question already holding a view. That is a reason to check our reasoning rather than dismiss it — but the reader should know it, and weigh the argument accordingly.
X.The short version
Clinicians under time pressure do not become worse doctors. They become narrower ones. The search contracts around the prototype, the answer stays right, and the check that existed to catch the rare thing quietly stops happening — while confidence falls, which tells us they can feel it going.
Every other high-stakes field that has confronted this has concluded the same thing: the fix is not more instruction beforehand, it is the right thing present at the moment. Medicine has the evidence, has borrowed the tool in the operating theatre, and has almost entirely failed to apply the lesson to how it educates.
The clinician does not need to know more. They need what they already know to be there when the clock is running.