Clinical Reasoning for Junior Doctors: A Ward Guide

You passed the exams. On paper, you can rank a differential in ten seconds.

Then it's 3am, half a story, obs drifting, collateral missing, and nothing fits the template. The nurse wants a plan.

Your ranked list stalls. Your brain pings between possibilities, the registrar's questions, and the bleep.

This post is about that gap, clinical reasoning under ward conditions rather than exam conditions, practical habits for getting unstuck.

What clinical reasoning actually means on a ward

Clinical reasoning is the process of turning what a patient tells you into a ranked set of possibilities and a next action, under time pressure and with incomplete information. Exams test the ranked list. The ward tests everything around it.

The method for building the list itself sits in its own guide, and this piece focuses on what you do before the list exists and after it stops helping.

Before the list, you frame the problem in the patient's words, define the time course, and decide what the immediate question actually is. You state what you don't know yet and what you won't assume.

After the list stops helping, you decide what single piece of information would move you, who needs to hear your uncertainty, and when to escalate. You document your reasoning plainly so the next person can pick it up at pace.

The exam version and the ward version are different jobs

Exams hand you a complete vignette. Everything relevant is on the page, the timing is kind, and there is a tidy answer if you rank the differentials the way the mark sheet expects. You are expected to find it.

On the ward you get a bleep, a name, a bed number, and half a story from someone who was not there for the first half. It feels like the call you take at 3am, not a case stem. No one hands you the missing lines.

The gap is plain. The exam tests whether you can order possibilities from complete information, the ward tests whether you can act on incomplete information without waiting for it to become complete, while keeping your model ready to change as new bits arrive.

The aim is not certainty. The job is reducing uncertainty, not removing it, enough to make a decision you can stand behind. You will commit while still unsure, then revisit as better information arrives. That is the work, not a failure of preparation.

Gathering is the half that actually fails

Clinical reasoning for junior doctors fails more on gathering than knowledge. You probably knew the diagnosis. You did not ask the question that would surface it.

Gerald Hughes is 72, an AI patient who volunteers a new, persistent, throbbing headache over the right temple for three to four weeks, completely unlike any he has had before. He thinks it is tension, eye strain or a trapped nerve.

If you probe, you learn his jaw aches and cramps when he chews and he has to stop and rest it. On a couple of occasions his right eye vision briefly greyed or blurred then returned, and both have worsened over the past week. Those two facts reorder the whole problem toward giant cell arteritis, which is sight threatening and urgent.

The discriminating fact is usually the one the patient does not think is related, so it never gets volunteered. He had not connected the jaw and eye symptoms with the headache until someone asked.

A new headache at 72 always needs a cause found, which makes his explanation unsafe to accept. A normal ESR does not fully exclude the diagnosis.

People routinely reach for tension type headache, migraine, trigeminal neuralgia, central retinal artery occlusion, or cluster headache. This is one of the ones that get missed at night.

Practise the gathering step Run targeted histories on an AI patient who only reveals facts when you ask. Live and early, voice or text, one free case a week, every case reviewed and signed off by a doctor. Start a free case

When a normal result should not reassure you

In clinical reasoning for junior doctors, a normal result only reassures you if the test could be abnormal at this point in the illness. Early in disease, several markers lag behind the physiology. A normal number is a snapshot of a moving process, not a verdict on it.

Frank Sullivan, 72, an AI patient in Diagnosica’s library, opens with severe central abdominal pain since this morning. Six hours in, it’s constant, excruciating, and hasn’t eased. He’s vomited, passed one loose stool, and thinks it’s something he ate.

He’s restless and can’t get comfortable. When you ask about his abdomen, the case reports it as soft, only mildly tender, with no guarding and no rebound. Pain out of proportion to the findings. You can’t lay hands on anyone in a simulator, the case reports the finding when you ask for it. Real hands on real abdomens stay with real patients, with a senior watching.

The lactate comes back normal. The CRP comes back normal. The debrief asks you this, word for word: “His lactate and CRP come back normal early, are you reassured? Why not?” CRP lags and is commonly normal early, rising later. A normal lactate early does not exclude ischaemic bowel. A normal abdominal film never excludes mesenteric ischaemia and must not be used to stop the work-up. If you want a friendly primer on interpreting normals and abnormals, Mind The Bleep’s guides to abnormal investigations are worth a read.

Here is the near miss. Amylase may be mildly raised in this case and gets relabelled as pancreatitis, then the work-up stops. Pancreatitis needs a threefold rise plus consistent imaging.

Before you send any test, say out loud what a normal result would mean. If a normal result wouldn’t move you, it’s not reassurance, it’s data arriving.

Holding two possibilities at once

Under pressure your brain wants one answer. The skill is carrying two, the working diagnosis and the can't-miss, in separate slots.

Andrew, 50, an AI patient, had a headache that hit maximum within seconds during exertion two hours ago. Migraine is tempting and is on the case's list of wrong answers, but it builds over minutes to hours and is usually a recognised, recurrent pattern, so the can't-miss is subarachnoid haemorrhage. First scan sensitivity is highest within six hours and a negative early scan does not fully exclude it.

Gerald, 72, an AI patient, reports a new right temporal headache for three to four weeks, with jaw cramping on chewing and brief greying of vision. Tension-type headache is the comfortable label and is on the case's wrong-answers list, but it does not cause jaw claudication or transient visual loss, and a new headache at 72 needs a cause sought, so the can't-miss is giant cell arteritis, sight-threatening and urgent.

Write both down. Next to the can't-miss, write the single thing that would put it to bed and ask what a test would actually change. If nothing can, it stays live and stays live out loud at handover.

What to do when your gut disagrees with the plan

The plan’s signed off and something in your head says no. Silence is the trap, then a retrospective note nobody reads.

First move, ask as a question, not a verdict. “Can I check my thinking with you, I’m not sure how the jaw pain fits the plan.” Questions invite answers, and verdicts invite defence.

Second, say the specific thing that doesn’t fit, not the vague feeling like “I’m worried”. “He said this came on within seconds” is a fact your senior now has to place.

Third, escalate without accusation. You’re asking for a second look, not filing a complaint, and if it’s still not resolved and you’re still worried, say so again, then go up one step.

Last move, write down what you thought and when. Not to build a case, but so that when the picture changes at 4am, the next person can see what was suspected and how long it’s been suspected for.

Harder to do than to write, especially three weeks in. For practical on-call language and FY1 context, see Mind The Bleep's FY1 guides.

How to practise this deliberately

Reading rarely changes nights. Reading about reasoning does not change what you do at 3am, saying your diagnosis out loud and then being challenged on it does. Here is a drill that trains it. Before you open the notes, say out loud what you think it is, what you cannot afford it to be, and what would change your mind today.

Then check yourself against what actually happened, not what you remember later. A quick note of your hunch, the red line you could not miss, and the one thing that would flip you, is enough.

On Diagnosica you talk through a case with an AI patient by voice or text, order investigations, commit to a diagnosis and a plan, then defend it to an AI senior in a debrief and, on the paid tier, get rubric scoring after every case calibrated to published mark sheets. The case script is fixed, and a finding stays hidden until you ask the question that surfaces it; it is live and early, rough edges expected, the free tier is one case a week, every case is reviewed and signed off by a doctor, and there is a no-signup demo on the landing page: talk or type, about 3 minutes.

Diagnosica does not do physical examination practice, and it is not a substitute for the wards. "It doesn't give advice about real patients, and it's not a diagnostic system. Treat every output as educational." If you are starting out, your first year on the wards pairs with this habit.

Educational use only. Not medical advice. AI-generated; verify clinically against primary sources.

Clinical review pending.

Debrief like it matters Commit to a diagnosis and plan, then defend it to an AI senior who tests your why. Every case is reviewed and signed off by a doctor, and you get one free case a week. Run a free case