There is a category of exam error that feels deeply unfair, and it is one of the most common failure modes for international graduates and for anyone using resources written for another country. You know the medicine. You reason correctly. You select an answer that is entirely defensible, that you have seen work in practice, and that any competent doctor in the country you trained in would endorse. And it is marked wrong, because the exam is set in a different health system, and in that system the rule is different. This is not a knowledge gap and it will not be fixed by learning more medicine. It needs its own diagnosis and its own remedy.
Key takeaways
- An answer can be clinically sound and still wrong for your exam, because guidance is national.
- Contamination clusters in four places: drug names, units, screening and prevention, and thresholds.
- Tag these errors separately, because they are not knowledge gaps and revising harder will not fix them.
- General AI tools are a major source of contamination, because they blend guidance across jurisdictions.
- Do not mix question banks from different countries in the final phase of preparation.
The four places contamination lives
The divergence is not random. It concentrates where the health system, rather than the disease, determines the answer.
Drug names. The same molecule has different names in different markets, and a paper written for one will not accept the other. Paracetamol and acetaminophen. Adrenaline and epinephrine. Salbutamol and albuterol. Answering fluently in the wrong nomenclature is the most superficial form of contamination and the easiest to fix, but it signals a deeper problem: you are reasoning inside the wrong system.
Units. This one is genuinely dangerous, because it can invert a clinical judgement rather than merely misname it. Glucose in millimoles per litre or milligrams per decilitre. Cholesterol on two entirely different scales. Creatinine reported differently. Glycated haemoglobin as a percentage or in millimoles per mole. A number that means "well controlled" in one system means something quite different in another, and a candidate who has internalised the wrong scale will confidently misclassify a patient.
Screening and prevention. This is pure health-system knowledge, and it is unlearnable by clinical reasoning. Which programmes exist, who is eligible, at what age they start, at what interval they run, and when they stop. These differ substantially between countries, and no amount of understanding pathology will let you infer your exam's answer.
Thresholds and pathways. When to refer, how urgently, along which route, and to whom. Referral pathways are constructed by health systems and have no natural existence, so the answer is national by definition. So too are the treatment thresholds in many guidelines, where the same evidence has been converted into different recommendations by different national bodies.
The insidious ones: risk scores and targets
Beyond the obvious categories sit the divergences that catch even careful candidates, because they involve tools that look universal and are not.
Cardiovascular risk is calculated with different instruments in different countries, and the instrument determines the answer to "does this patient need a statin". A candidate who has internalised one country's calculator and its thresholds will reason impeccably to the wrong conclusion in an exam that uses another.
Blood pressure targets have diverged between national bodies, sometimes substantially, and the same patient can be hypertensive under one guideline and not under another.
Even scores that share a name evolve differently. The stroke-risk scoring used in atrial fibrillation has been revised in different ways by different bodies, including changes to whether sex is counted as a risk factor at all, so the same patient can generate a different score and therefore a different anticoagulation decision depending on which version your exam is testing.
These are the errors that feel most unjust, because the reasoning was flawless. The reasoning was also grounded in the wrong document.
Where the contamination comes from
Three sources, and they are worth naming because the remedy differs.
Your own training. If you trained abroad, your defaults are the defaults of that system, and they are deeply internalised because they were correct for years. They will not feel like guesses. They will feel like knowledge, which is precisely why they are dangerous and why they survive revision untouched.
Resources written for another market. A question bank, a textbook or a revision course written for a different country teaches that country's rules, faithfully and well. Using it for a different exam imports the contamination wholesale.
General AI tools. This is a growing and under-recognised source. A general-purpose model trained on the whole internet has absorbed guidance from every jurisdiction, and unless it is explicitly grounded in the guidance your exam uses, it will blend them, often fluently and confidently. An answer that sounds authoritative and cites nothing is exactly the shape of this failure.
Tag these errors separately
Because this is not a knowledge gap, it must not be filed with your knowledge gaps.
Create a distinct category in your error log for imported-guidance errors, and use it whenever the diagnosis is "I knew the medicine and applied the rule from another system". The test is simple: could you state a coherent clinical rule that produced your answer? If yes, and it simply is not this country's rule, it belongs here.
Once you have twenty or thirty of these, look at them together. They will cluster, and the clusters are informative. Most candidates find their contamination concentrated in prescribing, or in referral, or in screening, rather than spread evenly, and a cluster is a single problem that can be corrected once.
The review protocol
For each contaminated error, do three things.
Write down the rule you applied, before you look at anything. This exposes the imported rule explicitly, which matters, because until it is on paper it will keep feeling like knowledge rather than like an assumption.
Find the rule your exam's jurisdiction actually uses, from the guidance your exam is grounded in rather than from a general search, and state the divergence in one line: "I would have done X, this exam expects Y."
Then re-test the principle in a different clinical setting. If you got a referral-threshold question wrong in dermatology, seek out referral-threshold questions in urology. That is how you find out whether you have corrected a rule or merely memorised an answer.
Do not mix banks late
One practical rule for the final phase.
Early in preparation, using material from more than one country is survivable and occasionally useful, because breadth is breadth and medicine is medicine. In the last four to six weeks, it is actively harmful. You are trying to install a single, consistent set of national defaults so firmly that they fire automatically under time pressure, and practising against a different country's rules in that window undermines exactly the automaticity you are building.
Pick the bank that is grounded in your exam's jurisdiction, and finish on it.
Where iatroX fits
iatroX's UK banks are grounded in NICE, CKS, SIGN and the SmPC, with the source attached to the explanation, so the rule you learn is the rule your exam is testing and you can see where it came from rather than taking it on trust. Its international banks are localised to their own markets rather than being a translated UK bank, which is the specific thing that causes contamination when a resource is repurposed across borders. Missed questions can be opened in the Socratic Tutor, which asks you to reason before it explains, and that is precisely the step that surfaces an imported rule, because a candidate with a contaminated default will state it confidently and then see exactly where it diverges. Try it with free sample questions at iatroX. For how to extract a rule from a source without disappearing into it, see source-grounded explanations.
Frequently asked questions
Why do I get questions wrong when my answer is clinically correct? Because guidance is national. Drug names, units, screening programmes, referral pathways and treatment thresholds all differ between countries, and an exam tests the rules of the system it is set in, not general clinical truth.
Which topics are most affected by cross-country differences? Prescribing and drug nomenclature, laboratory units, screening and prevention programmes, and referral thresholds and urgent pathways. Risk calculators and blood pressure targets also diverge, and those catch careful candidates because the tools look universal.
Are general AI tools safe to use for exam revision? Only with care. A general model has absorbed guidance from every jurisdiction and will blend them fluently unless it is explicitly grounded in the guidance your exam uses. Confident, uncited answers are the signature of this failure.
Can I use a question bank written for another country? Early on, with caution. In the final four to six weeks, no. You are trying to install one country's defaults so they fire automatically under pressure, and practising another country's rules in that window undermines exactly that.