Method
How the questions are written and checked
- Which sources, in which order
- Question style
- Explanation structure
- How questions are checked
- What AI assistance means here
- Not clinical advice; not affiliated
Which sources, in which order
Every question is anchored to a named source. When sources disagree, the hierarchy is:
- BSG and European guidelines (for example ESGE, EASL, ECCO, UEG). These are the documents the ESEGH blueprint is written against, so they take precedence.
- NICE guidance and technology appraisals, where they add a UK threshold or pathway the European guidance leaves open.
- Named trials and landmark studies, used when a guideline cites them for a specific number, and when no guideline covers the point.
Where a guideline has been updated, the current version is used and the date is given in the source line. Older guidance is not used as a source once it has been superseded.
Question style
The exam rarely asks you to make a diagnosis. It gives you the diagnosis and asks what you do with it. The bank is written the same way.
- The diagnosis is usually in the stem. The discriminating step is what is tested: the dose, the threshold, the interval, the next investigation, the point at which you stop or escalate.
- One best answer out of five. Distractors are chosen because they are plausible, not because they are absurd. Often one of them is the right action for a slightly different patient.
- The stem contains what you need. Investigations are given with units and reference ranges where they matter. There are no trick omissions.
- Topics are weighted to the blueprint. The ten blueprint categories are represented in roughly the proportions the exam uses, so a session of mixed questions feels like the paper.
Explanation structure
The explanation appears as soon as you answer. It has the same four parts every time, so you know where to look:
- Key points. Two to four short statements worth memorising. These are what the Anki export uses.
- Teaching. Paragraphs that work through the decision in the stem: what the guideline says, why, and where the common errors are. Tables are used where a comparison is clearer than prose.
- Why the distractors are wrong. Each of the other four options, with the specific reason it does not fit this patient. If an option would be correct in a different situation, that situation is named.
- Source. The guideline or trial, named with its year, so you can check it.
How questions are checked
This is the honest version.
- Claims are checked against source text. Doses, thresholds, intervals and recommendations in an explanation are compared with the wording of the guideline or trial they cite, not with memory or a textbook summary.
- A core set of 176 questions has been source-verified. These are the ST4 decision-pathway questions, covering the management decisions that appear most often. Each claim in each of these has been traced to its source.
- The wider bank is reviewed on a rolling basis. The remaining questions have been written to the same standard but have not all been through the line-by-line verification above. They are being worked through in blueprint order. Some errors will remain until they are.
- Error reporting is built in. Every question has a report button. Reports go directly to the author and are looked at individually.
- Corrections are versioned. When a question changes, its revision number increases and the change is recorded.
What this process does not do: it does not involve a second clinician reviewing every question, and it does not involve the exam board. If you find something wrong, the report button is the quickest route to getting it fixed.
What AI assistance means here
The questions and explanations are drafted with the help of large language models. In practice that means:
- The author sets the topic, the decision to be tested, the source to be used and the format. The model produces a draft stem, options and explanation to that specification.
- The author reads the draft, compares its claims with the source text, and edits or rejects it. Model output is never published unread.
- Models are also used for consistency checks across the bank: duplicated topics, option length, stem length, and whether every explanation actually names a source.
AI assistance is why the bank is large and consistent in format. It is not a source of authority, and nothing in the bank is correct because a model said it. The author is a gastroenterology registrar, not a consultant, and is preparing for the same exam. Treat the bank as a well-referenced revision tool written by a peer, and check anything that matters against the cited guideline.
Not clinical advice; not affiliated
This is exam preparation, not clinical advice. Doses, thresholds and pathways are given as the exam expects them and may be simplified. Do not use the bank to make decisions about patients. Use the current guideline, your local protocol and your senior.
GastroQBank is independent. It is not affiliated with, endorsed by or connected to MRCP(UK), the Federation of Royal Colleges of Physicians of the UK, the British Society of Gastroenterology (BSG) or the European Section and Board of Gastroenterology and Hepatology (ESBGH). References to the ESEGH, its blueprint and its format are descriptive. No question in the bank is taken from, or claims to resemble, a real exam item.