A best-of-five question bank can prove you know how to assess frailty, apply the Mental Capacity Act, deprescribe safely and work with a multidisciplinary team. It cannot prove you can do any of them in front of an examiner. That gap between a correct selected answer and a performed skill is exactly what the Diploma in Geriatric Medicine (DGM) Part 2 Clinical is built to test — and it is the part of DGM that ordinary MCQ practice cannot reach. This article names those skills and sets out how to train and measure each.
The official format map
The DGM, awarded by the Royal College of Physicians (RCP) in collaboration with the British Geriatrics Society (BGS), has two components (RCP, last checked 21 July 2026). Part 1, the Knowledge-Based Assessment, is an online paper of 100 best-of-five questions over three hours with no negative marking — a pure knowledge test. Part 2, the Clinical, is an OSCE of four live stations lasting roughly 90 minutes, held in person at the RCP assessment centre at The Spine in Liverpool. Candidates have four years from passing the KBA to complete the Clinical. The KBA measures knowledge; the Clinical measures performance. A question bank lives entirely on the first side of that line.
Knowledge versus performance: what a correct answer proves
When you select the right best-of-five option, you have demonstrated recognition: given a fully assembled stem, you can identify the correct action. What you have not demonstrated is the harder work the Clinical assesses — gathering the information yourself from a real or simulated patient, structuring an assessment in real time, communicating a difficult decision to a relative, and adapting when the scenario does not go to script. Recognition is necessary and it is not sufficient. Every skill below is one where a candidate can score well on the KBA and still be unready for the station that tests the same topic.
The under-tested skills, one by one
Frailty assessment in real time
Observable behaviour: conducting a structured, efficient comprehensive geriatric assessment (CGA) — eliciting the relevant functional, cognitive, social and physical information and synthesising it into a frailty judgement — rather than reciting a frailty score. Deliberate-practice task: run timed CGA rehearsals on real or simulated older patients, verbalising your reasoning as you go. Feedback source: a geriatrician or trained examiner marking against the domains the station rewards. Exit standard: you can complete a focused CGA to time and defend the resulting management plan without prompting.
Mental capacity assessment
Observable behaviour: applying a functional, decision-specific capacity assessment under the Mental Capacity Act — testing understanding, retention, weighing and communication for this decision — and handling Lasting Power of Attorney, advance decisions and best-interests reasoning. Deliberate-practice task: role-play capacity assessments for varied decisions (discharge, treatment refusal, finances) with a colleague playing the patient. Feedback source: a clinician who can judge both the legal framework and the communication. Exit standard: you can run a defensible capacity assessment for a specific decision and document the reasoning, not just quote the four-part test.
Medicines optimisation and the deprescribing conversation
Observable behaviour: conducting a structured medication review in a multimorbid older patient — identifying adverse drug reactions, interactions and inappropriate polypharmacy, and explaining a deprescribing plan to the patient or relative. Deliberate-practice task: work real medication lists and rehearse the conversation, sourcing every decision from the SmPC/eMC and NICE/CKS. Feedback source: a clinician for the pharmacology and a role-player for the communication. Exit standard: you can justify each change against a current UK source and communicate it clearly and safely to a lay listener.
Multidisciplinary coordination and communication
Observable behaviour: demonstrating the roles and limits of the wider team, and communicating a plan across a transfer of care — the interface that carries real weight in geriatric practice. Deliberate-practice task: rehearse handover and family-meeting scenarios where the correct answer depends on who does what. Feedback source: a clinician who has run these meetings. Exit standard: you can lead a simulated MDT or family discussion to a safe, coordinated plan under time pressure.
A worked station: deliberate practice on a capacity case
Take a single station — assessing capacity for a proposed discharge — and see what climbing the ladder actually looks like. Rung one (isolated skill): rehearse the mechanics until they are automatic — introduce yourself, establish the specific decision, then test understanding, retention, weighing and communication in turn, without notes. Rung two (coached case): run the full scenario with a colleague playing a patient with early dementia who wants to go home; a clinician stops you afterwards and marks whether you assessed capacity for this decision rather than globally, whether you explored the least-restrictive option, and whether your manner was respectful. Rung three (timed integrated case): repeat under the real station clock, adding a relative who disagrees, so you must handle both the assessment and the communication in the time allowed. Rung four (unseen simulation): a colleague hands you a capacity case you have not seen and marks blind against the RCP descriptors. The sequence is deliberate — you do not face an unseen, timed, disputed case until the mechanics are automatic, or you will rehearse anxiety rather than skill.
The four-week modality ladder
Do not jump straight to full mock stations. Climb a ladder, from isolated skill to realistic simulation.
| Week | Rung | What you do |
|---|---|---|
| 1 | Isolated skill | Drill one component at a time (a capacity test; a CGA structure) until the mechanics are automatic |
| 2 | Coached case | Run a full case on one skill with a clinician giving structured feedback after each attempt |
| 3 | Timed integrated case | Combine skills under the real nine-to-ten-minute station clock, marked against the rubric |
| 4 | Unseen simulation | Full mock circuit on cases you have not seen, marked blind by a colleague |
Knowledge maintenance runs underneath the ladder: keep the KBA-level content warm with short unseen MCQ blocks so your factual base does not decay while you drill performance. That is the appropriate role for a question bank here — the foundation, not the training ground.
When AI feedback helps, when it is unreliable, and when a clinician is required
Automated feedback is useful for the mechanical and factual layer: whether you named the components of a capacity assessment, whether a drug interaction is real, whether your CGA covered the expected domains. It becomes unreliable the moment judgement, empathy or clinical nuance is involved — an AI cannot reliably score whether you communicated a deprescribing decision humanely, or whether your best-interests reasoning was defensible in context, and it can produce fluent, confident feedback that is simply wrong. Before you trust any automated score, calibrate it against a human judgement on the same performance (calibrating AI feedback). A clinician or examiner is required for anything the OSCE rubric actually rewards: communication quality, real-time synthesis, and the safety of a plan. iatroX and other MCQ tools measure the knowledge layer honestly and do not claim to score the OSCE — treat that boundary as a feature, not a shortfall.
Getting examiner-grade feedback without a formal course
Not everyone can attend a course, but everyone can build a feedback loop, and the requirement is a marker who anchors to the standard rather than merely a friendly observer. Give your practice partner the published RCP station descriptors and ask them to mark against those specific behaviours, not against a general impression — the difference between "that was fine" and "you assessed global capacity when the station asked about one decision" is the difference between useless and useful feedback. Rotate markers so you are not calibrated to one person's quirks. Record yourself, with consent, and watch the communication stations back: candidates are routinely surprised by their own pace, jargon and interruptions. Where a clinician is unavailable, a structured checklist derived from the official descriptors still beats unstructured self-assessment. The one thing you cannot do is mark your own performance in real time — you are too busy performing to observe, which is exactly why the exam uses an external examiner.
A balanced case and task matrix
Candidates rehearse what they find comfortable, which means the confident communicator practises communication stations and avoids data interpretation, and vice versa. Force balance with a matrix: list the station types (communication, history, examination/short clinical, data and prescribing, capacity, frailty/CGA, MDT/transfer) down one axis and your practice count and last-reviewed date across it. Any cell that stays empty is the station most likely to catch you. Rotate deliberately so you are not practising only familiar scenarios with familiar colleagues.
How much practice is enough?
There is no official quota of cases, and any specific number would be invented. The useful answer is a standard, not a count: practise each station type until you can perform it to time, on an unseen case, marked to the rubric by someone other than yourself — and until your performance is stable across two or three consecutive unseen attempts rather than dependent on which case you happen to draw. In practice this means every cell in your case-and-task matrix has at least a handful of distinct rehearsals behind it, weighted toward the station types you find hardest, and that your most recent attempt at each was unseen and timed. Volume for its own sake is a trap: ten reruns of the same familiar case build confidence without building skill, whereas three genuinely different, rubric-marked cases tell you far more. Stop when the standard is met on unseen cases, not when a counter reaches a round number.
Red flags
Stop and reset if you notice any of these:
- Memorised scripts — reciting a rehearsed opening rather than responding to the patient in front of you.
- Repeated cases — practising the same three scenarios until your score reflects familiarity, not skill.
- Generic feedback — "that was good" with no reference to the rubric or a specific behaviour.
- Uncalibrated scoring — marks from an assessor who has not anchored to the station standard, or from an automated tool no one has checked against a human.
- No official-rubric check — practising without ever mapping your performance to the RCP station descriptors.
Each red flag inflates your sense of readiness while leaving the real gap untouched.
FAQ
How do I know whether I have covered the full DGM blueprint? For the KBA, map your practice against the RCP February 2025 syllabus domain by domain and confirm each is covered at an adequate unseen standard. For the Clinical, coverage means the case-and-task matrix above has no empty cells — every station type practised, timed and marked against the rubric. Covering the blueprint is a per-component judgement; the knowledge blueprint and the clinical station map are different maps, and completing one says nothing about the other.
Can one question bank be enough for DGM? No. A good bank can be enough for the knowledge component if it matches the RCP blueprint and you measure on unseen items, but it cannot train or assess the four Part 2 skills in this article — frailty assessment, capacity, medicines conversations and MDT coordination — because those are performances, not recognition tasks. For the diploma as a whole, a bank is a necessary foundation and never a sufficient one.
What should I measure instead of my overall Q-bank percentage for DGM? For the KBA, measure per-domain unseen accuracy, pacing and high-confidence error rate rather than a headline percentage. For the Clinical, measure something a percentage cannot capture at all: rubric-anchored performance on unseen stations, marked by a clinician — can you run a capacity assessment or a deprescribing conversation to standard, to time, on a case you have not seen? That observed-performance judgement is the readiness signal, and no MCQ score substitutes for it.
When should I stop doing new DGM questions? Stop adding KBA questions when your knowledge coverage is even, your unseen timed accuracy is stable and your error log is quiet — and then redirect that time to the Clinical if it is in scope. Doing yet more MCQs is a comfortable way to avoid the harder, more useful work of practising stations against a rubric. The signal to stop MCQs is often the signal to start OSCE practice in earnest.
Which DGM resource should I use for my weakest component? Match the modality to the skill. Weak KBA knowledge: a blueprint-mapped bank plus unseen transfer practice. Weak on a Clinical skill: coached, rubric-anchored practice with a clinician and role-players — a course, a study group, or supervised sessions — not a question bank. iatroX is a strong option for keeping the underlying knowledge warm and measuring it on unseen items; it is explicitly not an OSCE simulator, so pair it with a genuine performance-practice source for the Clinical. The DGM resource decision tree maps the options to profiles.
Editorial notes and references
Written by Dr Kolawole Tytler, NHS GP and founder of iatroX. Last checked 21 July 2026. Exam-format facts are from the RCP DGM pages and the February 2025 syllabus; vendor figures elsewhere in this series are vendor-reported as of 21 July 2026. Disclosure: iatroX operates a competing UK question bank; its role here is confined to the knowledge and unseen-MCQ-measurement layer that underpins these skills, and it is stated plainly that iatroX does not reproduce or replace the DGM Part 2 Clinical OSCE, which requires clinician-marked practice against the RCP rubric. Corrections via the feedback route on iatrox.com. References: RCP Diploma in Geriatric Medicine and syllabus (rcp.ac.uk); BGS DGM page (bgs.org.uk); and, internally, calibrating AI feedback before you trust the score, the iatroX DGM bank and the DGM content-gap checklist.