Strong surgeons rarely fail the academic viva on clinical knowledge. They falter on delivery — the phrasing, the traps, the moment an examiner pushes back. This is the chapter candidates ask for first, given away whole: the checklists, the eleven traps with the wrong and right answer side by side, the phrases that work and the ones that cost marks, and how to hold your position under pressure.
Print this page and tick off items as you prepare. This is your final checklist before the exam.
| Timing | Task | What to Extract |
|---|---|---|
| 0–4 min | Orientation (Pass 1) | Title, abstract, study design, main result, conclusion |
| 4–16 min | Interrogation (Pass 2) | PICO, methods details, Table 1, main results table, p-values, CIs |
| 16–24 min | Construction (Pass 3) | 2–3 strengths, 2–3 limitations, clinical application |
| 24–29 min | Rehearsal (Pass 4) | Opening summary rehearsed, first two answers ready, one number chosen |
Study Basics
Methods
Results
Critique
Thirty minutes is long enough to read every word — which is exactly the trap. A candidate who reads the whole paper slowly arrives with recall but no argument. Use the four passes, and protect the last one: rehearsal is what the extra time buys you.
What to Write Down:
You’ll have paper to make notes. Write:
| Phase | Duration | What Happens |
|---|---|---|
| Opening | 2 min | Your structured summary of the paper |
| Methods | 3 min | Questions about study design, randomisation, blinding |
| Results | 4 min | Interpreting tables, statistics, clinical significance |
| Critique | 3 min | Strengths, limitations, bias |
| Application | 3 min | Would you change practice? How does it apply? |
Structure your answers using SBAR (Situation, Background, Assessment, Recommendation):
Situation: “This trial provides Oxford Level 1b evidence for perioperative chemotherapy in gastric cancer.”
Background: “The absolute survival benefit is 13%, with a number needed to treat of 8, which is clinically meaningful in a disease with poor prognosis.”
Assessment: “However, the trial population excludes elderly patients with comorbidities who make up a significant proportion of my practice. Additionally, the ECF regimen is now superseded by FLOT.”
Recommendation: “I would offer perioperative chemotherapy to fit patients with resectable gastric cancer, using a modern regimen like FLOT. For elderly or frail patients, I’d have a shared decision-making discussion about the balance of benefits versus toxicity.”
Wrong answer: “The p-value is 0.001, so this result is very important.”
Right answer: “The result is statistically significant with p=0.001, meaning it’s unlikely due to chance. However, the absolute difference is only 1%, which may not be clinically meaningful. I would need to consider cost, patient preference, and quality of life before changing practice.”
Wrong answer: “The subgroup analysis shows benefit in patients under 65, so we should only treat younger patients.”
Right answer: “The subgroup analysis suggests a trend toward benefit in younger patients, but this was not the primary analysis and may be due to chance given multiple comparisons. Subgroup analyses should be hypothesis-generating only and require validation in a dedicated trial.”
Also know the converse: a non-significant interaction does not prove the effect is uniform. These tests are usually underpowered, so it means the trial could not detect a difference — a weaker claim than showing there is none.
Wrong answer: “This is a great RCT published in a top journal, so we should adopt this immediately.”
Right answer: “While this is a well-designed RCT, there are important limitations including [specific limitations]. These affect how I would apply the results to my patient population. I would consider the trial evidence alongside patient factors and shared decision-making.”
Wrong answer: “This trial has too many flaws — selection bias, lack of blinding, heterogeneous population. The results are worthless.”
Right answer: “Every trial has limitations, and this one is no exception. However, it provides valuable Level 1 evidence on an important clinical question. While I would be cautious about [specific limitation], the overall findings are informative and should influence practice when applied thoughtfully.”
Wrong answer: “There’s selection bias because of the inclusion criteria.”
Right answer: “There’s selection bias because the trial excluded patients over 75 with significant comorbidities. This limits generalisability because these excluded patients represent a large proportion of the real-world population we operate on, and they may respond differently to treatment.”
Wrong answer: “The curve shows 40% recurrence at five years.”
Right answer: “That is a Kaplan–Meier estimate, which treats deaths from other causes as censored observations. That assumes censoring is non-informative; when competing deaths are common, Kaplan–Meier overestimates the cumulative incidence of recurrence. I would want a cumulative incidence function instead.”
Wrong answer: “The AUC is 0.89, so it is a good model.”
Right answer: “The discrimination is good — the model ranks patients well. But AUC assesses discrimination, not calibration — it tells us little about whether the predicted probabilities are numerically accurate. A model can discriminate well and still tell a patient 89% when the observed rate is 66%. I would want to see the calibration plot before quoting a number in a consent conversation.”
Wrong answer: “The confidence interval sits within the margin, so non-inferiority is demonstrated.”
Right answer: “It met the margin, but the margin is the appraisal. My first question is how it was justified — preservation of effect, clinical meaningfulness, and whether it was pre-specified. A margin that tolerates a clinically unacceptable loss makes the result meaningless however tight the interval.”
Wrong answer: “Pneumonia was significantly reduced, p equals 0.04.”
Right answer: “That is one of four comparisons. Testing four outcomes at the 5% level carries roughly a one-in-five chance of at least one false positive. I would ask whether this outcome was pre-specified as primary and whether the threshold was adjusted — a Bonferroni correction here would require p below 0.0125.”
Wrong answer: “The funnel plot shows publication bias.”
Right answer: “It depends which funnel plot. One dot per study, assessed for asymmetry, addresses publication bias. One dot per unit against control limits is institutional outcome monitoring. They share a name and nothing else, so I would check which is being presented before commenting.”
Wrong answer: “The curves cross, so proportional hazards is violated.”
Right answer: “Crossing curves are a visual clue that the assumption may be violated, not proof of it. I would confirm formally with the Schoenfeld residuals test or a log-log plot. If the assumption genuinely fails, a single hazard ratio stops being a meaningful summary of the whole follow-up.”
Examiners rarely ask what they appear to be asking. Most questions are a fork: one branch checks recall, the other checks judgement. Recognising which branch you are on is most of the work. The left column is what you hear; the middle is what is being assessed; the right is where a strong answer goes first.
| If they ask… | Primarily testing… | Go here first |
|---|---|---|
| “How was the non-inferiority margin justified?” | Whether you recognise that justifying the margin is central to interpreting the trial | Preservation of effect, clinical meaningfulness, pre-specified |
| “The effect was larger in the elderly — comment.” | Whether you reach for the interaction test or fall for subgroup p-values | “Is there a formal test for interaction?” |
| “The curves cross at 18 months.” | Whether you can name the proportional hazards assumption | Schoenfeld residuals or a log-log plot |
| “The AUC is 0.89 — is this a good model?” | Whether you distinguish discrimination from calibration | “Good ranking — how does it calibrate?” |
| “Would you change your practice?” | Applicability, patient values and shared decision-making — not recall | SBAR, then your own population and its caveats |
| “40% had recurred by five years.” | Whether you consider competing risks in an older or high-mortality cohort | Cumulative incidence, not Kaplan–Meier |
| “Pneumonia was reduced, p = 0.04.” | Whether you notice this is one of several comparisons | Pre-specified primary outcome? Adjusted threshold? |
| “What does the funnel plot show?” | Whether you recognise which funnel plot is shown before interpreting it | Studies and asymmetry, or units and control limits? |
| “Is this result significant?” | Whether you separate statistical from clinical significance | “Statistically yes; clinically it depends on…” |
| “Would these results apply to your patient?” | External validity, not internal validity | Trial population versus the patient in front of me |
| “The relative risk fell by 50%.” | Whether you ask for the absolute benefit | ARR and NNT — and the baseline risk they depend on |
| “The secondary endpoint was significant.” | Whether you notice it was not the primary outcome | What was the primary outcome, and was it met? |
| “What are the weaknesses?” | Proportion — balance, not a demolition | Two or three material limitations, then what it still shows |
Almost every question above tests the same instinct: do you accept the number you are given, or do you ask what had to be true for that number to mean anything? Candidates who perform strongly usually move beyond naming the statistical test. They consider the assumptions, limitations and clinical implications before commenting on the result itself.
Bluffing. If you don’t know, say so and add what you do know: “I’m not certain of that figure, but the trial showed a survival benefit — I’d want to check the confidence interval.” Honest uncertainty costs far less than invention.
Professional Demeanour
“I’m not familiar with that specific statistical test, but my understanding is that it’s used for [general category]. Could you clarify what aspect you’d like me to discuss?”
or
“Based on what I can see in the methods section, my reading would be [interpretation], though I would want to check that before relying on it.”
Knowing the content is not the same as delivering it. Three habits separate a candidate who knows the material from one who sounds like they do.
Examiners are listening for whether you can identify what matters. Give the answer in your first sentence, then support it. “This is a non-inferiority trial and its conclusion is that sentinel node biopsy alone was not worse — I’d want to check the margin and whether the population matches mine” tells the examiner more in ten seconds than a minute of describing the methods before reaching a view.
The recognition drills in this book are built for exactly this. What you see, what it means, key pattern — then the opener. Practise saying the opener first, not last.
Bluffing is the one unrecoverable error: the examiner is a specialist and will see through it. But bare uncertainty wastes the question. Acknowledge the limit, then pivot to reasoning you can defend.
“I’m not certain of that specific figure. What I can say is that the confidence interval crosses one, so the trial has not demonstrated a difference — and I’d want to know whether it was powered to detect the difference that would matter clinically.”
That answer concedes nothing you know and demonstrates the reasoning the question was testing. It is almost always worth more than a guess that happens to be right.
Repeated challenge is frequently a test of whether you will abandon a correct answer under pressure rather than a signal that you are wrong. Decide which is happening and say so.
If you can still defend your position: “I take the point, but I’d still say the analysis is complete-case rather than intention-to-treat, because 52 patients with no outcome data were excluded.” If they have shown you something you had missed: “That changes my answer — I hadn’t accounted for the imputation, and with it the analysis is intention-to-treat after all.”
Changing your mind for a reason is a strength. Changing it because someone frowned is not.
A two-second pause before answering reads as consideration, not hesitation. Rushing to fill silence is how candidates talk themselves into claims they cannot defend.
You now have the phrases, the traps and the fork. What sits underneath them is the part that lets you produce them about a paper you have never seen: the four-pass reading method, the statistics from first principles, sixteen chart types and how to read each in five seconds, and worked model critiques of CROSS, MAGIC, Z0011 and the Dutch TME trial — plus a companion set of 100 flashcards, free on this site.
Get the Complete Handbook →amac-medical.com · Think Like an Examiner · Critique with Confidence
Examination revision aid only — not a clinical guideline. Evidence cut-off June 2026. Not affiliated with, endorsed by, or connected to the Royal Colleges of Surgeons, the Intercollegiate Surgical Curriculum Programme, or any examining body.