Evidence-Based Medicine and Critical Appraisal — Viva

Long Case Viva Defence

Candidate's opening statement (model answer)

"Mrs K is a 75-year-old woman with non-valvular atrial fibrillation (CHA2DS2-VASc 5), stage 3b chronic kidney disease (eGFR 32), and a recent mechanical fall with a fractured wrist, who presents a classic evidence-based medicine problem: she is at high risk of ischaemic stroke (her CHA2DS2-VASc gives an annual risk of 6 to 7 per cent), but she also has an elevated bleeding risk (HAS-BLED 4) and she sits near the renal exclusion threshold of the pivotal direct oral anticoagulant trials. [1]

Her main problems are:

1. Non-valvular atrial fibrillation with a high annual stroke risk (CHA2DS2-VASc 5), the primary indication for the decision.
2. Stage 3b chronic kidney disease (eGFR 32), which raises the applicability question because the pivotal DOAC trials largely excluded eGFR below 25 to 30, and which affects the dosing and the bleeding risk.
3. A recent mechanical fall with a fractured wrist, which raises the traumatic bleeding risk and the fall-related hesitation about anticoagulation.
4. An elevated bleeding risk (HAS-BLED 4) that must be weighed against the stroke benefit in absolute terms.
5. The need for a shared decision that integrates the evidence, the clinical judgement, and her values. [1]

My approach is to formulate the focused PICO question, to appraise the applicability of the pivotal DOAC evidence to Mrs K with particular attention to the renal and fall caveats, to compute the absolute benefit (NNT for stroke) and harm (NNH for major bleeding) for her baseline risk, to state the GRADE strength and quality of the recommendation, to share the decision with her in plain language using natural frequencies, and to document the reasoning and the monitoring plan. The overarching principle is that evidence informs the estimate of effect, clinical expertise adjusts it for the individual, and the patient's values shape the final decision." [1]

Examiner probing questions and model answers

Q1: "Walk me through how you would apply the evidence for apixaban to this patient, given that she sits near the renal exclusion threshold." [1]

"I would apply the three-step EBM process of appraisal, application and shared decision [1]. First, the appraisal. The pivotal apixaban trial demonstrated a reduction in stroke or systemic embolism and a lower rate of major bleeding (including intracranial haemorrhage) compared with warfarin. The trial excluded patients with an eGFR below 25 mL per minute; Mrs K at 32 is just inside the threshold, so the trial's renal subgroup data apply. The relative risk reduction for stroke is consistent across the renal and elderly subgroups, so the relative effect can be applied; what changes is the absolute benefit, because her baseline risk is high.

Second, the application. The CHA2DS2-VASc of 5 gives an annual stroke risk of approximately 6 to 7 per cent without anticoagulation. Apixaban reduces stroke by approximately 80 per cent versus no therapy, so the ARR is approximately 4.8 per cent per year and the NNT is approximately 21. Against this, her major bleeding risk on apixaban is elevated by the CKD and the fall, to approximately 2 to 3 per cent per year, with an NNH for a major bleed of approximately 33 to 50. The balance favours anticoagulation, and apixaban is preferred over warfarin because its intracranial haemorrhage rate is lower, which matters given the fall. The dose must be checked against the three renal criteria (age 80 or above, weight 60 kg or below, creatinine 133 micromol per litre or above); if any two are present, the dose is 2.5 mg twice daily. [1]

Third, the shared decision. I present the numbers in natural frequencies — without treatment, about 6 or 7 in 100 like her would have a stroke this year; with apixaban, about 1 or 2 in 100; so for every 21 treated, one stroke is prevented, against a major bleed for every 33 to 50 treated. I invite her to weigh the benefit and the harm against her own values." [1]

Q2: "Her son tells you he has read on the internet that blood thinners are dangerous in older people who fall. How do you address this?" [1]

"I would address the concern directly and with the evidence. The presumed safety concern about anticoagulation in fallers is not supported by the data: a patient who falls has the same elevated stroke risk as a patient who does not, and the stroke-prevention benefit of anticoagulation persists. The traumatic intracranial haemorrhage risk added by anticoagulation in a faller is real but modest, and the net balance (stroke prevented versus bleed caused) remains favourable in a high-CHA2DS2-VASc patient like Mrs K. The choice of apixaban over warfarin is itself a response to the fall concern, because apixaban's intracranial haemorrhage rate is lower. I would explain this to the son in plain terms, I would invite Mrs K (the patient, whose decision it is) to weigh the numbers, and I would document the conversation. If the fall risk is modifiable — a physiotherapy referral, a home-safety assessment, a medication review for sedating drugs — I would address those in parallel, because reducing the fall risk reduces the bleeding risk and makes the decision easier." [1]

Q3: "How would you weigh the GRADE strength and quality of the recommendation for apixaban in this patient?" [1]

"The GRADE rating for apixaban versus warfarin or no therapy for stroke prevention in atrial fibrillation is a strong recommendation on high-quality evidence [2]. The evidence starts at high quality (large, well-conducted randomised controlled trials), and there is no serious risk of bias, no serious inconsistency, no serious imprecision. There is a mild indirectness concern because Mrs K sits near the renal exclusion threshold and has a fall history not specifically captured, but the relative effect is consistent across the renal and elderly subgroups, so the indirectness does not downgrade the quality below high. The recommendation is strong because the desirable effect (a disabling stroke prevented, NNT 21) clearly outweighs the undesirable effect (a major bleed, NNH 33 to 50, with apixaban's lower intracranial haemorrhage rate than warfarin). The strength means I can offer anticoagulation as the default, while sharing the decision because of the individual considerations."

Q4: "Suppose Mrs K declines the anticoagulation after the conversation. What is your response?" [1]

"I respect her decision, provided she has capacity for it and the decision is informed. I confirm her understanding by asking her to paraphrase the benefit and the harm in her own words. I document the capacity assessment (she understands the stroke risk, the benefit of apixaban, the bleeding risk, and the consequences of declining; she retains the information, weighs it without compulsion, and communicates a clear decision), the conversation, the numbers I presented, and her reasoning. I continue the doctor-patient relationship — declining one treatment does not mean declining care. I optimise the modifiable risk factors (blood-pressure control, diabetes control, fall reduction), I offer to revisit the decision at any time, and I schedule a review. The principle is that EBM integrates evidence with patient values; a patient who has weighed the evidence and chosen otherwise has exercised her autonomy, and my role is to support her." [1]

Q5: "What if a colleague on the ward round cites a small observational study suggesting DOACs are harmful in CKD and refuses to anticoagulate? How do you respond?" [1]

"I would respond by placing the evidence in the hierarchy. A small observational study sits far below the large randomised controlled trials in the hierarchy of evidence, and observational studies of harm are subject to residual confounding (sicker patients, or patients with more advanced CKD, may have been preferentially given a DOAC, producing a spurious harm signal). I would ask whether the study adjusted for the confounding, whether the finding is biologically plausible and consistent with the trial subgroup data (which show consistent stroke prevention across the renal spectrum), and whether it has been replicated. I would present the trial evidence and its GRADE rating, I would compute the NNT and NNH for the patient in front of us, and I would make a recommendation grounded in the best available evidence while acknowledging the uncertainty. If the disagreement persists, I would seek a senior opinion. The discipline of EBM is to weigh evidence by its quality and its applicability, not by its convenience." [1]

Short Case Discussion

Forest-plot interpretation and QUADAS-2 appraisal

Examiner instruction: "You are shown a forest plot from a meta-analysis of a new antiplatelet agent versus clopidogrel for secondary prevention after myocardial infarction, and a diagnostic accuracy study of a high-sensitivity troponin assay for suspected acute coronary syndrome. Interpret the forest plot and appraise the diagnostic study." [1]

Candidate's model answer — forest plot: [1]

"I read the forest plot in the following order. First, the title and the outcome: this is the pooled estimate for the composite of cardiovascular death, myocardial infarction or stroke, and the measure is an odds ratio, so the line of no effect is at 1, with benefit (favouring the new agent) to the left of 1. Second, each row is one trial: the square is the trial's point estimate (the size of the square is proportional to the trial's weight in the pooled analysis), and the horizontal line is the trial's 95 per cent confidence interval. Third, the pooled estimate is the diamond at the bottom; the width of the diamond is the 95 per cent confidence interval of the pooled result. Here the diamond sits to the left of 1 and does not cross the line of no effect, so the pooled result is statistically significant at the 5 per cent level — the new agent reduces the composite outcome. Fourth, the heterogeneity: the I-squared is reported below the plot; if it is below 25 per cent the trials are homogeneous and a fixed-effects model is appropriate, if it is above 50 per cent the trials are heterogeneous and a random-effects model with subgroup exploration is required [7]. I would also check for funnel-plot asymmetry, which would suggest publication bias and an inflated estimate. The bottom line is that the new agent appears beneficial, but the magnitude of the absolute benefit (the NNT) must be sought separately, and the result must be interpreted in the light of the heterogeneity and any publication bias."

Candidate's model answer — QUADAS-2 appraisal: [1]

"For the diagnostic accuracy study of high-sensitivity troponin, I would apply the QUADAS-2 tool, which assesses risk of bias and applicability across four domains [5]. First, patient selection: were the patients a consecutive or random sample representative of the population in whom the test will be used (the emergency department population with suspected acute coronary syndrome), or were they selected in a way that could introduce spectrum bias (for example, only admitted patients, or only those with a clear ECG change)? Second, the index test: was the troponin assay interpreted without knowledge of the reference standard (blinding), and was the threshold pre-specified or derived from the same data (which would inflate the accuracy)? Third, the reference standard: was the final diagnosis (the adjudicated myocardial infarction, based on serial troponin, the ECG and the imaging) applied independently and blinded to the index test, and is it likely to correctly classify the target condition? Fourth, flow and timing: was the reference standard applied to all patients (no partial or differential verification bias), and was the interval between the index test and the reference standard appropriate (not so long that the condition changed)?

For each domain I judge the risk of bias as low, high or unclear, and for the first three I also judge the applicability (does the study answer my clinical question?). The sensitivity and specificity are then reported with their confidence intervals, and I apply the likelihood ratios to the patient's pre-test probability to compute the post-test probability. The key exam point is that sensitivity and specificity are properties of the test in a defined population, and a study with spectrum bias (a selected population) will over-estimate both, so the applicability judgement is as important as the risk-of-bias judgement." [1]

Examiner follow-up: "The diagnostic study reports a sensitivity of 99 per cent and a specificity of 80 per cent. A patient in your emergency department has a pre-test probability of 10 per cent and a positive troponin. What is the post-test probability, and what does it mean clinically?" [1]

"I apply the likelihood ratio. The positive likelihood ratio is sensitivity divided by (1 minus specificity), which is 0.99 divided by 0.20, equal to approximately 4.95 — call it 5. A likelihood ratio of 5 raises the probability of disease from 10 per cent (the pre-test probability) to approximately 35 per cent (the post-test probability), using the likelihood-ratio nomogram or the Bayesian conversion. So a positive troponin in a low-pre-test-probability patient raises the probability to about one in three — enough to warrant further investigation (serial troponins, observation, and imaging) but not enough to diagnose myocardial infarction in isolation. The clinical message is that a positive test in a low-risk patient is far less informative than a positive test in a high-risk patient, because the post-test probability depends on the pre-test probability. This is why the diagnostic pathway for suspected acute coronary syndrome integrates the pre-test probability (the history, the risk factors, the ECG) with the troponin, rather than treating the troponin in isolation. The negative likelihood ratio here would be (1 minus sensitivity) divided by specificity, which is 0.01 divided by 0.80, equal to 0.0125 — a very low value, meaning a negative troponin usefully lowers the probability of infarction, which is why a negative high-sensitivity troponin at the appropriate interval is such a powerful rule-out test." [1]