📚 FRCS Viva Flashcards

The absolute difference in event rates between control and experimental groups.

Example: If control death rate = 20% and treatment death rate = 15%:

ARR = 20% - 15% = 5%

The proportional reduction in risk compared to the control group.

Example: ARR = 5%, CER = 20%:

RRR = 5/20 = 25%

How many patients need treatment to prevent one adverse event.

Example: ARR = 5% = 0.05:

NNT = 1/0.05 = 20

Meaning: Treat 20 patients to prevent 1 death

p-value = the probability of observing a result at least as extreme as this one, if the null hypothesis were true.

It is not "the probability the result is due to chance," nor the probability the null hypothesis is true — that is the classic viva trap.

• p < 0.05: Statistically significant (reject null hypothesis)
• p < 0.001: Very strong evidence against the null
• p > 0.05: Not significant (fail to reject null)

CI = Range within which the true effect likely lies (95% confident)

For differences:

• CI crosses 0? → Not significant
• CI doesn't cross 0? → Significant

For ratios (OR, RR, HR):

• CI crosses 1.0? → Not significant
• CI doesn't cross 1.0? → Significant

Ability of a test to correctly identify people WHO HAVE the disease.

Think: SnNout - High Sensitivity rules Negative result out

Ability of a test to correctly identify people WHO DON'T HAVE the disease.

Think: SpPin - High Specificity rules positive result in

Type I Error (α): False positive

→ Saying there's an effect when there isn't (rejecting true null hypothesis)

→ Set by p-value threshold (usually 0.05 = 5% risk)

Type II Error (β): False negative

→ Missing a real effect (failing to reject false null hypothesis)

→ Related to statistical power (Power = 1 - β)

Power = Ability to detect a real effect if it exists

Power = 1 - β (Type II error)

Standard target: 80% power

Increased by:

• Larger sample size
• Larger effect size
• Less variability in data

Interpretation:

• RR = 1: No difference
• RR < 1: Protective effect
• RR > 1: Increased risk

Example: RR = 0.75 means 25% reduction in risk

From 2×2 table: Odds of exposure in cases vs controls

Interpretation (same direction as RR):

• OR = 1: No association
• OR < 1: Protective
• OR > 1: Risk factor

HR = the ratio of the hazard (event rate at a given instant) in one group vs the other, across the whole follow-up.

Used in survival analysis (time-to-event data)

Interpretation:

• HR = 0.66: 34% reduction in hazard of death
• HR = 1: No difference
• HR = 1.5: 50% increase in hazard

Comparing 2 independent groups, continuous data:

Independent t-test:

• ✓ Data normally distributed
• ✓ Similar variances

Mann-Whitney U test:

• ✓ Non-normal distribution
• ✓ Skewed data
• ✓ Ordinal data

Comparing 2 categorical variables:

Chi-square test:

• ✓ Large sample (all expected cell counts > 5)
• ✓ Quick calculation

Fisher's exact test:

• ✓ Small sample (any cell < 5)
• ✓ More accurate for small numbers

Analysis of Variance (ANOVA)

Use: Comparing means of 3+ groups

Requirements:

• Continuous data
• Normal distribution
• Independent groups

If significant → Post-hoc tests to see which specific pairs differ

Linear Regression:

• Continuous outcome (e.g., length of stay)
• Predicts actual values

Logistic Regression:

• Binary outcome (yes/no, e.g., complication)
• Predicts probability (0-1)
• Results = Odds Ratios

Cox Proportional Hazards Regression

Use: Time-to-event data (survival analysis)

Outcome: Hazard Ratios (HR) for each predictor

Example: Identifying independent predictors of recurrence-free survival while adjusting for age, stage, grade

A result can have p<0.05 but be too small to matter in practice.

Example: New drug reduces operative time by 2 minutes

• ✓ p = 0.001 (statistically significant)
• ✗ But 2 minutes is clinically trivial

ITT: Analyze patients in the groups they were randomized to, regardless of what actually happened

Why?

• Preserves randomization
• Reflects real-world practice
• Prevents bias

Example: Patient randomized to surgery but refused → still counted in surgery group

I² = Measure of how different the studies are

Interpretation:

• I² < 25%: Low (studies consistent)
• I² 25-50%: Moderate
• I² > 50%: Substantial (explore causes before pooling)

Prevalence: Proportion with disease AT ONE TIME POINT

• Snapshot
• Example: 10% of population has diabetes

Incidence: NEW cases over a TIME PERIOD

• Rate
• Example: 50 new diabetes cases per 1000 people per year

Correlation: Two variables are associated/related

Causation: One variable CAUSES the other

Classic example: Ice cream sales correlate with drowning deaths

• ✓ Correlated
• ✗ Ice cream doesn't cause drowning
• ✓ Confounder = summer weather

The diamond at the bottom of a forest plot shows the pooled result from all studies combined.

Interpretation:

• Width = confidence interval
• Center = pooled estimate
• Crosses the line of no effect? = Not significant
• Doesn't cross it? = Significant

How many patients need to be exposed to cause one additional harm

Example: New drug increases bleeding by 3%

NNH = 1/0.03 = 33

Meaning: Treat 33 patients to cause 1 additional bleed

Standard Deviation (SD):

• Describes spread of DATA
• How much individual values vary

Standard Error (SE):

• Describes uncertainty of the MEAN
• How precise our estimate is
• SE = SD / √n

When: Sample doesn't represent target population

Example: Trial only enrolls fit, healthy patients but results applied to elderly with comorbidities

Prevention:

• Broad inclusion criteria
• Consecutive enrollment
• Document who was excluded and why

When: Groups treated differently beyond the intervention

Example: Intervention group gets more clinic visits, closer monitoring, extra support

Prevention:

• Blinding of participants and clinicians
• Standardized protocols
• Equal attention to both groups

When: Outcomes measured differently between groups

Example: Assessor knows treatment allocation and "looks harder" for complications in one group

Prevention:

• Blinding of outcome assessors
• Objective outcome measures
• Standardized assessment protocols

When: Systematic differences in dropouts/withdrawals

Example: Sicker patients preferentially drop out from treatment group because of side effects

Prevention:

• Minimize dropout (good follow-up)
• Intention-to-treat analysis
• Report reasons for withdrawal

When: Participants remember past exposures differently

Example: Patients with cancer remember risk factors (smoking, diet) better than healthy controls

Most common in: Case-control studies

Prevention:

• Use objective records
• Prospective data collection

When: Positive studies more likely to be published than negative ones

Result: Meta-analyses overestimate treatment effects

Detection: Funnel plot asymmetry

Prevention:

• Search trial registries
• Contact authors for unpublished data
• Grey literature searches

Definition: Variable associated with BOTH exposure AND outcome that distorts the apparent relationship

Classic Example: Coffee → Lung cancer?

• ✗ Coffee doesn't cause cancer
• ✓ Smoking is the confounder
• Smokers drink more coffee AND get cancer

At Design Stage:

1. Randomization (RCTs) - best method
2. Matching - match cases/controls on confounders
3. Restriction - only include homogeneous group

Example of restriction: Only include non-smokers to eliminate smoking as confounder

At Analysis Stage:

1. Stratification - analyze subgroups separately
2. Multivariable regression - adjust for multiple confounders
3. Propensity score matching - advanced statistical technique

When: Early detection appears to improve survival without actually changing death

Example: Screening detects cancer 2 years earlier

• Without screening: Diagnosis age 60 → Death age 65 (5-yr survival)
• With screening: Diagnosis age 58 → Death age 65 (7-yr survival)
• ✗ Patient still dies age 65!

Solution: Use mortality, not survival from diagnosis

When: Screening preferentially detects slow-growing cancers

Why: Aggressive cancers progress quickly and present between screens

Result: Screened cancers have better prognosis (but only because we caught the slow ones)

Solution: Randomized screening trials

When: Assessor's expectations influence measurements

Example: Surgeon who performed novel technique assesses outcomes

Prevention:

• Independent outcome assessment
• Blinding of assessors
• Objective outcome measures

When: People change behavior because they know they're being observed

Example: Patients in trial comply better with medication than real-world patients

Result: Trial overestimates real-world effectiveness

Solution: Pragmatic trial design, minimal monitoring

When: Selective reporting of favorable outcomes

Example: Trial pre-specifies primary outcome A, but reports secondary outcome B because it was significant

Prevention:

• Pre-registration of protocols
• Report all pre-specified outcomes
• Check trial registry

Randomization:

• Prevents selection bias
• Balances known/unknown confounders
• Creates comparable groups

Blinding:

• Prevents performance & detection bias
• Prevents differential treatment
• Prevents biased assessment

Single-blind: Patient doesn't know (doctor does)

Double-blind: Patient AND doctor don't know

Triple-blind: Patient, doctor, AND outcome assessor don't know

Impossible to blind:

• Surgeon doing the operation
• Patient if different incisions
• Patient if awake

CAN still blind:

• Outcome assessors
• Data analysts
• Sometimes patients (if under GA, similar wounds)

When: Treatment decision is based on prognosis

Example: Sicker patients get more aggressive treatment

• Compare laparoscopic vs open surgery
• Surgeons do open on sicker patients
• Open appears to have worse outcomes
• But really = sicker patients!

Internal Validity: Are results true for THIS study population?

• Threatened by bias
• Threatened by confounding

External Validity (Generalizability): Do results apply to OTHER populations?

• Threatened by selection bias
• Depends on how representative sample is

When: Errors in measuring/classifying variables

Examples:

• Misclas sification of disease status
• Inaccurate measurement tools
• Different criteria used in groups

Types: Recall bias, observer bias, measurement error

Direction: Forward (Exposure → Outcome)

Setup:

1. Identify exposed vs unexposed groups
2. Follow forward in time
3. Compare incidence of outcomes

Measure: Relative Risk

Example: Follow smokers vs non-smokers → lung cancer

Direction: Backward (Outcome → Exposure)

Setup:

1. Identify cases (with outcome)
2. Select controls (without outcome)
3. Look back at exposures

Measure: Odds Ratio (NOT relative risk!)

Example: Lung cancer patients vs controls → smoking history

Best for RARE OUTCOMES

Why: You start with cases that already have the rare outcome

Other advantages:

• Quick
• Inexpensive
• Small sample needed
• Multiple risk factors at once

Main problem: Recall bias

Other limitations:

• Cannot calculate incidence/prevalence
• Difficult to prove temporal relationship
• Selection of controls is challenging
• Can only study one outcome

1. Random allocation to intervention or control
2. Concurrent control group for comparison
3. Prospective follow-up to measure outcomes

Gold standard because:

• Randomization balances confounders
• Establishes causation
• Minimizes bias

Prospective Cohort:

• Identify groups NOW
• Follow forward into future
• Higher quality data

Retrospective Cohort:

• Look back at historical data
• Use existing records
• Faster, cheaper
• Lower quality data

Design: Snapshot at one time point

Measures: Prevalence (not incidence)

Example: Survey of current smoking rates

Advantage: Quick, cheap

Limitation: Cannot establish temporal relationship or causation

Systematic Review:

• Comprehensive search
• Quality assessment
• Narrative synthesis
• May or may not pool data

Meta-Analysis:

• Statistical pooling of results
• Calculates combined estimate
• Shown on forest plot

Don't pool if:

• High heterogeneity (I² > 50%)
• Different populations
• Different interventions
• Different outcomes measured
• Methodological issues

P = Population/Patient

I = Intervention

C = Comparison/Control

O = Outcome

Example: In patients with gastric cancer (P), does perioperative chemotherapy (I) vs surgery alone (C) improve survival (O)?

Method: Randomize within subgroups to ensure balance of important prognostic factors

Example: Stratify by cancer stage

• Stage I patients randomized
• Stage II patients randomized
• Stage III patients randomized

Result: Equal stage distribution in both groups

Method: Randomize in blocks to ensure equal group sizes

Example: Block of 4

• Block 1: AABB
• Block 2: BABA
• Block 3: ABAB

Result: Balanced allocation throughout trial

Method: Randomize groups (hospitals, wards) not individuals

When used: Intervention is at group level

Example: Surgical safety checklist implementation

• Can't randomize individual patients
• Randomize hospitals

Analysis: Need to account for clustering

Definition: Keeping randomization sequence hidden until patient enrolled

Prevents: Selection bias

How:

• Central randomization (phone/web)
• Sequentially numbered sealed envelopes
• Third-party allocation

Intention-to-Treat (ITT):

• Analyze all as randomized
• Includes dropouts, crossovers
• Reflects real-world
• Preferred

Per-Protocol:

• Only those who completed
• Excludes non-compliant
• Overestimates benefit

Level 1: Systematic review of RCTs (1a) or a single high-quality RCT (1b)

Level 2: Cohort study, or a lower-quality / underpowered RCT (2a/2b)

Level 3: Case-control study

Level 4: Case series

Level 5: Expert opinion

Goal: Show new treatment is "not worse than" standard (within acceptable margin)

Example: New operation is less invasive - want to show outcomes not significantly worse

Pre-specify: Non-inferiority margin (acceptable difference)

Definition: Combining multiple outcomes into one endpoint

Example: MACE (Major Adverse Cardiac Events)

• Death OR MI OR stroke
• Patient counted if ANY occur

Advantage: Increases power, fewer patients needed

Problem: Mixing important (death) with less important outcomes

Design: Real-world conditions, broad inclusion, minimal intervention

vs Explanatory trial: Ideal conditions, strict criteria, close monitoring

Pragmatic advantages:

• Better external validity
• More generalizable
• Reflects actual practice

Definition: Planned analysis before trial completion

Reasons:

• Early benefit detected → stop trial early
• Clear harm detected → stop trial
• Futility (no chance of benefit) → stop

Important: Must be pre-planned, use adjusted significance levels

The 4-Sentence Opening:

1. "This is a [study design] published in [journal] in [year]"
2. "The PICO is: [state PICO clearly]"
3. "The main finding was [primary outcome with numbers]"
4. "This provides Level [X] evidence for/against [intervention]"

❌ "Age causes complications"

✅ Say: "Age correlates with complications"

❌ "p=0.05 so it's clinically important"

✅ Say: "While statistically significant, we must consider clinical significance"

❌ "The paper says..."

✅ Say: "The authors found/reported..."

❌ "I don't know"

✅ Say: "Based on what I can see in the methods..."

S = Situation (what you see)

"This is a... The main finding is..."

B = Background (context)

"This is important because current practice is..."

A = Assessment (your critique)

"The strengths are... However, I'm concerned about..."

R = Recommendation (conclusion)

"This suggests we should/shouldn't..."

Good viva response:

"I would check if randomization was appropriate by examining Table 1 for baseline balance. The groups should be similar for age, sex, comorbidities, and disease severity. If there are significant differences, randomization may have failed or the sample was too small to achieve balance. I'd also check if allocation was concealed and if the randomization method is described."

Professional response:

"True double-blinding is challenging in surgical trials because the operating surgeon necessarily knows the procedure performed. However, we can still minimize bias by blinding the outcome assessors, data analysts, and potentially the patients if incisions are similar and they're under general anaesthetic. The key is documented, standardized outcome assessment by independent, blinded assessors."

Where to look:

1. Abstract - usually stated clearly
2. Methods section - "Primary endpoint was..."
3. Sample size calculation - based on primary outcome
4. Results - reported first

Viva response structure:

"While the result is statistically significant (p=[value]), I need to consider clinical significance. The absolute difference is [ARR], meaning [interpretation]. The NNT is [value], which means treating [X] patients prevents one event. Given the [risks/costs/burden] of treatment, I would consider this [clinically meaningful / clinically trivial]."

Good answer:

"I would check if a sample size calculation was performed. The authors should report the assumed effect size, significance level (usually 0.05), power (usually 80-90%), and expected dropout rate. If the calculated sample wasn't achieved, the study may be underpowered to detect a real effect, increasing risk of type II error."

Template response:

"The generalizability depends on how representative this sample is. I note the patients were [describe population]. In my practice, patients tend to be [older/younger/sicker/healthier]. The [strict inclusion criteria / single center / specialized setting] may limit external validity. However, the internal validity appears sound, so results likely apply to similar populations."

Always prepared to give:

2 Strengths:

• Start positive ("The main strength is...")
• Be specific (RCT design, large sample, good follow-up)

2 Weaknesses:

• Use diplomatic language ("However, I'm concerned about...")
• Be specific (selection bias, lack of blinding, short follow-up)

Professional way to mention:

"I note the study was funded by [pharmaceutical company / device manufacturer]. While this doesn't automatically invalidate results, it does warrant careful scrutiny of the methods and consideration of whether outcomes favor the sponsor. I would look for independent replication and check if protocol was pre-registered."

6-Step Description:

1. "This Kaplan-Meier curve shows..."
2. "There are [N] groups: [describe]"
3. "The curves [do/don't] separate early, suggesting..."
4. "At [time point], the survival is [%] vs [%]"
5. "The hazard ratio is [value] with CI [range], p=[value]"
6. "This represents a [X]% relative reduction in mortality"

Good response:

"The wide confidence interval indicates imprecision in the estimate. This usually reflects a small sample size, high variability, or few events. While the point estimate suggests [effect], the CI ranges from [lower] to [upper], meaning the true effect could be anywhere in this range. I would interpret this result cautiously and consider whether additional data is needed."

Structured answer:

"The loss to follow-up was [X]%. Generally, <5% is acceptable, 5-20% introduces potential bias, and >20% threatens validity. I'm concerned because: (1) dropout was differential between groups [[%] vs [%]], and (2) reasons suggest sicker patients may have dropped out. This could introduce attrition bias. Ideally, I'd want sensitivity analysis assuming worst-case scenarios."

0-2 min: Quick scan - title, abstract, figures

2-7 min: Deep dive - methods (PICO, design, randomization), Table 1 (baseline), main results table

7-10 min: Prepare critique - identify 2 strengths, 2 weaknesses, calculate NNT if possible, note discussion points

Never say "I don't know" - instead:

• "Based on what I see in the methods section..."
• "The paper doesn't specify this, but typically..."
• "I would need to examine [X] to answer that fully..."
• "My understanding is that... but I'd want to verify..."

Critical points to make:

"While the subgroup analysis suggests [finding], I'm cautious because: (1) it appears post-hoc rather than pre-specified, (2) multiple subgroups increase risk of false positives, and (3) each subgroup is underpowered. Subgroup findings should be hypothesis-generating only and require validation in dedicated trials."

The complete answer:

"The control event rate was [X]% and the experimental rate was [Y]%, giving an absolute risk reduction of [ARR]%. This represents a relative risk reduction of [RRR]%. The number needed to treat is [NNT], meaning we need to treat [N] patients to prevent one [outcome]. Given the [burden/cost/risks] of treatment, I consider this [clinically significant/marginal]."

The Perfect 60-Second Pitch:

"This [design] of [N] patients with [condition] compared [intervention] to [control] and found [primary outcome with numbers]. The [ARR/RRR/NNT] was [values]. Main strengths are [2 strengths]. However, I'm concerned about [2 limitations]. This provides Level [X] evidence that in [population], [intervention] [does/doesn't] improve [outcome], and I [would/wouldn't] change practice based on this."

Balanced response template:

"This is Level [X] evidence showing [finding]. Before changing practice I would consider: (1) Are these results internally valid? (2) Do they apply to my patients? (3) Are benefits clinically meaningful given risks/costs? (4) Has this been replicated? Based on this single study, I would [await further evidence / discuss with MDT / consider in selected patients]."

Trap 1: Confusing correlation with causation

Trap 2: Saying "relative risk" for case-control

Trap 3: Ignoring clinical vs statistical significance

Trap 4: Not checking baseline balance (Table 1)

Trap 5: Assuming subgroup analysis is valid

Professional stalling phrases:

• "Let me just check the methods section..."
• "That's a good question. Looking at Table 1..."
• "I'd approach this systematically..."
• "Before answering, I want to clarify..."
• "Let me think through the PICO..."

Professional uncertainty phrases:

• "The paper doesn't clearly specify..."
• "My understanding is... but I'd want to verify..."
• "Based on standard practice, I would expect..."
• "I can't be certain from what's presented, but..."
• "I would need to see [X] to answer definitively..."

The 8-Point Critique:

1. Study design appropriate?
2. Sample size adequate?
3. Baseline balance (Table 1)?
4. Primary outcome clearly defined?
5. Blinding/bias minimization?
6. Analysis appropriate (ITT)?
7. Results clinically meaningful?
8. Generalizable to my patients?

The bridge phrase:

"So statistically, [summarize numbers]. Clinically, this means for my patients, [translate to practice]. I would [counseling approach / treatment modification / shared decision-making]. The key message for the patient is [simple language explanation]."

Complete forest plot answer:

"This forest plot meta-analysis of [N] studies shows [intervention] compared to [control] for [outcome]. Each square represents a study, with size reflecting weight. The diamond shows the pooled estimate of [value] with 95% CI [range]. This [does/doesn't] cross the line of no effect (1.0 for ratio measures, 0 for mean differences), indicating [significance]. The I² of [X]% suggests [low/moderate/high] heterogeneity."

Professional responses:

• Stay calm and courteous
• Don't get defensive
• "That's a fair point. Let me reconsider..."
• "I see your concern. Looking at it differently..."
• Stick to what you know
• Don't make things up

Before entering the room:

1. □ Deep breath, relax shoulders
2. □ Smile, make eye contact
3. □ Have pen & paper ready
4. □ Remember: PICO first
5. □ Use 10-min reading time wisely
6. □ 2 strengths, 2 weaknesses prepared
7. □ Speak slowly and clearly

If you realize you're wrong:

"Actually, let me correct that. On reflection..." OR "I apologize, I misspoke. What I meant to say was..."

If examiner corrects you:

"Thank you for clarifying. So if [restate correctly]..."

In the last 90 seconds:

• Stand or sit tall with a relaxed, upright posture
• Slow, controlled breathing to steady your nerves
• Remind yourself: "I've prepared. I know this."
• A shoulder roll or short walk to release tension

Final 30 seconds template:

"In summary, this [study design] of [N] patients shows [main finding]. The [ARR/NNT] is [value], which I consider [clinically significant/not significant] given [rationale]. The main limitation is [key weakness]. This provides Level [X] evidence, and based on this, I would [clinical action]."

Honest fallback:

"I'm not familiar with this specific statistical method. However, I can see the p-value is [X] and confidence interval is [Y], suggesting [interpretation]. The clinical finding appears to be [describe]. I would discuss this with a statistician to fully understand the analysis."

Comparison response:

"Multi-center trials have better external validity because they include diverse populations, surgical techniques, and settings. However, single-center studies often have more consistent protocols and data quality. This [multi/single]-center design means results are [more/less] generalizable but [more/less] internally consistent."

Professional transitions:

• "Moving to the methods..."
• "Looking at the baseline characteristics..."
• "Turning to the results..."
• "In terms of clinical applicability..."
• "From a methodological standpoint..."
• "Considering the limitations..."