What the MCCQE tests in epidemiology and biostatistics
Epidemiology and biostatistics sit inside the population health part of the MCC Examination Objectives. The exam is not looking for a statistician. It wants a physician who can read a study, judge whether a test result changes the plan, explain risk to a patient in plain numbers, and decide whether a screening program does more good than harm. Questions are short and usually have one clear correct answer if you set up the numbers carefully.
Most items fall into four groups. Calculation questions give you a 2x2 table or a trial result and ask for sensitivity, predictive value, relative risk, odds ratio, absolute risk reduction or number needed to treat. Design questions describe a research goal and ask which study type fits, or describe a study and ask what type it is. Bias and validity questions describe a flaw and ask you to name it or fix it. Application questions ask what a result means for the patient in front of you.
In the blueprint these questions mostly sit in the health promotion and illness prevention dimension, with physician activities of assessment and communication. A question about explaining a screening result to an anxious patient is still an epidemiology question: you need to know that a positive result in a low-risk population is often a false positive, and you need to say so clearly.
- Expect to do the arithmetic yourself; the numbers are chosen so the working is short
- Questions often hide the key number in the stem, such as the prevalence in a new population
- Interpretation matters as much as arithmetic: a correct number with the wrong meaning still loses the mark
How the MCCQE frames these questions: three mini cases
Case one: a family physician reads that a new blood marker for colorectal cancer has a sensitivity of 95%. A 52-year-old patient with no symptoms asks whether she should have it instead of her provincial screening test. The exam is not asking you to praise a high sensitivity. It wants you to ask how the marker performed in a screening population, what its specificity is, whether it reduces deaths, and whether it fits the provincial program. A test validated in symptomatic hospital patients will usually perform worse in average-risk people.
Case two: a drug company sales representative leaves a brochure saying a new lipid drug cuts heart attacks by 30%. The baseline risk of the patients in the trial was low. The stem then asks what you should tell a patient. The best answer converts the relative figure into an absolute risk reduction and an NNT over the trial period, and frames it in natural frequencies such as out of 100 people like you. This blends calculation with communication, which is exactly how the blueprint mixes physician activities.
Case three: a public health unit notices more cases of a gastrointestinal illness after a community event. The question may ask for the attack rate, the next step in the outbreak investigation, or the best study design to find the source. Here a retrospective cohort of everyone who attended, comparing food-specific attack rates, is usually the efficient choice because the whole exposed population is known. If attendance cannot be listed, a case-control approach is used instead.
- Expect clinical framing around a statistical core
- The correct answer often adds a practical step: confirm, explain, or refer to the provincial program
- Read the last sentence first so you know which measure to calculate
Study designs: which one answers which question
Match the design to the question. If the outcome is rare or takes decades to appear, a case-control study is efficient: you start with people who have the disease and people who do not, then look back for exposure. Its natural measure is the odds ratio. It is quick and cheap but vulnerable to recall bias and to poor choice of controls.
If the exposure is rare, or you need incidence and the time order of events, use a cohort study. You start with exposed and unexposed people who do not yet have the outcome and follow them forward (prospective) or use records to reconstruct the follow-up (retrospective). Its natural measure is relative risk, and it can study several outcomes from one exposure. Loss to follow-up is its main weakness.
A cross-sectional study measures exposure and outcome at the same moment. It gives prevalence and is useful for planning services, but it cannot show which came first. An ecological study compares groups, not individuals, so a link seen between countries may not hold for people (the ecological fallacy). Case reports and case series raise hypotheses but cannot test them because there is no comparison group.
The randomized controlled trial is the strongest single design for treatment questions because randomization balances both known and unknown confounders. Blinding protects against information bias, and allocation concealment protects against selection at the moment of enrolment. A systematic review with meta-analysis pools several trials and sits at the top of most evidence hierarchies, but only if the included studies are of good quality and reasonably similar.
- Rare disease: case-control, odds ratio
- Rare exposure or need incidence: cohort, relative risk
- Prevalence or burden at one point in time: cross-sectional
- Does a treatment work: randomized controlled trial
- Is it unethical to randomize the exposure (for example smoking): observational design
Diagnostic tests: sensitivity, specificity and predictive values
Draw the 2x2 table every time. Put disease status across the top (present, absent) and the test result down the side (positive, negative). The cells are true positives (a), false positives (b), false negatives (c) and true negatives (d). Sensitivity is a/(a+c): of the people who have the disease, how many test positive. Specificity is d/(b+d): of the people without the disease, how many test negative. These are properties of the test and do not change with prevalence.
Predictive values are what patients care about. Positive predictive value is a/(a+b): if my test is positive, how likely is it that I have the disease. Negative predictive value is d/(c+d). Both depend heavily on prevalence. Take the same test from a specialist clinic to a general population screening program and the PPV falls, often sharply, while the NPV rises. This is the single most tested idea in the subject.
A highly sensitive test is good for ruling out disease when negative (SnNout), which is why it is used first in screening and triage. A highly specific test is good for ruling in disease when positive (SpPin), which is why it is used to confirm. Moving a cut-off to catch more true cases raises sensitivity and lowers specificity. Running two tests in series (both must be positive) increases specificity; running them in parallel (either positive counts) increases sensitivity.
Likelihood ratios combine sensitivity and specificity into one number that does not depend on prevalence. LR+ is sensitivity/(1 - specificity); LR- is (1 - sensitivity)/specificity. An LR+ above about 10 or an LR- below about 0.1 usually changes management. Multiply pretest odds by the LR to get post-test odds, then convert back to probability if the question asks for it.
- Sensitivity = TP / all with disease
- Specificity = TN / all without disease
- PPV = TP / all positive tests; NPV = TN / all negative tests
- Lower prevalence: PPV down, NPV up, sensitivity and specificity unchanged
- Odds = probability / (1 - probability)
Worked walkthrough: from pretest probability to a decision
Suppose a patient's pretest probability of a condition is 20% and a test has a positive likelihood ratio of 6. Convert probability to odds: 0.2 / 0.8 = 0.25. Multiply by the LR: 0.25 x 6 = 1.5. Convert back to probability: 1.5 / (1 + 1.5) = 0.6, so 60%. The positive result has tripled the probability, but it has not confirmed the diagnosis. If the treatment threshold is above 60%, you need a further test.
Now take the same test with an LR- of 0.2 and a negative result. Post-test odds are 0.25 x 0.2 = 0.05, which is a probability of about 5%. Whether that is low enough to stop depends on how dangerous a missed case would be. For a condition such as pulmonary embolism, clinical decision rules and D-dimer are combined so that the post-test probability falls below the level where imaging is needed. The exam uses this logic often: the test is only useful if the result moves the patient across a decision threshold.
A receiver operating characteristic (ROC) curve plots sensitivity against 1 - specificity for every possible cut-off. A test whose curve sits close to the top left corner discriminates well, and the area under the curve summarizes this; 0.5 means no better than chance. Moving the cut-off slides you along the curve: you gain sensitivity and lose specificity, or the reverse. The choice depends on whether missing a case or causing a false alarm is more harmful.
- Odds = p / (1 - p); probability = odds / (1 + odds)
- Post-test odds = pretest odds x LR
- A test is worth ordering only if a result could change management
- Serious missed diagnosis: favour high sensitivity and a low LR-
Solving any diagnostic test question
- 1Name the measure askedIs it a test property (sensitivity, specificity, LR) or a result in this population (PPV, NPV)?
- 2Pick a round populationUse 1,000 or 10,000 people and apply the stated prevalence to split diseased and non-diseased.
- 3Fill the 2x2 tableSensitivity gives true positives; specificity gives true negatives. Subtract to get the false cells.
- 4Divide down the right row or columnPPV and NPV run across rows (test result); sensitivity and specificity run down columns (disease).
- 5Sanity-check and interpretLow prevalence means low PPV. Ask whether the result crosses a treatment or testing threshold.
Measures of frequency, risk and effect
Incidence counts new cases in a population at risk over a period of time. Prevalence counts all existing cases at one point (point prevalence) or across a period (period prevalence). In a stable situation, prevalence is roughly incidence multiplied by average disease duration. That is why a new treatment that keeps people alive longer without curing them raises prevalence, and why a disease that is quickly fatal or quickly cured has a low prevalence even with high incidence.
Relative risk is the incidence in the exposed divided by the incidence in the unexposed. A relative risk of 1 means no association. The odds ratio approximates the relative risk when the outcome is rare, which is why case-control studies can still answer risk questions. Hazard ratios come from survival analysis and are read the same way: below 1 favours the treatment when the outcome is bad.
For treatment decisions, absolute numbers matter most. Absolute risk reduction (ARR) is the control event rate minus the treatment event rate. Relative risk reduction (RRR) is ARR divided by the control event rate. Number needed to treat (NNT) is 1/ARR, rounded up, and it always carries a time frame and an outcome. Number needed to harm works the same way with the absolute risk increase of an adverse effect. A drug with a 50% RRR can have a large or tiny NNT depending on the baseline risk, which is why the MCCQE likes to ask you to explain benefit in absolute terms.
Attributable risk is the risk difference between exposed and unexposed and tells you how much disease the exposure causes among the exposed. Population attributable risk adds how common the exposure is, so it tells a public health planner how much disease would disappear if the exposure were removed from the whole community. A weak risk factor that is very common can matter more to the population than a strong but rare one.
- ARR = control event rate - treatment event rate
- RRR = ARR / control event rate
- NNT = 1 / ARR (round up, state time frame)
- Prevalence is about incidence times duration
Bias, confounding and chance
Bias is a systematic error that pushes the result in one direction. Selection bias happens when the people studied differ from the target population or when groups are chosen differently, for example comparing hospital cases with community controls. The healthy worker effect is a selection bias: employed people are healthier than the general population, so occupational cohorts can hide harm. Loss to follow-up becomes bias when the people who drop out differ from those who stay.
Information bias comes from measuring exposure or outcome wrongly. Recall bias is typical in case-control studies: parents of a child with a birth defect search their memory harder than parents of a healthy child. Observer bias occurs when an assessor who knows the group assignment interprets findings differently. Blinding is the fix for both observer and participant expectations.
Screening studies have their own biases. Lead-time bias makes survival look longer only because diagnosis came earlier, even if death occurs at the same time. Length-time bias happens because screening picks up slow-growing disease more often than aggressive disease, which makes screen-detected cases look better. Overdiagnosis is the detection of disease that would never have caused symptoms. The correct outcome to judge a screening program is disease-specific mortality (and ideally all-cause mortality) in a randomized comparison, not survival from diagnosis.
A confounder is linked to both the exposure and the outcome and is not on the causal pathway. Coffee drinkers seem to get more lung cancer because coffee drinkers smoke more. Control it at the design stage with randomization, restriction or matching, or at the analysis stage with stratification or multivariable adjustment. Effect modification is different: the effect truly differs between subgroups, and you should report it, not remove it.
Chance is handled with statistics. A p-value is the probability of seeing a result this extreme if there were truly no effect. A 95% confidence interval that crosses 1 for a ratio, or 0 for a difference, means the result is not statistically significant. A type I error (alpha) is a false positive finding; a type II error (beta) is missing a real effect, usually because the sample was too small. Power is 1 - beta. Statistical significance is not the same as clinical importance.
- Recall bias: case-control studies; fix with objective records
- Lead-time and length-time bias: screening studies; judge by mortality in an RCT
- Confounding: fix with randomization, restriction, matching, stratification or adjustment
- Intention-to-treat analysis keeps the benefit of randomization; per-protocol analysis can reintroduce bias
Bias red flags in exam stems
Describing data and choosing a statistical test
In a normal distribution the mean, median and mode are equal, and about 68% of values fall within 1 standard deviation of the mean and about 95% within 2. A positively skewed distribution has a long tail to the right, so the mean is pulled above the median; length of hospital stay and income are typical examples. For skewed data the median and interquartile range describe the centre and spread better than the mean and standard deviation.
Standard deviation describes how spread out individual values are. Standard error describes how precisely the sample mean estimates the true mean, and it shrinks as the sample grows. Confidence intervals are built from the standard error, which is why bigger trials give narrower intervals.
The MCCQE does not ask you to run tests, but it may ask which one fits. Compare two means with a t-test (paired if the same people are measured twice). Compare three or more means with analysis of variance. Compare proportions in categories with a chi-square test. Correlation describes the strength of a linear relationship between two continuous variables and does not prove cause. Regression predicts one variable from others, and multivariable regression is one way to adjust for confounders.
Survival analysis handles time-to-event data in which some participants are still event-free at the end of follow-up (censored). A Kaplan-Meier curve shows the share still event-free over time, and the hazard ratio compares the rate of events between groups. Median survival is the time at which half the group has had the event.
- Skewed data: report median and interquartile range
- Two means: t-test; three or more: ANOVA; proportions: chi-square
- Correlation is not causation
- Larger sample: smaller standard error, narrower confidence interval
Critical appraisal and applying evidence to a patient
Appraise any study with three questions. Are the results valid (right design, randomization, concealment, blinding, complete follow-up, intention-to-treat)? What are the results (size of effect, absolute numbers, precision of the confidence interval)? Do they apply to my patient (similar age, sex, risk, setting, and outcomes that matter to the patient)? Exam questions often test the last step: a trial done in high-risk men may not justify treating a low-risk woman.
Watch for surrogate outcomes. A drug that lowers a lab value is not proven to reduce heart attacks, fractures or death until a trial shows it. Composite outcomes can hide that the benefit came from the least important component. Subgroup findings that were not planned in advance are hypothesis-generating only.
The Canadian Task Force on Preventive Health Care uses the GRADE approach and labels recommendations as strong or conditional (previously weak). A conditional recommendation means the balance of benefit and harm depends on patient values, so the right answer on the exam is usually shared decision-making rather than a blanket order. Non-inferiority trials ask whether a new option is not unacceptably worse than the standard, often because it is cheaper, safer or easier; they do not prove superiority.
- Validity, results, applicability
- Prefer patient-important outcomes over surrogate markers
- Read absolute numbers before relative numbers
- Conditional recommendation: discuss values and preferences
Screening principles and Canadian practice
Screening means testing people without symptoms. It is only worthwhile when the condition is an important health problem, has a detectable early phase, has an acceptable test and an effective treatment that works better when started early, and when the program can follow up abnormal results. These ideas come from the classic Wilson and Jungner principles, and the MCCQE expects you to apply them rather than recite them.
Every screening program has harms: false positives, anxiety, invasive follow-up tests, overdiagnosis and overtreatment. The benefit is a reduction in deaths or serious illness. The CTFPHC weighs both and publishes recommendations by age and risk group. Provinces and territories run organized programs, especially for cancer screening, and details such as recall intervals and test type can differ between provinces. On the exam, use the Canadian recommendation for an average-risk patient unless the stem gives a reason for higher risk.
Canadian and US recommendations differ in places, and this is a common source of wrong answers for candidates who studied from US material. The CTFPHC recommends against routine PSA screening for prostate cancer, while US guidance offers individual decision-making in some age groups. The CTFPHC also tends to recommend against screening where evidence of benefit is weak, for example routine screening for some conditions in low-risk adults. When the two countries disagree, answer with the Canadian position and be ready to support shared decision-making where the recommendation is conditional.
- Judge a screening program by mortality in randomized trials, not by survival time after diagnosis
- A positive screening test is not a diagnosis: arrange the confirmatory test
- Higher-risk patients (family history, genetic syndromes) leave the average-risk pathway
- Explain false positives before the test, not only after
Screening: Canadian answers versus US habits
Canadian practice (CTFPHC, provincial programs)
- Recommends against routine PSA screening for prostate cancer
- Lung cancer: low-dose CT for high-risk smokers aged 55 to 74, annually for up to 3 consecutive years
- Organized provincial programs send invitations and recalls
- Strong or conditional recommendations using GRADE; conditional means shared decision
Common US habit (USPSTF and others)
- PSA is an individual decision for men aged 55 to 69
- Lung cancer: annual low-dose CT from age 50 with a lower pack-year threshold
- Screening is often opportunistic and driven by insurers and individual clinicians
- Letter grades A to D and I, with different wording for the same evidence
Levels of prevention, causation and outbreak steps
Primary prevention stops disease before it starts, such as immunization or smoking cessation. Secondary prevention finds disease early in people without symptoms, which is screening. Tertiary prevention limits complications in established disease, such as cardiac rehabilitation. Quaternary prevention protects patients from unnecessary or harmful medical activity, a theme that fits Choosing Wisely Canada. Some frameworks add primordial prevention, which targets the social and environmental conditions that create risk factors in the first place.
To judge whether an association is likely causal, use the Bradford Hill considerations: strength, consistency, specificity, temporality, biological gradient (dose response), plausibility, coherence, experiment and analogy. Only temporality is essential: the cause must come before the effect. Exam stems often test this with a cross-sectional study that cannot show which came first.
An outbreak investigation follows a predictable order: confirm the diagnosis and that an outbreak exists, write a case definition, find and count cases, describe them by time, place and person (an epidemic curve helps), form a hypothesis, test it with an analytic study, put control measures in place as early as the evidence allows, and communicate findings. In Canada, physicians report notifiable diseases to the local medical officer of health under provincial public health law. The attack rate is the number of new cases divided by the population at risk during the outbreak.
- Screening is secondary prevention
- Temporality is the one essential causal criterion
- Control measures do not wait for the full analysis if the source is clear
- Notifiable diseases are reported to local public health, not only to the hospital
Explaining numbers to patients: the communication angle
Many epidemiology items end with a patient question: should I take this pill, should I have this test, what does my result mean. The best answer uses absolute numbers, a stated time frame and natural frequencies. Saying out of 100 people like you who take this for 5 years, about 4 avoid a stroke and 96 get no benefit from it on this outcome is clearer and more honest than saying it cuts strokes by 40%.
Present benefits and harms in the same format so the patient can compare them. Mention false positives and what follow-up a positive screen would trigger before the test is done, so the patient can make an informed choice. When the CTFPHC recommendation is conditional, the expected answer is a shared decision that reflects the patient's values. Decision aids are a reasonable choice in these stems. Avoid percentages alone with patients who have limited numeracy or English, and check understanding with teach-back.
- Same denominator, same time frame for benefit and harm
- Natural frequencies over percentages
- Discuss false positives before screening
- Conditional recommendation: shared decision-making
How to read epidemiology stems quickly and safely
Start by finding what the question actually asks: a property of the test (sensitivity, specificity), a property of the result in this population (PPV, NPV), a risk measure (RR, OR), or a treatment measure (ARR, NNT). Then build the table. If the stem gives percentages instead of counts, invent a convenient population such as 1,000 or 10,000 people and fill the cells. This turns every predictive value question into simple division.
Look for the hidden prevalence. A stem may give sensitivity and specificity from a hospital study and then describe a screening clinic where only 1 in 100 people has the disease. The answer changes completely. Look for the time frame in trial questions; an NNT without a time frame is incomplete. Check whether the question wants a percentage, a ratio or a whole number, and round NNT up.
For design and bias questions, underline the direction of the study. Starting from disease and looking back is case-control. Starting from exposure and following forward is cohort. Assigning the exposure is a trial. Then ask what could make the result wrong: how subjects were chosen, how data were collected, what else differs between groups, and whether the outcome measure was fair.
- Step 1: name the measure being asked for
- Step 2: build the 2x2 table with round numbers
- Step 3: check prevalence, time frame and units
- Step 4: sanity-check the answer (a PPV cannot exceed 100%, an NNT is a whole number)
A one-week revision plan for this subject
Day 1: draw the 2x2 table from memory and work ten sensitivity, specificity and predictive value items. Day 2: change prevalence across five versions of the same question and convert pretest to post-test probability with likelihood ratios. Day 3: risk measures, ARR, RRR, NNT and NNH from both percentages and raw counts. Day 4: study designs and their natural measures, then bias names with one example each. Day 5: screening principles, lead-time and length-time bias, and the main CTFPHC positions compared with US guidance. Day 6: data description, test selection, survival curves and outbreak steps. Day 7: a timed mixed block from this question bank, then rewrite every wrong answer as a one-line rule.
- Keep each day to about 60-90 minutes
- Do 10 minutes of mental arithmetic drills daily
- Redo missed questions two days later
| Measure | Formula | What it tells you |
|---|---|---|
| Sensitivity | TP / (TP + FN) | Share of people with the disease who test positive; high value rules out when negative |
| Specificity | TN / (TN + FP) | Share of people without the disease who test negative; high value rules in when positive |
| Positive predictive value | TP / (TP + FP) | Chance a positive result is a true case; falls as prevalence falls |
| Negative predictive value | TN / (TN + FN) | Chance a negative result is truly negative; rises as prevalence falls |
| Positive likelihood ratio | Sensitivity / (1 - specificity) | How much a positive result raises the odds of disease |
| Relative risk | Incidence exposed / incidence unexposed | Strength of association in cohort studies and trials |
| Odds ratio | (a x d) / (b x c) | Association in case-control studies; close to RR when the outcome is rare |
| Absolute risk reduction | Control event rate - treatment event rate | Real size of benefit for the patient |
| Number needed to treat | 1 / ARR, rounded up | Patients treated for a stated time to prevent one event |
| Prevalence (stable state) | Incidence x average duration | Why longer survival without cure raises prevalence |
Worked examples in MCCQE style
Pick an answer to see why each option is right or wrong. These examples are written for this page; the bank holds the full set for epidemiology & biostatistics.
Example 1
A new rapid blood test for a bacterial infection was validated in an emergency department. It has a sensitivity of 90% and a specificity of 80%. A community health centre plans to use it on 1,000 adults who have mild symptoms, in a group where the expected prevalence of the infection is 10%.
What is the positive predictive value of the test in this community population?
-
This is the prevalence, which is the pretest probability, not the probability of disease after a positive result.
-
Of 100 people with infection, 90 test positive. Of 900 without infection, 20% (180) test positive. PPV = 90 / (90 + 180) = 33%.
-
This is the specificity, a property of the test that does not answer the patient's question about a positive result.
-
This is the sensitivity. It tells you how many infected people are detected, not how many positives are truly infected.
-
This is close to the negative predictive value: 720 true negatives out of 730 negative results, about 98.6%.
Show answer and reasoning
Answer: B. About 33%
- A. This is the prevalence, which is the pretest probability, not the probability of disease after a positive result.
- B. Of 100 people with infection, 90 test positive. Of 900 without infection, 20% (180) test positive. PPV = 90 / (90 + 180) = 33%.
- C. This is the specificity, a property of the test that does not answer the patient's question about a positive result.
- D. This is the sensitivity. It tells you how many infected people are detected, not how many positives are truly infected.
- E. This is close to the negative predictive value: 720 true negatives out of 730 negative results, about 98.6%.
Build the 2x2 table with real counts. In a low-prevalence population even a good test gives many false positives, so PPV drops.
Based on: Standard diagnostic test calculation; MCC Examination Objectives (population health)
Example 2
A randomized controlled trial enrols 4,000 adults with atherosclerotic disease. Over 5 years, stroke occurs in 10% of patients in the placebo group and in 6% of patients in the treatment group. A patient in your family practice asks how much the medication would actually help someone like her.
How many patients need to be treated for 5 years to prevent one stroke?
-
This confuses the absolute risk reduction (4 percentage points) with the number needed to treat.
-
This uses the control event rate of 10% (1/0.10) instead of the difference between groups.
-
This is 1/0.06, the inverse of the treatment event rate, which has no meaning as an NNT.
-
ARR = 10% - 6% = 4% = 0.04. NNT = 1 / 0.04 = 25 patients treated for 5 years to prevent one stroke.
-
This takes the relative risk reduction of 40% and treats it as the NNT, mixing relative and absolute measures.
Show answer and reasoning
Answer: D. 25 patients
- A. This confuses the absolute risk reduction (4 percentage points) with the number needed to treat.
- B. This uses the control event rate of 10% (1/0.10) instead of the difference between groups.
- C. This is 1/0.06, the inverse of the treatment event rate, which has no meaning as an NNT.
- D. ARR = 10% - 6% = 4% = 0.04. NNT = 1 / 0.04 = 25 patients treated for 5 years to prevent one stroke.
- E. This takes the relative risk reduction of 40% and treats it as the NNT, mixing relative and absolute measures.
NNT = 1 / ARR, always with the time frame. Explain benefit to patients in absolute terms, not as a relative reduction.
Based on: Evidence-based medicine principles for communicating treatment benefit
Example 3
A province introduces screening for a cancer using an imaging test. Five years later, a report shows that average survival from diagnosis is 3 years longer in screen-detected patients than in patients diagnosed after symptoms. However, the death rate from this cancer in the province has not changed.
Which explanation best accounts for these findings?
-
Screening moves the date of diagnosis earlier, so survival from diagnosis looks longer even if death comes at the same time. Unchanged mortality is the giveaway.
-
Recall bias affects self-reported exposures in case-control studies. Survival dates come from records, not memory.
-
This is a selection bias in occupational studies where employed people are healthier than the general population.
-
Age could confound a comparison, but it would not explain longer survival with an unchanged population death rate.
-
Losing patients from follow-up could distort survival, but nothing in the stem suggests missing data.
Show answer and reasoning
Answer: A. Lead-time bias
- A. Screening moves the date of diagnosis earlier, so survival from diagnosis looks longer even if death comes at the same time. Unchanged mortality is the giveaway.
- B. Recall bias affects self-reported exposures in case-control studies. Survival dates come from records, not memory.
- C. This is a selection bias in occupational studies where employed people are healthier than the general population.
- D. Age could confound a comparison, but it would not explain longer survival with an unchanged population death rate.
- E. Losing patients from follow-up could distort survival, but nothing in the stem suggests missing data.
Longer survival after screening means little on its own. Judge screening by disease-specific mortality in a randomized comparison.
Based on: Screening principles used by the Canadian Task Force on Preventive Health Care
Example 4
Investigators study a rare congenital heart defect. They identify 100 infants with the defect and 100 infants without it from the same hospitals. Mothers' prescription records show that 40 of the case mothers and 20 of the control mothers filled a prescription for a particular antiepileptic drug in the first trimester.
What is the odds ratio for the association between the drug and the defect?
-
This does not come from any correct arrangement of the table and underestimates the association.
-
This is the ratio of exposed cases to exposed controls (40/20), which ignores the unexposed cells.
-
Odds of exposure in cases = 40/60; in controls = 20/80. OR = (40 x 80) / (60 x 20) = 3,200 / 1,200, about 2.7.
-
This would follow from swapping cells in the cross-product, a common slip when the table is not drawn.
-
This doubles the ratio of exposure proportions without using the odds formula.
Show answer and reasoning
Answer: C. About 2.7
- A. This does not come from any correct arrangement of the table and underestimates the association.
- B. This is the ratio of exposed cases to exposed controls (40/20), which ignores the unexposed cells.
- C. Odds of exposure in cases = 40/60; in controls = 20/80. OR = (40 x 80) / (60 x 20) = 3,200 / 1,200, about 2.7.
- D. This would follow from swapping cells in the cross-product, a common slip when the table is not drawn.
- E. This doubles the ratio of exposure proportions without using the odds formula.
In a case-control study, compute the odds ratio as ad/bc. Records rather than recall reduce information bias.
Based on: Standard case-control analysis; MCC Examination Objectives (population health)
Example 5
A multicentre randomized trial compares a new antiplatelet drug with standard therapy in 6,000 patients after a minor stroke. The primary outcome is recurrent stroke at 1 year. The reported hazard ratio is 0.85 with a 95% confidence interval of 0.70 to 1.03.
Which interpretation of this result is most accurate?
-
The point estimate suggests a 15% relative reduction, but the interval is wide and includes no effect, so there is no certainty.
-
The 95% interval crosses 1, so the result is not significant at the 5% level. The data fit modest benefit, no effect or slight harm.
-
The upper limit of 1.03 is a boundary of uncertainty, not an estimate of harm in most patients.
-
Absence of significance is not proof of no effect. The trial may have been underpowered for a modest benefit.
-
A point estimate below 1 is not enough. Significance depends on whether the interval excludes 1.
Show answer and reasoning
Answer: B. The result is not statistically significant at 5%
- A. The point estimate suggests a 15% relative reduction, but the interval is wide and includes no effect, so there is no certainty.
- B. The 95% interval crosses 1, so the result is not significant at the 5% level. The data fit modest benefit, no effect or slight harm.
- C. The upper limit of 1.03 is a boundary of uncertainty, not an estimate of harm in most patients.
- D. Absence of significance is not proof of no effect. The trial may have been underpowered for a modest benefit.
- E. A point estimate below 1 is not enough. Significance depends on whether the interval excludes 1.
For a ratio, check whether the confidence interval includes 1. Non-significant does not mean proven equivalent.
Based on: Evidence-based medicine principles for interpreting trial results
Quick recall
Tap a card to turn it over. If you can answer all of them without looking, move on to timed questions.
Where candidates lose marks
- Relative instead of absolute
A 40% relative reduction sounds large but may mean an NNT of 100 or more in low-risk patients. Report ARR and NNT when asked about benefit.
- Ignoring the new prevalence
Sensitivity and specificity carry over to a new setting, but PPV must be recalculated with the new prevalence.
- Survival as proof of screening
Longer survival after diagnosis can come from lead-time or length-time bias. Mortality reduction is the real test.
- Mixing up designs
Starting from cases and looking back is case-control even if the data are collected now. Starting from exposure is cohort.
- Significance equals importance
A small p-value can come with a trivial effect in a huge trial. Look at the effect size and the confidence interval.
How to study epidemiology & biostatistics
Learn the 2x2 table layout and write the formulas from memory until you can do it in under a minute.
Do five predictive value questions where you change only the prevalence, and watch PPV and NPV move.
Practise NNT and NNH from trial results with percentages and with raw counts.
Make a one-page sheet matching each study design to its measure, strengths and typical bias.
Read the CTFPHC summaries for the main cancer and chronic disease screening topics and note where they differ from US guidance.
Finish with timed mixed blocks from this question bank and review every calculation you got wrong.
Sources and further reading
Check the current edition of each guideline before your exam. This page is study material, not clinical advice.