Module 2 — Bias, Confounding & Effect Modification

Question 1 of 60

A case-control study evaluates whether prior oral-contraceptive use is associated with venous thromboembolism. Cases are recruited from all hospitals in a city, but controls are recruited only from a thrombosis specialty clinic where many patients have strong family histories of clotting disorders. Which problem is most likely?

A. Selection bias B. Recall bias C. Confounding D. Lead-time bias E. Effect modification

Correct answer: A — Selection bias

A–E option analysis

A.Correct. The control group has an abnormal baseline risk profile and does not represent the source population from which the cases arose. This can distort the exposure distribution among controls and bias the odds ratio.
B.Incorrect. Recall bias concerns differential memory of past exposure; the main problem here is how controls were selected. The control group has an abnormal baseline risk profile and does not represent the source population from which the cases arose. This can distort the exposure distribution among controls and bias the odds ratio.
C.Incorrect. Confounding may also occur, but the defining defect is that controls do not represent the population that produced the cases. The control group has an abnormal baseline risk profile and does not represent the source population from which the cases arose. This can distort the exposure distribution among controls and bias the odds ratio.
D.Incorrect. Lead-time bias is a screening-related distortion caused by earlier diagnosis. The control group has an abnormal baseline risk profile and does not represent the source population from which the cases arose. This can distort the exposure distribution among controls and bias the odds ratio.
E.Incorrect. Effect modification is a real difference in effect across subgroups, not a flaw in participant selection. The control group has an abnormal baseline risk profile and does not represent the source population from which the cases arose. This can distort the exposure distribution among controls and bias the odds ratio.

Disease / clinical review: Venous thromboembolism risk is influenced by estrogen exposure, inherited thrombophilia, surgery, immobility, cancer, pregnancy, and other factors. Case-control validity depends heavily on appropriate control selection.

Clinical safety note: Do not change clinical decisions from a biased observational association without checking whether cases and controls came from comparable source populations.

MCCQE exam trap: Controls should represent the exposure distribution in the source population that produced the cases.

Authoritative references:

Question 2 of 60

Researchers estimate smoking prevalence among Canadian adults by surveying volunteers who click a social-media advertisement titled “Tell us about your smoking habits.” What is the greatest threat to validity?

A. Recall bias B. Interviewer bias C. Volunteer selection bias D. Non-differential misclassification E. Length-time bias

Correct answer: C — Volunteer selection bias

A–E option analysis

A.Incorrect. The question concerns who enters the study rather than whether participants remember smoking history inaccurately. People who choose to respond to a smoking survey may differ systematically from nonresponders in smoking status, health interest, age, or other characteristics, producing volunteer selection bias.
B.Incorrect. There is no interviewer-driven differential questioning in the scenario. People who choose to respond to a smoking survey may differ systematically from nonresponders in smoking status, health interest, age, or other characteristics, producing volunteer selection bias.
C.Correct. People who choose to respond to a smoking survey may differ systematically from nonresponders in smoking status, health interest, age, or other characteristics, producing volunteer selection bias.
D.Incorrect. Misclassification concerns measurement error after participants are sampled; the central problem is self-selection into the survey. People who choose to respond to a smoking survey may differ systematically from nonresponders in smoking status, health interest, age, or other characteristics, producing volunteer selection bias.
E.Incorrect. Length-time bias occurs in screening when slower-progressing disease is preferentially detected. People who choose to respond to a smoking survey may differ systematically from nonresponders in smoking status, health interest, age, or other characteristics, producing volunteer selection bias.

Disease / clinical review: Smoking prevalence is a population estimate. Representative probability-based sampling is far more important than simply obtaining a large number of responses.

Clinical safety note: Population-health decisions should not rely on an unrepresentative convenience sample when prevalence estimates could affect policy or resource allocation.

MCCQE exam trap: A very large convenience sample can still be badly biased.

Authoritative references:

Question 3 of 60

A cohort study enrolls only workers who have remained employed at a chemical plant for at least 15 years and compares their mortality with that of the general population. Mortality appears lower among the workers. What best explains this finding?

A. Healthy-worker effect B. Recall bias C. Berkson bias D. Lead-time bias E. Regression to the mean

Correct answer: A — Healthy-worker effect

A–E option analysis

A.Correct. Employed populations are often healthier than the general population because people who are severely ill or disabled are less likely to enter or remain employed. This selection phenomenon can make occupational cohorts appear healthier.
B.Incorrect. The exposure and outcome are not being distorted primarily by memory. Employed populations are often healthier than the general population because people who are severely ill or disabled are less likely to enter or remain employed. This selection phenomenon can make occupational cohorts appear healthier.
C.Incorrect. Berkson bias typically arises from hospital-based selection in case-control studies. Employed populations are often healthier than the general population because people who are severely ill or disabled are less likely to enter or remain employed. This selection phenomenon can make occupational cohorts appear healthier.
D.Incorrect. Lead-time bias concerns apparent longer survival due to earlier diagnosis. Employed populations are often healthier than the general population because people who are severely ill or disabled are less likely to enter or remain employed. This selection phenomenon can make occupational cohorts appear healthier.
E.Incorrect. Regression to the mean concerns repeated measurements after selecting extreme values, not employment-based selection. Employed populations are often healthier than the general population because people who are severely ill or disabled are less likely to enter or remain employed. This selection phenomenon can make occupational cohorts appear healthier.

Disease / clinical review: Occupational-health studies often require an appropriate employed comparison group because employment itself selects for a healthier population.

Clinical safety note: Do not interpret lower crude mortality in workers as proof that a workplace exposure is safe; comparator selection matters.

MCCQE exam trap: Comparing workers directly with the general population can underestimate occupational harm.

Authoritative references:

Question 4 of 60

A study of chronic kidney disease recruits participants only from a tertiary nephrology clinic. Investigators then claim that 70% of all people with chronic kidney disease have severe symptoms. What is the main problem?

A. Detection bias B. Recall bias C. Confounding by age D. Ecological fallacy E. Referral selection bias

Correct answer: E — Referral selection bias

A–E option analysis

A.Incorrect. Detection bias concerns differential outcome ascertainment rather than referral into a specialty clinic. Tertiary centers disproportionately receive complicated or severe cases. A clinic-based sample therefore may not represent the full spectrum of chronic kidney disease in the community.
B.Incorrect. The main issue is not memory of prior exposures. Tertiary centers disproportionately receive complicated or severe cases. A clinic-based sample therefore may not represent the full spectrum of chronic kidney disease in the community.
C.Incorrect. Age may confound some associations, but it does not explain why a tertiary clinic overrepresents severe disease. Tertiary centers disproportionately receive complicated or severe cases. A clinic-based sample therefore may not represent the full spectrum of chronic kidney disease in the community.
D.Incorrect. The claim is based on an individual-level clinic sample, not group-level ecological data. Tertiary centers disproportionately receive complicated or severe cases. A clinic-based sample therefore may not represent the full spectrum of chronic kidney disease in the community.
E.Correct. Tertiary centers disproportionately receive complicated or severe cases. A clinic-based sample therefore may not represent the full spectrum of chronic kidney disease in the community.

Disease / clinical review: Chronic kidney disease ranges from asymptomatic early disease to advanced symptomatic disease. Referral patterns strongly influence the severity mix seen in tertiary care.

Clinical safety note: Overestimating severity can distort counselling, resource planning, and treatment expectations for patients with earlier-stage disease.

MCCQE exam trap: Specialty-clinic prevalence is not population prevalence.

Authoritative references:

Question 5 of 60

Investigators compare asthma prevalence among children attending an emergency department with prevalence among all children in the surrounding city. They conclude that asthma affects 35% of local children. Which bias is most important?

A. Recall bias B. Confounding C. Observer bias D. Lead-time bias E. Selection bias from a nonrepresentative clinical sample

Correct answer: E — Selection bias from a nonrepresentative clinical sample

A–E option analysis

A.Incorrect. The bias arises before measurement because emergency-department attendees differ from the general pediatric population. Emergency-department patients are selected because of illness and healthcare-seeking behavior, so they cannot be assumed to represent all children in the community.
B.Incorrect. Confounding is not the main reason the prevalence estimate is invalid. Emergency-department patients are selected because of illness and healthcare-seeking behavior, so they cannot be assumed to represent all children in the community.
C.Incorrect. Observer expectations are not the main issue. Emergency-department patients are selected because of illness and healthcare-seeking behavior, so they cannot be assumed to represent all children in the community.
D.Incorrect. No screening-related survival comparison is involved. Emergency-department patients are selected because of illness and healthcare-seeking behavior, so they cannot be assumed to represent all children in the community.
E.Correct. Emergency-department patients are selected because of illness and healthcare-seeking behavior, so they cannot be assumed to represent all children in the community.

Disease / clinical review: Asthma is common in children, but emergency departments overrepresent patients with acute symptoms and higher healthcare utilization.

Clinical safety note: Using a biased prevalence estimate can misdirect public-health resources and exaggerate disease burden.

MCCQE exam trap: Prevalence estimates require a denominator sampled from the target population.

Authoritative references:

Question 6 of 60

A case-control study of alcohol use and pancreatitis selects controls from a detoxification unit. Which direction of bias is most likely if alcohol use is much more common among controls than in the source population?

A. Bias away from the null B. No bias because controls do not have pancreatitis C. Lead-time bias D. Bias toward the null E. Effect modification

Correct answer: D — Bias toward the null

A–E option analysis

A.Incorrect. Overrepresentation of the exposure among controls usually reduces the exposure contrast between cases and controls. If controls are unusually likely to have the exposure, the odds of exposure among controls rises and the observed odds ratio is typically pulled closer to 1.
B.Incorrect. Control validity depends on representing the source population, not simply being disease-free. If controls are unusually likely to have the exposure, the odds of exposure among controls rises and the observed odds ratio is typically pulled closer to 1.
C.Incorrect. This is unrelated to screening. If controls are unusually likely to have the exposure, the odds of exposure among controls rises and the observed odds ratio is typically pulled closer to 1.
D.Correct. If controls are unusually likely to have the exposure, the odds of exposure among controls rises and the observed odds ratio is typically pulled closer to 1.
E.Incorrect. The problem is distorted control exposure prevalence, not a true subgroup interaction. If controls are unusually likely to have the exposure, the odds of exposure among controls rises and the observed odds ratio is typically pulled closer to 1.

Disease / clinical review: Alcohol is a major cause of acute and chronic pancreatitis, but a biased control group can seriously distort the apparent strength of association.

Clinical safety note: Do not use a biased odds ratio to minimize or exaggerate clinically established risk.

MCCQE exam trap: Ask how the control-selection mechanism changes exposure prevalence among controls.

Authoritative references:

Question 7 of 60

A study evaluates a new fall-prevention program in long-term-care homes. Homes volunteering for the program already have unusually strong safety cultures and more staffing than nonparticipating homes. Which bias is most concerning when comparing outcomes?

A. Recall bias B. Self-selection bias C. Observer bias D. Non-differential misclassification E. Lead-time bias

Correct answer: B — Self-selection bias

A–E option analysis

A.Incorrect. The program exposure and falls are not mainly limited by memory. Facilities choosing to participate may systematically differ in staffing, leadership, baseline fall risk, or safety practices from nonparticipating homes, making crude outcome differences difficult to attribute to the program.
B.Correct. Facilities choosing to participate may systematically differ in staffing, leadership, baseline fall risk, or safety practices from nonparticipating homes, making crude outcome differences difficult to attribute to the program.
C.Incorrect. Outcome ascertainment could matter, but voluntary participation is the more fundamental selection problem. Facilities choosing to participate may systematically differ in staffing, leadership, baseline fall risk, or safety practices from nonparticipating homes, making crude outcome differences difficult to attribute to the program.
D.Incorrect. The main issue is not equal measurement error across groups. Facilities choosing to participate may systematically differ in staffing, leadership, baseline fall risk, or safety practices from nonparticipating homes, making crude outcome differences difficult to attribute to the program.
E.Incorrect. This does not involve screening or survival from diagnosis. Facilities choosing to participate may systematically differ in staffing, leadership, baseline fall risk, or safety practices from nonparticipating homes, making crude outcome differences difficult to attribute to the program.

Disease / clinical review: Fall risk in long-term care is affected by mobility, cognition, medications, environmental hazards, staffing, and prevention systems.

Clinical safety note: Interventions that appear effective in self-selected high-performing facilities may not have the same effect in facilities with fewer resources.

MCCQE exam trap: Voluntary participation can create major baseline differences even before the intervention starts.

Authoritative references:

Question 8 of 60

A prospective pregnancy cohort requires participants to attend a university research center every month during pregnancy. Women with unstable housing and transportation barriers are much less likely to enroll. Which concern is greatest?

A. Recall bias B. Length-time bias C. Regression dilution D. Selection bias limiting generalizability and possibly internal validity E. Interviewer bias

Correct answer: D — Selection bias limiting generalizability and possibly internal validity

A–E option analysis

A.Incorrect. The primary issue is differential enrollment, not recollection. The enrollment burden selectively excludes people with unstable housing or transportation barriers. This can reduce external validity and can also bias exposure-outcome estimates if enrollment is related to both exposure and outcome risk.
B.Incorrect. Length-time bias is a screening phenomenon. The enrollment burden selectively excludes people with unstable housing or transportation barriers. This can reduce external validity and can also bias exposure-outcome estimates if enrollment is related to both exposure and outcome risk.
C.Incorrect. Regression dilution is related to measurement error in continuous exposures. The enrollment burden selectively excludes people with unstable housing or transportation barriers. This can reduce external validity and can also bias exposure-outcome estimates if enrollment is related to both exposure and outcome risk.
D.Correct. The enrollment burden selectively excludes people with unstable housing or transportation barriers. This can reduce external validity and can also bias exposure-outcome estimates if enrollment is related to both exposure and outcome risk.
E.Incorrect. The problem arises before interviews occur. The enrollment burden selectively excludes people with unstable housing or transportation barriers. This can reduce external validity and can also bias exposure-outcome estimates if enrollment is related to both exposure and outcome risk.

Disease / clinical review: Pregnancy outcomes are influenced by medical and social determinants, including housing, income, transportation, nutrition, substance exposure, and access to prenatal care.

Clinical safety note: Research procedures should minimize unnecessary barriers so vulnerable groups are not systematically excluded from evidence used to guide care.

MCCQE exam trap: Barriers to participation can change both who is represented and the observed association.

Authoritative references:

Question 9 of 60

In a vaccine-effectiveness study, vaccinated adults are much more likely than unvaccinated adults to seek testing whenever they develop mild respiratory symptoms. Which problem can arise?

A. Recall bias B. Healthy-worker effect C. Length-time bias D. Selection into outcome ascertainment E. Ecological fallacy

Correct answer: D — Selection into outcome ascertainment

A–E option analysis

A.Incorrect. The key issue is differential testing and entry into the observed outcome set, not memory. Differential healthcare-seeking and testing can determine who has an outcome detected. If testing probability depends on vaccination status and symptoms, observed case rates can be biased.
B.Incorrect. Employment status is not the selection mechanism here. Differential healthcare-seeking and testing can determine who has an outcome detected. If testing probability depends on vaccination status and symptoms, observed case rates can be biased.
C.Incorrect. No screening survival comparison is being made. Differential healthcare-seeking and testing can determine who has an outcome detected. If testing probability depends on vaccination status and symptoms, observed case rates can be biased.
D.Correct. Differential healthcare-seeking and testing can determine who has an outcome detected. If testing probability depends on vaccination status and symptoms, observed case rates can be biased.
E.Incorrect. The data are individual-level. Differential healthcare-seeking and testing can determine who has an outcome detected. If testing probability depends on vaccination status and symptoms, observed case rates can be biased.

Disease / clinical review: Vaccine-effectiveness estimates depend on comparable opportunity for infection detection across vaccination groups.

Clinical safety note: Biased ascertainment can either overestimate or underestimate vaccine benefit and should not be used uncritically for public-health decisions.

MCCQE exam trap: Who gets tested can be as important as how the test performs.

Authoritative references:

Question 10 of 60

A hospital-based case-control study examines obesity and osteoarthritis. Both cases and controls are selected from orthopedic clinics, where obesity is common because it contributes to many musculoskeletal disorders. What selection problem is classically associated with this situation?

A. Recall bias B. Berkson bias C. Healthy-worker effect D. Lead-time bias E. Attrition bias

Correct answer: B — Berkson bias

A–E option analysis

A.Incorrect. This concerns memory of prior exposure, not hospital-based selection. Berkson bias can occur when both exposure and disease affect the probability of hospital admission or clinic attendance, distorting the exposure-disease association in a hospital-based sample.
B.Correct. Berkson bias can occur when both exposure and disease affect the probability of hospital admission or clinic attendance, distorting the exposure-disease association in a hospital-based sample.
C.Incorrect. That phenomenon compares employed populations with the general population. Berkson bias can occur when both exposure and disease affect the probability of hospital admission or clinic attendance, distorting the exposure-disease association in a hospital-based sample.
D.Incorrect. This is a screening-related distortion. Berkson bias can occur when both exposure and disease affect the probability of hospital admission or clinic attendance, distorting the exposure-disease association in a hospital-based sample.
E.Incorrect. Attrition requires differential loss after enrollment. Berkson bias can occur when both exposure and disease affect the probability of hospital admission or clinic attendance, distorting the exposure-disease association in a hospital-based sample.

Disease / clinical review: Osteoarthritis and obesity are both common in orthopedic populations, so clinic-based exposure distributions may differ from those in the community.

Clinical safety note: Do not infer community-level risk from a hospital-selected sample without checking whether the control condition is related to the exposure.

MCCQE exam trap: Hospital controls are convenient but can be dangerous when the exposure affects the reason for hospitalization.

Authoritative references:

Question 11 of 60

Mothers of infants with congenital heart defects are asked to remember all medications taken during the first trimester. Mothers of unaffected infants are asked the same questions, but mothers of affected infants provide substantially more detailed exposure histories. Which bias is most likely?

A. Selection bias B. Observer bias C. Non-differential misclassification D. Recall bias E. Lead-time bias

Correct answer: D — Recall bias

A–E option analysis

A.Incorrect. The defining feature is differential recollection after participants are selected. Cases may search their memory more intensely than controls, leading to differential recall of prior exposures and biased exposure estimates.
B.Incorrect. The difference is in participant memory rather than assessor expectations. Cases may search their memory more intensely than controls, leading to differential recall of prior exposures and biased exposure estimates.
C.Incorrect. The measurement error differs according to outcome status, so it is not non-differential. Cases may search their memory more intensely than controls, leading to differential recall of prior exposures and biased exposure estimates.
D.Correct. Cases may search their memory more intensely than controls, leading to differential recall of prior exposures and biased exposure estimates.
E.Incorrect. No screening or survival comparison is involved. Cases may search their memory more intensely than controls, leading to differential recall of prior exposures and biased exposure estimates.

Disease / clinical review: Congenital anomaly research frequently relies on retrospective pregnancy exposure histories, making standardized records preferable when available.

Clinical safety note: Do not communicate a causal medication risk from a recall-prone association without stronger corroborating evidence.

MCCQE exam trap: Case-control studies are particularly vulnerable when exposure history depends on memory.

Authoritative references:

Question 12 of 60

An interviewer knows which participants have lung cancer and probes cases much more aggressively than controls about prior occupational exposures. What bias is present?

A. Recall bias B. Selection bias C. Interviewer bias D. Confounding E. Attrition bias

Correct answer: C — Interviewer bias

A–E option analysis

A.Incorrect. Recall bias is participant-driven memory difference; here the interviewer changes the intensity of questioning. Knowledge of outcome status changes how exposure information is elicited, creating systematic differential measurement by the interviewer.
B.Incorrect. Participants have already been selected; the distortion occurs during measurement. Knowledge of outcome status changes how exposure information is elicited, creating systematic differential measurement by the interviewer.
C.Correct. Knowledge of outcome status changes how exposure information is elicited, creating systematic differential measurement by the interviewer.
D.Incorrect. A third variable is not the main issue. Knowledge of outcome status changes how exposure information is elicited, creating systematic differential measurement by the interviewer.
E.Incorrect. No differential loss to follow-up is described. Knowledge of outcome status changes how exposure information is elicited, creating systematic differential measurement by the interviewer.

Disease / clinical review: Occupational lung-cancer research depends on accurate exposure histories, including duration and intensity of exposure.

Clinical safety note: Biased questioning can inflate apparent occupational risk and should be avoided through standardized, blinded assessment when possible.

MCCQE exam trap: Blinding interviewers to case status can reduce interviewer bias when feasible.

Authoritative references:

Question 13 of 60

In a cohort study, both exposed and unexposed participants underreport daily sodium intake by about the same amount. What type of measurement error is this?

A. Differential misclassification B. Recall bias C. Non-differential misclassification D. Selection bias E. Effect modification

Correct answer: C — Non-differential misclassification

A–E option analysis

A.Incorrect. Differential misclassification means the error differs between comparison groups. When measurement error is similar across comparison groups, it is classified as non-differential. For a dichotomous exposure this often biases associations toward the null, although exceptions exist.
B.Incorrect. Although memory may contribute, the key classification is that the measurement error is similar across exposure groups. When measurement error is similar across comparison groups, it is classified as non-differential. For a dichotomous exposure this often biases associations toward the null, although exceptions exist.
C.Correct. When measurement error is similar across comparison groups, it is classified as non-differential. For a dichotomous exposure this often biases associations toward the null, although exceptions exist.
D.Incorrect. The problem occurs during measurement, not enrollment. When measurement error is similar across comparison groups, it is classified as non-differential. For a dichotomous exposure this often biases associations toward the null, although exceptions exist.
E.Incorrect. Effect modification is a real difference in effect across strata. When measurement error is similar across comparison groups, it is classified as non-differential. For a dichotomous exposure this often biases associations toward the null, although exceptions exist.

Disease / clinical review: High sodium intake can affect blood pressure and cardiovascular risk. Dietary exposure measurement is difficult and often benefits from repeated or objective measures.

Clinical safety note: Measurement error can hide clinically important exposure-outcome relationships and should be considered before declaring no association.

MCCQE exam trap: Do not assume non-differential error always biases toward the null in every setting.

Authoritative references:

Question 14 of 60

In a diabetes study, patients who develop nephropathy are subsequently questioned in far greater detail about prior NSAID use than patients who remain free of nephropathy. What type of misclassification is most likely?

A. Non-differential misclassification B. Selection bias C. Random error only D. Differential exposure misclassification E. Ecological fallacy

Correct answer: D — Differential exposure misclassification

A–E option analysis

A.Incorrect. The intensity of exposure ascertainment differs by outcome status. Exposure measurement depends on outcome status, so errors in classifying NSAID use can differ between those with and without nephropathy.
B.Incorrect. The participants are already in the cohort; the main distortion occurs during exposure measurement. Exposure measurement depends on outcome status, so errors in classifying NSAID use can differ between those with and without nephropathy.
C.Incorrect. The systematic difference in questioning makes this more than random error. Exposure measurement depends on outcome status, so errors in classifying NSAID use can differ between those with and without nephropathy.
D.Correct. Exposure measurement depends on outcome status, so errors in classifying NSAID use can differ between those with and without nephropathy.
E.Incorrect. The data are individual-level. Exposure measurement depends on outcome status, so errors in classifying NSAID use can differ between those with and without nephropathy.

Disease / clinical review: Diabetic kidney disease risk is influenced by glycemic control, blood pressure, duration of diabetes, medications, and other nephrotoxic exposures.

Clinical safety note: Do not attribute kidney injury to an exposure based on differential post-outcome questioning alone.

MCCQE exam trap: Differential misclassification can bias an association toward or away from the null.

Authoritative references:

Question 15 of 60

A trial compares two wound dressings. Nurses know treatment allocation and are also responsible for deciding whether a wound is 'healed,' a partly subjective outcome. Which bias is most likely?

A. Selection bias B. Recall bias C. Observer ascertainment bias D. Confounding E. Healthy-worker effect

Correct answer: C — Observer ascertainment bias

A–E option analysis

A.Incorrect. Randomized participants are already enrolled; the concern is outcome assessment. Knowledge of treatment assignment can influence subjective judgments of healing, producing observer or ascertainment bias.
B.Incorrect. Patients are not being asked to remember prior exposures. Knowledge of treatment assignment can influence subjective judgments of healing, producing observer or ascertainment bias.
C.Correct. Knowledge of treatment assignment can influence subjective judgments of healing, producing observer or ascertainment bias.
D.Incorrect. Randomization addresses baseline confounding but cannot prevent biased subjective outcome assessment. Knowledge of treatment assignment can influence subjective judgments of healing, producing observer or ascertainment bias.
E.Incorrect. Employment selection is irrelevant. Knowledge of treatment assignment can influence subjective judgments of healing, producing observer or ascertainment bias.

Disease / clinical review: Wound healing can be influenced by perfusion, diabetes, infection, pressure, nutrition, and wound care. Objective or standardized outcome criteria improve reliability.

Clinical safety note: Biased healing assessments can lead clinicians to adopt an ineffective dressing or reject an effective one.

MCCQE exam trap: Blinding the outcome assessor is especially important when outcomes are subjective.

Authoritative references:

Question 16 of 60

Researchers classify hypertension using a single blood-pressure measurement taken immediately after participants climb three flights of stairs. What is the main information problem?

A. Selection bias B. Recall bias C. Confounding D. Effect modification E. Exposure/outcome misclassification from poor measurement

Correct answer: E — Exposure/outcome misclassification from poor measurement

A–E option analysis

A.Incorrect. The sample-selection process is not the main issue. An unstandardized blood-pressure measurement can incorrectly classify hypertension status because recent exertion transiently elevates blood pressure.
B.Incorrect. Blood pressure is being measured rather than recalled. An unstandardized blood-pressure measurement can incorrectly classify hypertension status because recent exertion transiently elevates blood pressure.
C.Incorrect. A third variable is not the defining problem. An unstandardized blood-pressure measurement can incorrectly classify hypertension status because recent exertion transiently elevates blood pressure.
D.Incorrect. The concern is inaccurate classification, not a real difference in effect across subgroups. An unstandardized blood-pressure measurement can incorrectly classify hypertension status because recent exertion transiently elevates blood pressure.
E.Correct. An unstandardized blood-pressure measurement can incorrectly classify hypertension status because recent exertion transiently elevates blood pressure.

Disease / clinical review: Hypertension diagnosis requires standardized measurement and often confirmation across repeated measurements or validated out-of-office readings.

Clinical safety note: Misclassification can expose patients to unnecessary treatment or falsely reassure those with true hypertension.

MCCQE exam trap: Bad measurement can create bias even when the study design itself is appropriate.

Authoritative references:

Question 17 of 60

A study uses an electronic health-record code for 'heart failure' as the outcome, but the code has only moderate sensitivity. The same coding algorithm is applied equally to exposed and unexposed patients. What is the main concern?

A. Differential misclassification B. Selection bias C. Recall bias D. Effect modification E. Non-differential outcome misclassification

Correct answer: E — Non-differential outcome misclassification

A–E option analysis

A.Incorrect. The same imperfect algorithm is applied to both exposure groups. An imperfect outcome definition applied similarly across groups creates non-differential outcome misclassification, which can reduce the ability to detect a true association.
B.Incorrect. The principal defect is outcome measurement, not selection. An imperfect outcome definition applied similarly across groups creates non-differential outcome misclassification, which can reduce the ability to detect a true association.
C.Incorrect. No participant memory is involved. An imperfect outcome definition applied similarly across groups creates non-differential outcome misclassification, which can reduce the ability to detect a true association.
D.Incorrect. This is not a subgroup difference in causal effect. An imperfect outcome definition applied similarly across groups creates non-differential outcome misclassification, which can reduce the ability to detect a true association.
E.Correct. An imperfect outcome definition applied similarly across groups creates non-differential outcome misclassification, which can reduce the ability to detect a true association.

Disease / clinical review: Heart failure diagnosis often requires symptoms, signs, imaging, biomarkers, and clinical judgment. Coding-only definitions may miss or misclassify cases.

Clinical safety note: Misclassified outcomes can distort estimates of treatment effectiveness or disease risk.

MCCQE exam trap: Administrative codes are not automatically equivalent to clinically validated outcomes.

Authoritative references:

Question 18 of 60

Participants in a weight-loss study know they are being observed and temporarily improve their diet and exercise because of the study visits. What phenomenon is this?

A. Recall bias B. Selection bias C. Hawthorne effect D. Regression to the mean E. Lead-time bias

Correct answer: C — Hawthorne effect

A–E option analysis

A.Incorrect. The behavior changes because participants know they are observed, not because they remember past exposures differently. The Hawthorne effect refers to changes in behavior that occur because participants know they are being observed or studied.
B.Incorrect. The effect occurs after enrollment. The Hawthorne effect refers to changes in behavior that occur because participants know they are being observed or studied.
C.Correct. The Hawthorne effect refers to changes in behavior that occur because participants know they are being observed or studied.
D.Incorrect. That concerns repeated measurement of extreme values. The Hawthorne effect refers to changes in behavior that occur because participants know they are being observed or studied.
E.Incorrect. No screening-related survival distortion is involved. The Hawthorne effect refers to changes in behavior that occur because participants know they are being observed or studied.

Disease / clinical review: Weight-management outcomes are sensitive to adherence, self-monitoring, social support, and study contact.

Clinical safety note: Short-term improvement caused by observation alone can make an intervention appear more effective than it will be in routine practice.

MCCQE exam trap: Observed behavior during a study may not reflect usual behavior outside the study.

Authoritative references:

Question 19 of 60

A researcher measuring handgrip strength knows which participants received a rehabilitation intervention and unconsciously encourages those participants more strongly during testing. What bias is most likely?

A. Selection bias B. Observer-expectancy bias C. Recall bias D. Confounding E. Attrition bias

Correct answer: B — Observer-expectancy bias

A–E option analysis

A.Incorrect. The problem arises during measurement, not recruitment. Knowledge of group assignment changes the examiner's behavior and can systematically influence measured performance.
B.Correct. Knowledge of group assignment changes the examiner's behavior and can systematically influence measured performance.
C.Incorrect. Participants are not recalling historical exposure. Knowledge of group assignment changes the examiner's behavior and can systematically influence measured performance.
D.Incorrect. The differential encouragement is a measurement process, not a third-variable distortion. Knowledge of group assignment changes the examiner's behavior and can systematically influence measured performance.
E.Incorrect. No differential dropout is described. Knowledge of group assignment changes the examiner's behavior and can systematically influence measured performance.

Disease / clinical review: Rehabilitation outcomes such as strength and function require standardized testing procedures and ideally blinded assessors.

Clinical safety note: Biased functional assessment can misrepresent rehabilitation benefit and affect treatment decisions.

MCCQE exam trap: Objective measurement is not immune to bias if the assessor changes testing conditions.

Authoritative references:

Question 20 of 60

In a smoking study, participants with chronic cough are more likely to deny smoking because they fear criticism, whereas asymptomatic participants report smoking more honestly. What type of bias is most likely?

A. Social-desirability information bias B. Selection bias C. Lead-time bias D. Healthy-worker effect E. Effect modification

Correct answer: A — Social-desirability information bias

A–E option analysis

A.Correct. Participants may systematically underreport stigmatized behaviors to present themselves more favorably, creating information bias. Here the misreporting may also differ by symptom status.
B.Incorrect. The problem is inaccurate reporting after participants are enrolled. Participants may systematically underreport stigmatized behaviors to present themselves more favorably, creating information bias. Here the misreporting may also differ by symptom status.
C.Incorrect. No screening-related earlier diagnosis is involved. Participants may systematically underreport stigmatized behaviors to present themselves more favorably, creating information bias. Here the misreporting may also differ by symptom status.
D.Incorrect. Employment status is irrelevant. Participants may systematically underreport stigmatized behaviors to present themselves more favorably, creating information bias. Here the misreporting may also differ by symptom status.
E.Incorrect. The true biological effect is not being shown to differ by subgroup. Participants may systematically underreport stigmatized behaviors to present themselves more favorably, creating information bias. Here the misreporting may also differ by symptom status.

Disease / clinical review: Smoking is strongly related to chronic cough and respiratory disease, so underreporting can weaken or distort observed associations.

Clinical safety note: Do not falsely reassure a patient or population based on exposure data known to be vulnerable to social-desirability bias.

MCCQE exam trap: Self-report of stigmatized exposures may require confidential or objective validation methods.

Authoritative references:

Question 21 of 60

A 3-year randomized diabetes trial loses 4% of participants in the control group and 28% in the intervention group because of gastrointestinal adverse effects. Investigators analyze only those who complete the study. What bias is most concerning?

A. Attrition bias B. Recall bias C. Selection bias at enrollment D. Lead-time bias E. Ecological fallacy

Correct answer: A — Attrition bias

A–E option analysis

A.Correct. Differential loss to follow-up, especially when related to treatment adverse effects and possibly outcomes, can destroy comparability between groups and bias the treatment estimate.
B.Incorrect. The concern is differential loss after enrollment, not memory. Differential loss to follow-up, especially when related to treatment adverse effects and possibly outcomes, can destroy comparability between groups and bias the treatment estimate.
C.Incorrect. The major imbalance develops during follow-up rather than at baseline recruitment. Differential loss to follow-up, especially when related to treatment adverse effects and possibly outcomes, can destroy comparability between groups and bias the treatment estimate.
D.Incorrect. No screening survival issue is involved. Differential loss to follow-up, especially when related to treatment adverse effects and possibly outcomes, can destroy comparability between groups and bias the treatment estimate.
E.Incorrect. The data are individual-level. Differential loss to follow-up, especially when related to treatment adverse effects and possibly outcomes, can destroy comparability between groups and bias the treatment estimate.

Disease / clinical review: Diabetes medications can cause adverse effects that influence adherence. Both efficacy and tolerability affect real-world usefulness.

Clinical safety note: Do not interpret a completer-only analysis as reliable when dropout is substantial and treatment-related.

MCCQE exam trap: High and unequal dropout is a major red flag even in a randomized trial.

Authoritative references:

Question 22 of 60

A cohort study of homelessness and mortality loses many participants who move frequently, while stably housed participants are almost completely retained. Why is this problematic?

A. Recall bias B. Loss-to-follow-up bias C. Observer bias D. Non-differential misclassification E. Length-time bias

Correct answer: B — Loss-to-follow-up bias

A–E option analysis

A.Incorrect. The problem is differential retention, not memory. If loss to follow-up is related to exposure and mortality risk, the observed association can be substantially biased.
B.Correct. If loss to follow-up is related to exposure and mortality risk, the observed association can be substantially biased.
C.Incorrect. Outcome assessment style is not the primary issue. If loss to follow-up is related to exposure and mortality risk, the observed association can be substantially biased.
D.Incorrect. The main problem is missing outcome data among one exposure group. If loss to follow-up is related to exposure and mortality risk, the observed association can be substantially biased.
E.Incorrect. No screening program is involved. If loss to follow-up is related to exposure and mortality risk, the observed association can be substantially biased.

Disease / clinical review: Housing instability is itself associated with healthcare access, chronic disease, injury risk, and mortality. Retaining mobile populations is methodologically and ethically important.

Clinical safety note: Evidence about vulnerable populations can become misleading if those at highest risk are disproportionately lost.

MCCQE exam trap: Loss to follow-up matters most when it is related to both exposure and outcome risk.

Authoritative references:

Question 23 of 60

Researchers study long-term survival after severe sepsis but include only patients who survived long enough to attend a 3-month follow-up clinic. What bias is introduced?

A. Survivorship bias B. Recall bias C. Confounding by age D. Observer bias E. Ecological fallacy

Correct answer: A — Survivorship bias

A–E option analysis

A.Correct. By restricting the sample to people who survived to follow-up, the study excludes early deaths and creates a selected survivor population that is healthier than the full cohort.
B.Incorrect. The central issue is conditioning on survival, not memory. By restricting the sample to people who survived to follow-up, the study excludes early deaths and creates a selected survivor population that is healthier than the full cohort.
C.Incorrect. Age may influence survival, but the design itself excludes early deaths. By restricting the sample to people who survived to follow-up, the study excludes early deaths and creates a selected survivor population that is healthier than the full cohort.
D.Incorrect. The outcome is distorted by who remains available for analysis, not assessor expectations. By restricting the sample to people who survived to follow-up, the study excludes early deaths and creates a selected survivor population that is healthier than the full cohort.
E.Incorrect. This is not a group-level inference problem. By restricting the sample to people who survived to follow-up, the study excludes early deaths and creates a selected survivor population that is healthier than the full cohort.

Disease / clinical review: Severe sepsis has high early mortality and substantial long-term morbidity among survivors.

Clinical safety note: Survivor-only estimates can understate true mortality and complication burden, affecting counselling and follow-up planning.

MCCQE exam trap: Ask who had to survive or remain event-free long enough to enter the analysis.

Authoritative references:

Question 24 of 60

A study of postoperative complications excludes all patients who die within 48 hours of surgery because they 'do not have enough follow-up.' What is the major concern?

A. Recall bias B. Non-differential misclassification C. Confounding D. Hawthorne effect E. Survivorship selection bias

Correct answer: E — Survivorship selection bias

A–E option analysis

A.Incorrect. No historical recall is involved. Excluding early deaths removes patients with the worst outcomes and can make the procedure appear safer than it truly is.
B.Incorrect. The key problem is exclusion based on early survival. Excluding early deaths removes patients with the worst outcomes and can make the procedure appear safer than it truly is.
C.Incorrect. Confounding may exist, but conditioning on survival is the direct design problem. Excluding early deaths removes patients with the worst outcomes and can make the procedure appear safer than it truly is.
D.Incorrect. Participants' behavior is not changing because of observation. Excluding early deaths removes patients with the worst outcomes and can make the procedure appear safer than it truly is.
E.Correct. Excluding early deaths removes patients with the worst outcomes and can make the procedure appear safer than it truly is.

Disease / clinical review: Postoperative mortality and major complications are patient-important safety outcomes and should be incorporated into outcome definitions.

Clinical safety note: Removing severe harms from analysis can directly mislead clinical risk-benefit decisions.

MCCQE exam trap: Early deaths are outcomes, not inconvenient missing data.

Authoritative references:

Question 25 of 60

A cohort evaluates whether a medication causes arrhythmia. Patients must remain on the medication for 6 months before they are classified as exposed, and any arrhythmias occurring during those 6 months are counted in the unexposed group. Which bias is likely?

A. Recall bias B. Immortal-time bias C. Lead-time bias D. Berkson bias E. Effect modification

Correct answer: B — Immortal-time bias

A–E option analysis

A.Incorrect. The issue is how exposure time is classified, not participant memory. To be classified as exposed, patients must survive and remain event-free long enough to reach 6 months. Misclassifying this guaranteed survival period can make exposure look artificially protective.
B.Correct. To be classified as exposed, patients must survive and remain event-free long enough to reach 6 months. Misclassifying this guaranteed survival period can make exposure look artificially protective.
C.Incorrect. Lead-time bias is linked to earlier diagnosis in screening. To be classified as exposed, patients must survive and remain event-free long enough to reach 6 months. Misclassifying this guaranteed survival period can make exposure look artificially protective.
D.Incorrect. Berkson bias is a hospital-selection phenomenon. To be classified as exposed, patients must survive and remain event-free long enough to reach 6 months. Misclassifying this guaranteed survival period can make exposure look artificially protective.
E.Incorrect. There is no evidence that the true treatment effect differs across subgroups. To be classified as exposed, patients must survive and remain event-free long enough to reach 6 months. Misclassifying this guaranteed survival period can make exposure look artificially protective.

Disease / clinical review: Arrhythmia risk can be influenced by medications, structural heart disease, electrolytes, and comorbidity. Time-varying exposure analysis is often necessary.

Clinical safety note: Immortal-time bias can falsely reassure clinicians about medication safety.

MCCQE exam trap: Time-dependent exposures require correct person-time classification.

Authoritative references:

Question 26 of 60

A registry compares cancer survival in people who use an online patient portal with those who do not. Portal users are younger, wealthier, more educated, and healthier at baseline. What broad bias problem is most likely?

A. Recall bias only B. Lead-time bias only C. Selection bias with confounding D. Non-differential misclassification only E. Effect modification only

Correct answer: C — Selection bias with confounding

A–E option analysis

A.Incorrect. The major differences are baseline selection and confounding, not memory. Use of the portal is self-selected and strongly associated with prognostic factors. Crude survival differences therefore cannot be attributed to portal use alone.
B.Incorrect. Earlier diagnosis is not the only or primary issue described. Use of the portal is self-selected and strongly associated with prognostic factors. Crude survival differences therefore cannot be attributed to portal use alone.
C.Correct. Use of the portal is self-selected and strongly associated with prognostic factors. Crude survival differences therefore cannot be attributed to portal use alone.
D.Incorrect. Measurement error is not the central problem. Use of the portal is self-selected and strongly associated with prognostic factors. Crude survival differences therefore cannot be attributed to portal use alone.
E.Incorrect. No true subgroup-specific treatment effect has been demonstrated. Use of the portal is self-selected and strongly associated with prognostic factors. Crude survival differences therefore cannot be attributed to portal use alone.

Disease / clinical review: Cancer survival depends on stage, tumor biology, treatment, comorbidity, access, and socioeconomic factors.

Clinical safety note: Do not conclude a digital intervention improves survival without accounting for major baseline differences and access inequities.

MCCQE exam trap: Digital-health exposure is often a marker of access, literacy, socioeconomic status, and baseline health.

Authoritative references:

Question 27 of 60

A longitudinal depression study retains nearly all participants with mild symptoms but loses many participants with severe symptoms who are hospitalized or disengage from care. What bias is most likely?

A. Recall bias B. Differential attrition bias C. Selection bias only at baseline D. Length-time bias E. Ecological fallacy

Correct answer: B — Differential attrition bias

A–E option analysis

A.Incorrect. The principal problem is who remains in follow-up. When dropout is strongly related to severity and likely outcome, the remaining sample becomes systematically different and outcome estimates may be overly favorable.
B.Correct. When dropout is strongly related to severity and likely outcome, the remaining sample becomes systematically different and outcome estimates may be overly favorable.
C.Incorrect. The imbalance arises after enrollment. When dropout is strongly related to severity and likely outcome, the remaining sample becomes systematically different and outcome estimates may be overly favorable.
D.Incorrect. No screening detection process is described. When dropout is strongly related to severity and likely outcome, the remaining sample becomes systematically different and outcome estimates may be overly favorable.
E.Incorrect. The analysis is individual-level. When dropout is strongly related to severity and likely outcome, the remaining sample becomes systematically different and outcome estimates may be overly favorable.

Disease / clinical review: Depression severity affects hospitalization, functioning, treatment adherence, and suicide risk.

Clinical safety note: Underrepresenting the sickest participants can dangerously underestimate illness burden or overestimate treatment effectiveness.

MCCQE exam trap: Missing data are not random when severe illness predicts dropout.

Authoritative references:

Question 28 of 60

A study of opioid overdose risk includes only people who successfully complete a 30-day detoxification program. Investigators then generalize the results to all people with opioid use disorder. What is the major concern?

A. Recall bias B. Selection bias from conditioning on treatment completion C. Observer bias D. Non-differential misclassification E. Effect modification

Correct answer: B — Selection bias from conditioning on treatment completion

A–E option analysis

A.Incorrect. The defining problem is selection based on completing treatment. Treatment completers are a selected group who may differ in motivation, severity, social support, tolerance, and relapse risk from people who do not complete detoxification.
B.Correct. Treatment completers are a selected group who may differ in motivation, severity, social support, tolerance, and relapse risk from people who do not complete detoxification.
C.Incorrect. Outcome assessment is not the central issue. Treatment completers are a selected group who may differ in motivation, severity, social support, tolerance, and relapse risk from people who do not complete detoxification.
D.Incorrect. Measurement error is not the main problem. Treatment completers are a selected group who may differ in motivation, severity, social support, tolerance, and relapse risk from people who do not complete detoxification.
E.Incorrect. The study has not shown that a true effect differs by subgroup. Treatment completers are a selected group who may differ in motivation, severity, social support, tolerance, and relapse risk from people who do not complete detoxification.

Disease / clinical review: Opioid overdose risk can increase after abstinence because tolerance falls. Completion status and post-discharge follow-up are clinically important.

Clinical safety note: Generalizing from selected treatment completers may underestimate overdose risk in people who disengage from care.

MCCQE exam trap: Conditioning on successful completion can produce a healthier, more adherent study population.

Authoritative references:

Question 29 of 60

A study finds coffee consumption is associated with myocardial infarction. Coffee drinkers are substantially more likely to smoke, and smoking independently increases myocardial infarction risk. What is smoking in this relationship?

A. An effect modifier necessarily B. A mediator C. Selection bias D. Random error E. A confounder

Correct answer: E — A confounder

A–E option analysis

A.Incorrect. Effect modification requires evidence that the coffee effect truly differs by smoking strata, not merely that smoking is related to both variables. Smoking is associated with the exposure and independently affects the outcome, so it can distort the observed coffee–myocardial infarction association.
B.Incorrect. A mediator lies on the causal pathway from coffee to myocardial infarction; smoking is not caused by coffee in the proposed model. Smoking is associated with the exposure and independently affects the outcome, so it can distort the observed coffee–myocardial infarction association.
C.Incorrect. The issue is distortion by a third variable, not who entered the study. Smoking is associated with the exposure and independently affects the outcome, so it can distort the observed coffee–myocardial infarction association.
D.Incorrect. The systematic relationship is not explained by sampling variation alone. Smoking is associated with the exposure and independently affects the outcome, so it can distort the observed coffee–myocardial infarction association.
E.Correct. Smoking is associated with the exposure and independently affects the outcome, so it can distort the observed coffee–myocardial infarction association.

Disease / clinical review: Smoking is a strong cardiovascular risk factor and can confound many lifestyle associations.

Clinical safety note: Do not give causal lifestyle advice from a crude association when a strong confounder has not been controlled.

MCCQE exam trap: A confounder is not simply 'another risk factor'; it must be related to both exposure and outcome and not lie on the causal pathway.

Authoritative references:

Question 30 of 60

Researchers find that exercise is associated with lower mortality. Exercisers are younger on average than nonexercisers, and age strongly predicts mortality. What variable should be considered a potential confounder?

A. Mortality B. Age C. Exercise D. Follow-up time E. Study sample size

Correct answer: B — Age

A–E option analysis

A.Incorrect. Mortality is the outcome, not a third variable causing distortion. Age is associated with both exercise behavior and mortality and can create a spurious or distorted crude association.
B.Correct. Age is associated with both exercise behavior and mortality and can create a spurious or distorted crude association.
C.Incorrect. Exercise is the exposure of interest. Age is associated with both exercise behavior and mortality and can create a spurious or distorted crude association.
D.Incorrect. Follow-up time affects opportunity to observe events but is not the key third variable described. Age is associated with both exercise behavior and mortality and can create a spurious or distorted crude association.
E.Incorrect. Sample size affects precision, not confounding in this scenario. Age is associated with both exercise behavior and mortality and can create a spurious or distorted crude association.

Disease / clinical review: Physical activity is associated with important health benefits, but age and baseline health strongly influence both activity levels and mortality.

Clinical safety note: Adjustment for age is important before translating observational exercise associations into quantitative risk estimates.

MCCQE exam trap: Always identify exposure, outcome, and the third variable separately.

Authoritative references:

Question 31 of 60

A study examines whether prenatal vitamin use lowers neural-tube-defect risk. Women who take vitamins are also more likely to attend early prenatal care, avoid smoking, and have higher income. What problem is most important?

A. Recall bias only B. Lead-time bias C. Selection bias only D. Confounding by health behavior and socioeconomic factors E. Effect modification

Correct answer: D — Confounding by health behavior and socioeconomic factors

A–E option analysis

A.Incorrect. Recall may matter, but the scenario explicitly describes systematic baseline differences related to exposure and outcome. Vitamin use may be a marker for multiple health-promoting behaviors and social advantages that independently affect pregnancy outcomes.
B.Incorrect. Neural-tube defects are not being made to appear better by earlier diagnosis. Vitamin use may be a marker for multiple health-promoting behaviors and social advantages that independently affect pregnancy outcomes.
C.Incorrect. The issue described is third-variable distortion among enrolled participants. Vitamin use may be a marker for multiple health-promoting behaviors and social advantages that independently affect pregnancy outcomes.
D.Correct. Vitamin use may be a marker for multiple health-promoting behaviors and social advantages that independently affect pregnancy outcomes.
E.Incorrect. No evidence is given that the true vitamin effect differs across subgroups. Vitamin use may be a marker for multiple health-promoting behaviors and social advantages that independently affect pregnancy outcomes.

Disease / clinical review: Folate supplementation reduces neural-tube-defect risk, but observational estimates can still be confounded by broader health behaviors and access to care.

Clinical safety note: Use established evidence and guideline-based folate recommendations rather than relying on an uncontrolled observational estimate.

MCCQE exam trap: Healthy-user confounding is common in observational studies of preventive behaviors.

Authoritative references:

Question 32 of 60

Investigators report that statin users have lower mortality than nonusers. Statin users also see physicians more often, receive more preventive care, and adhere better to medications in general. Which phenomenon is this?

A. Recall bias B. Healthy-user confounding C. Berkson bias D. Lead-time bias E. Effect modification

Correct answer: B — Healthy-user confounding

A–E option analysis

A.Incorrect. No differential memory is described. People who adhere to preventive medications often engage in other health-promoting behaviors and healthcare use that can independently improve outcomes.
B.Correct. People who adhere to preventive medications often engage in other health-promoting behaviors and healthcare use that can independently improve outcomes.
C.Incorrect. There is no hospital-based case-control selection mechanism. People who adhere to preventive medications often engage in other health-promoting behaviors and healthcare use that can independently improve outcomes.
D.Incorrect. Earlier diagnosis is not the defining issue. People who adhere to preventive medications often engage in other health-promoting behaviors and healthcare use that can independently improve outcomes.
E.Incorrect. The differences are confounding characteristics, not proven subgroup-specific treatment effects. People who adhere to preventive medications often engage in other health-promoting behaviors and healthcare use that can independently improve outcomes.

Disease / clinical review: Statins reduce cardiovascular risk in appropriate patients, but observational mortality comparisons require careful adjustment for baseline risk, adherence, and healthcare behavior.

Clinical safety note: Do not overestimate medication benefit from uncontrolled healthy-user effects.

MCCQE exam trap: Medication use in observational data can be a marker for many unmeasured behaviors.

Authoritative references:

Question 33 of 60

A study evaluates whether chronic NSAID use causes kidney disease. Patients with chronic arthritis are more likely to use NSAIDs and also have systemic inflammatory disease that can itself affect kidney function. What is the systemic inflammatory disease?

A. A mediator necessarily B. The outcome C. A potential confounder D. Selection bias E. Randomization

Correct answer: C — A potential confounder

A–E option analysis

A.Incorrect. It is not necessarily caused by NSAID use and therefore is not clearly on the causal pathway. The inflammatory disease may influence NSAID use and independently affect kidney outcomes, creating confounding by indication or underlying disease severity.
B.Incorrect. Kidney disease is the outcome; inflammatory disease is a third variable. The inflammatory disease may influence NSAID use and independently affect kidney outcomes, creating confounding by indication or underlying disease severity.
C.Correct. The inflammatory disease may influence NSAID use and independently affect kidney outcomes, creating confounding by indication or underlying disease severity.
D.Incorrect. The distortion comes from a third variable rather than enrollment alone. The inflammatory disease may influence NSAID use and independently affect kidney outcomes, creating confounding by indication or underlying disease severity.
E.Incorrect. Randomization is a design method, not a variable in the causal structure. The inflammatory disease may influence NSAID use and independently affect kidney outcomes, creating confounding by indication or underlying disease severity.

Disease / clinical review: NSAIDs can cause kidney injury, but underlying inflammatory disease, age, volume status, and comorbidity also influence renal risk.

Clinical safety note: Clinical decisions should consider both medication toxicity and the underlying disease that prompted treatment.

MCCQE exam trap: Confounding by indication is especially important in medication-effectiveness research.

Authoritative references:

Question 34 of 60

Researchers find that people prescribed an antiarrhythmic drug have higher mortality than people not prescribed the drug. Patients receiving the drug had much more severe arrhythmias at baseline. What is the best explanation?

A. Recall bias B. Lead-time bias C. Non-differential misclassification D. Effect modification E. Confounding by indication

Correct answer: E — Confounding by indication

A–E option analysis

A.Incorrect. Treatment assignment is based on clinical severity, not memory. Sicker patients are more likely to receive the drug and also more likely to die, so treatment indication and disease severity confound the drug–mortality association.
B.Incorrect. No screening survival distortion is described. Sicker patients are more likely to receive the drug and also more likely to die, so treatment indication and disease severity confound the drug–mortality association.
C.Incorrect. The main issue is baseline prognosis, not measurement error. Sicker patients are more likely to receive the drug and also more likely to die, so treatment indication and disease severity confound the drug–mortality association.
D.Incorrect. No subgroup-specific true drug effect is demonstrated. Sicker patients are more likely to receive the drug and also more likely to die, so treatment indication and disease severity confound the drug–mortality association.
E.Correct. Sicker patients are more likely to receive the drug and also more likely to die, so treatment indication and disease severity confound the drug–mortality association.

Disease / clinical review: Antiarrhythmic drugs can have important benefits and harms, but baseline rhythm severity and structural heart disease strongly influence prognosis.

Clinical safety note: Do not conclude that a drug causes mortality simply because the treated group starts with higher clinical risk.

MCCQE exam trap: Treatment groups in observational data are rarely exchangeable by default.

Authoritative references:

Question 35 of 60

A cohort finds that breastfeeding is associated with higher childhood IQ. Mothers who breastfeed longer also have higher average educational attainment. Maternal education is associated with childhood cognitive outcomes. What role can maternal education play?

A. Mediator necessarily B. Confounder C. Outcome D. Selection bias E. Effect modifier necessarily

Correct answer: B — Confounder

A–E option analysis

A.Incorrect. Maternal education precedes breastfeeding and is not caused by breastfeeding. Maternal education is associated with breastfeeding behavior and with child cognitive outcomes, so it may confound the crude association.
B.Correct. Maternal education is associated with breastfeeding behavior and with child cognitive outcomes, so it may confound the crude association.
C.Incorrect. Child IQ is the outcome, not maternal education. Maternal education is associated with breastfeeding behavior and with child cognitive outcomes, so it may confound the crude association.
D.Incorrect. The scenario identifies a third variable related to exposure and outcome. Maternal education is associated with breastfeeding behavior and with child cognitive outcomes, so it may confound the crude association.
E.Incorrect. Effect modification would require evidence that the breastfeeding effect differs across education strata. Maternal education is associated with breastfeeding behavior and with child cognitive outcomes, so it may confound the crude association.

Disease / clinical review: Child development is influenced by genetic, social, nutritional, educational, and environmental factors, making observational causal interpretation complex.

Clinical safety note: Do not overstate causal developmental benefits from a single uncontrolled observational association.

MCCQE exam trap: Temporal order helps distinguish confounders from mediators.

Authoritative references:

Question 36 of 60

A study shows that air pollution is associated with asthma admissions. Polluted neighborhoods also have higher poverty, older housing, and more smoking exposure. What should investigators do before claiming a causal pollution effect?

A. Ignore other variables because the association is statistically significant B. Exclude all participants with asthma C. Match on every measured variable regardless of causal role D. Use only a larger sample size E. Assess and control important confounders

Correct answer: E — Assess and control important confounders

A–E option analysis

A.Incorrect. Statistical significance does not eliminate confounding. Socioeconomic conditions, housing, and smoking may be related to both pollution exposure and asthma outcomes and therefore require causal consideration and appropriate control.
B.Incorrect. That would remove the outcome of interest. Socioeconomic conditions, housing, and smoking may be related to both pollution exposure and asthma outcomes and therefore require causal consideration and appropriate control.
C.Incorrect. Overadjustment or matching on mediators/colliders can introduce bias. Socioeconomic conditions, housing, and smoking may be related to both pollution exposure and asthma outcomes and therefore require causal consideration and appropriate control.
D.Incorrect. A larger sample improves precision but does not remove systematic confounding. Socioeconomic conditions, housing, and smoking may be related to both pollution exposure and asthma outcomes and therefore require causal consideration and appropriate control.
E.Correct. Socioeconomic conditions, housing, and smoking may be related to both pollution exposure and asthma outcomes and therefore require causal consideration and appropriate control.

Disease / clinical review: Asthma exacerbations are influenced by pollution, allergens, tobacco smoke, housing conditions, viral infections, and access to care.

Clinical safety note: Public-health recommendations should consider the full causal structure rather than relying on an unadjusted neighborhood association.

MCCQE exam trap: A large, precise estimate can still be precisely wrong because of confounding.

Authoritative references:

Question 37 of 60

Investigators study whether obesity causes knee osteoarthritis. They adjust for knee pain severity measured after osteoarthritis has already developed. Why can this be problematic?

A. It always eliminates confounding B. It is required for every observational study C. It converts the study into a randomized trial D. Adjustment for a consequence of the outcome can introduce bias E. It only affects precision, not validity

Correct answer: D — Adjustment for a consequence of the outcome can introduce bias

A–E option analysis

A.Incorrect. Adjusting for post-outcome variables can create rather than solve bias. Variables measured after the outcome or caused by the outcome can be colliders or consequences rather than confounders. Conditioning on them can distort the association.
B.Incorrect. Adjustment should be guided by a causal model, not performed indiscriminately. Variables measured after the outcome or caused by the outcome can be colliders or consequences rather than confounders. Conditioning on them can distort the association.
C.Incorrect. Statistical adjustment cannot create random assignment. Variables measured after the outcome or caused by the outcome can be colliders or consequences rather than confounders. Conditioning on them can distort the association.
D.Correct. Variables measured after the outcome or caused by the outcome can be colliders or consequences rather than confounders. Conditioning on them can distort the association.
E.Incorrect. Inappropriate adjustment can change the estimate systematically. Variables measured after the outcome or caused by the outcome can be colliders or consequences rather than confounders. Conditioning on them can distort the association.

Disease / clinical review: Obesity is an established risk factor for knee osteoarthritis, while pain severity is largely a manifestation of disease rather than a baseline confounder.

Clinical safety note: Poor causal adjustment can obscure true risk relationships and confuse counselling about modifiable risk factors.

MCCQE exam trap: Do not adjust for every available variable; adjust for variables justified by the causal structure.

Authoritative references:

Question 38 of 60

A study evaluates whether low birth weight increases adult hypertension risk. Investigators adjust for adult body mass index, which may partly lie on the causal pathway from early growth to later blood pressure. What concern arises?

A. Recall bias B. Selection bias C. Effect modification D. Random error E. Overadjustment for a mediator

Correct answer: E — Overadjustment for a mediator

A–E option analysis

A.Incorrect. The issue is causal adjustment, not memory. If adult body mass index is caused partly by early growth and itself influences blood pressure, adjusting for it may remove part of the total effect investigators want to estimate.
B.Incorrect. No enrollment distortion is described. If adult body mass index is caused partly by early growth and itself influences blood pressure, adjusting for it may remove part of the total effect investigators want to estimate.
C.Incorrect. The concern is blocking part of a causal pathway, not a true subgroup difference. If adult body mass index is caused partly by early growth and itself influences blood pressure, adjusting for it may remove part of the total effect investigators want to estimate.
D.Incorrect. The problem is systematic analytic bias, not chance. If adult body mass index is caused partly by early growth and itself influences blood pressure, adjusting for it may remove part of the total effect investigators want to estimate.
E.Correct. If adult body mass index is caused partly by early growth and itself influences blood pressure, adjusting for it may remove part of the total effect investigators want to estimate.

Disease / clinical review: Hypertension reflects lifelong genetic, developmental, behavioral, and metabolic influences.

Clinical safety note: Analytic choices should match the clinical question: total effect and direct effect are different estimands.

MCCQE exam trap: Confounders should be controlled; mediators usually should not be adjusted for when estimating the total causal effect.

Authoritative references:

Question 39 of 60

Researchers assess whether a new diabetes drug reduces cardiovascular events in observational data. They adjust for HbA1c measured 1 year after treatment starts. Treatment itself lowers HbA1c. Why might this adjustment be inappropriate when estimating the total treatment effect?

A. HbA1c is always a confounder B. The study becomes underpowered C. It creates recall bias D. HbA1c may be a mediator E. It guarantees no bias

Correct answer: D — HbA1c may be a mediator

A–E option analysis

A.Incorrect. A variable affected by treatment cannot be a baseline confounder for that treatment. Post-treatment HbA1c may lie on the pathway through which the drug affects cardiovascular outcomes. Adjusting for it can block part of the treatment effect.
B.Incorrect. Power is not the principal causal problem. Post-treatment HbA1c may lie on the pathway through which the drug affects cardiovascular outcomes. Adjusting for it can block part of the treatment effect.
C.Incorrect. HbA1c is measured objectively. Post-treatment HbA1c may lie on the pathway through which the drug affects cardiovascular outcomes. Adjusting for it can block part of the treatment effect.
D.Correct. Post-treatment HbA1c may lie on the pathway through which the drug affects cardiovascular outcomes. Adjusting for it can block part of the treatment effect.
E.Incorrect. Post-treatment adjustment can itself create bias. Post-treatment HbA1c may lie on the pathway through which the drug affects cardiovascular outcomes. Adjusting for it can block part of the treatment effect.

Disease / clinical review: Diabetes treatments can influence cardiovascular risk through glycemia and nonglycemic pathways.

Clinical safety note: Estimating the wrong causal effect can misrepresent a drug's clinical benefit.

MCCQE exam trap: Do not adjust for a post-treatment variable automatically just because it predicts outcome.

Authoritative references:

Question 40 of 60

A study finds a relationship between alcohol use and liver cancer. Hepatitis B infection is more common among heavy drinkers in the sample and independently causes liver cancer. What is the best next step?

A. Ignore hepatitis B because alcohol is the exposure of interest B. Exclude all participants with liver cancer C. Stratify or otherwise adjust for hepatitis B infection D. Increase sample size only E. Adjust for a post-cancer variable instead

Correct answer: C — Stratify or otherwise adjust for hepatitis B infection

A–E option analysis

A.Incorrect. A strong third-variable risk factor can distort the alcohol association. Hepatitis B is a plausible confounder because it is associated with the exposure distribution in this sample and is independently associated with liver cancer.
B.Incorrect. That removes the outcome. Hepatitis B is a plausible confounder because it is associated with the exposure distribution in this sample and is independently associated with liver cancer.
C.Correct. Hepatitis B is a plausible confounder because it is associated with the exposure distribution in this sample and is independently associated with liver cancer.
D.Incorrect. Precision does not remove confounding. Hepatitis B is a plausible confounder because it is associated with the exposure distribution in this sample and is independently associated with liver cancer.
E.Incorrect. Post-outcome adjustment may create additional bias. Hepatitis B is a plausible confounder because it is associated with the exposure distribution in this sample and is independently associated with liver cancer.

Disease / clinical review: Alcohol-associated liver disease and chronic viral hepatitis are both important causes of cirrhosis and hepatocellular carcinoma.

Clinical safety note: Risk counselling should account for multiple liver-cancer risk factors rather than attributing all risk to one exposure.

MCCQE exam trap: Confounding control can be performed at design or analysis, but the variable must first be identified correctly.

Authoritative references:

Question 41 of 60

A researcher asks whether sex is a confounder of the association between a drug and rash. Men and women receive the drug equally often, but rash risk differs by sex. Does sex meet the classic definition of a confounder in this scenario?

A. No, because it is not associated with the exposure B. Yes, because it predicts the outcome C. Yes, because sex should always be adjusted for D. Yes, because any baseline variable is a confounder E. No, because confounders must be caused by the exposure

Correct answer: A — No, because it is not associated with the exposure

A–E option analysis

A.Correct. A confounder must be associated with the exposure and independently associated with the outcome, while not being an intermediate on the causal pathway.
B.Incorrect. Predicting the outcome alone is not enough for classic confounding. A confounder must be associated with the exposure and independently associated with the outcome, while not being an intermediate on the causal pathway.
C.Incorrect. Covariate adjustment should be based on causal relevance, not automatic inclusion. A confounder must be associated with the exposure and independently associated with the outcome, while not being an intermediate on the causal pathway.
D.Incorrect. Baseline status alone does not define confounding. A confounder must be associated with the exposure and independently associated with the outcome, while not being an intermediate on the causal pathway.
E.Incorrect. Confounders generally precede exposure and are not caused by it. A confounder must be associated with the exposure and independently associated with the outcome, while not being an intermediate on the causal pathway.

Disease / clinical review: Drug rashes may vary by biological and clinical characteristics, but causal adjustment depends on whether those factors also influence treatment exposure.

Clinical safety note: Unnecessary adjustment can reduce efficiency or introduce bias if the variable is a collider or mediator.

MCCQE exam trap: An outcome predictor is not automatically a confounder.

Authoritative references:

Question 42 of 60

In an observational study of hormone therapy and stroke, women receiving therapy are younger and healthier at baseline than nonusers. The crude analysis suggests therapy is protective. After age and health status are adjusted for, the association disappears. What does this suggest?

A. The adjusted model must be wrong B. The exposure definitely has no biological effect C. Effect modification has been proven D. The crude association was confounded E. Recall bias has been eliminated

Correct answer: D — The crude association was confounded

A–E option analysis

A.Incorrect. A change after controlling important baseline differences is consistent with confounding. Large changes between crude and appropriately adjusted estimates can indicate that baseline differences were confounding the original association.
B.Incorrect. An adjusted null association does not prove absolute absence of any effect. Large changes between crude and appropriately adjusted estimates can indicate that baseline differences were confounding the original association.
C.Incorrect. Effect modification requires different stratum-specific effects rather than simply disappearance after adjustment. Large changes between crude and appropriately adjusted estimates can indicate that baseline differences were confounding the original association.
D.Correct. Large changes between crude and appropriately adjusted estimates can indicate that baseline differences were confounding the original association.
E.Incorrect. Statistical adjustment does not solve recall bias. Large changes between crude and appropriately adjusted estimates can indicate that baseline differences were confounding the original association.

Disease / clinical review: Stroke risk is strongly influenced by age, blood pressure, smoking, vascular disease, and other factors.

Clinical safety note: Treatment counselling should rely on well-controlled evidence because crude observational comparisons can be misleading.

MCCQE exam trap: Compare crude and adjusted estimates, but judge adjustment using causal reasoning rather than change-in-estimate alone.

Authoritative references:

Question 43 of 60

A randomized trial assigns participants to treatment by a concealed random sequence. Which major epidemiologic problem is randomization primarily intended to reduce?

A. Confounding B. Recall bias C. Observer bias D. Attrition bias E. Publication bias

Correct answer: A — Confounding

A–E option analysis

A.Correct. Randomization tends to distribute known and unknown baseline prognostic factors between groups by chance, reducing confounding.
B.Incorrect. Randomization does not improve memory of past exposures. Randomization tends to distribute known and unknown baseline prognostic factors between groups by chance, reducing confounding.
C.Incorrect. Blinding, not randomization, primarily addresses observer expectations. Randomization tends to distribute known and unknown baseline prognostic factors between groups by chance, reducing confounding.
D.Incorrect. Randomization does not prevent differential dropout after assignment. Randomization tends to distribute known and unknown baseline prognostic factors between groups by chance, reducing confounding.
E.Incorrect. Randomization within one study cannot determine which studies are published. Randomization tends to distribute known and unknown baseline prognostic factors between groups by chance, reducing confounding.

Disease / clinical review: Therapeutic trials use randomization to strengthen causal inference about treatment effects.

Clinical safety note: Even randomized trials still need good follow-up, valid outcome measurement, and safety monitoring.

MCCQE exam trap: Randomization does not guarantee perfect balance in every finite sample.

Authoritative references:

Question 44 of 60

Before starting a cohort study of alcohol and liver disease, investigators enroll only participants without chronic hepatitis B or C infection. Which confounding-control method is this?

A. Matching B. Restriction C. Stratification D. Randomization E. Multivariable regression

Correct answer: B — Restriction

A–E option analysis

A.Incorrect. Matching pairs or balances participants on confounder levels rather than excluding an entire level. Restriction limits enrollment to one level of a potential confounder, thereby preventing that variable from varying within the study.
B.Correct. Restriction limits enrollment to one level of a potential confounder, thereby preventing that variable from varying within the study.
C.Incorrect. Stratification is an analysis or sampling approach using separate levels of a variable. Restriction limits enrollment to one level of a potential confounder, thereby preventing that variable from varying within the study.
D.Incorrect. Exposure is not being assigned at random. Restriction limits enrollment to one level of a potential confounder, thereby preventing that variable from varying within the study.
E.Incorrect. No statistical model is being used in the design stage. Restriction limits enrollment to one level of a potential confounder, thereby preventing that variable from varying within the study.

Disease / clinical review: Viral hepatitis and alcohol can both cause chronic liver disease and liver cancer, so controlling viral infection may be important in observational research.

Clinical safety note: Design-stage control should not exclude clinically important populations without considering equity and applicability.

MCCQE exam trap: Restriction controls confounding but can reduce generalizability and prevents studying the restricted variable's effect.

Authoritative references:

Question 45 of 60

A case-control study of smoking and lung cancer pairs each case with a control of the same age and sex. Which method is being used to control confounding?

A. Restriction B. Matching C. Randomization D. Stratification only E. Blinding

Correct answer: B — Matching

A–E option analysis

A.Incorrect. Restriction would limit the study to one age or sex category rather than pairing participants. Matching makes cases and controls similar with respect to selected potential confounders such as age and sex.
B.Correct. Matching makes cases and controls similar with respect to selected potential confounders such as age and sex.
C.Incorrect. The study is observational and does not randomly assign smoking. Matching makes cases and controls similar with respect to selected potential confounders such as age and sex.
D.Incorrect. Stratification could be used analytically, but the described design technique is matching. Matching makes cases and controls similar with respect to selected potential confounders such as age and sex.
E.Incorrect. Blinding addresses measurement or performance bias rather than confounding by age and sex. Matching makes cases and controls similar with respect to selected potential confounders such as age and sex.

Disease / clinical review: Lung-cancer risk varies strongly with age and smoking exposure, making careful confounder control important.

Clinical safety note: Overmatching can reduce efficiency or obscure a true association.

MCCQE exam trap: Matched variables require appropriate analysis and cannot usually be evaluated as exposures in the same way.

Authoritative references:

Question 46 of 60

Investigators examine the association between obesity and hypertension separately among adults aged 18–39, 40–59, and ≥60 years. What confounding-control technique is this?

A. Restriction B. Matching C. Stratification D. Randomization E. Allocation concealment

Correct answer: C — Stratification

A–E option analysis

A.Incorrect. Restriction would include only one age category. Stratification estimates the exposure-outcome association within levels of a potential confounder and can reveal whether crude confounding is present.
B.Incorrect. Matching would pair or balance participants based on age. Stratification estimates the exposure-outcome association within levels of a potential confounder and can reveal whether crude confounding is present.
C.Correct. Stratification estimates the exposure-outcome association within levels of a potential confounder and can reveal whether crude confounding is present.
D.Incorrect. Obesity is not assigned at random. Stratification estimates the exposure-outcome association within levels of a potential confounder and can reveal whether crude confounding is present.
E.Incorrect. Allocation concealment applies to randomized treatment assignment. Stratification estimates the exposure-outcome association within levels of a potential confounder and can reveal whether crude confounding is present.

Disease / clinical review: Obesity and hypertension both vary with age, so age can confound their crude association.

Clinical safety note: Stratum-specific interpretation can be more clinically meaningful than a single crude estimate.

MCCQE exam trap: Stratification can also reveal effect modification if stratum-specific effects truly differ.

Authoritative references:

Question 47 of 60

A large cohort examines 12 potential confounders simultaneously when estimating the effect of air pollution on mortality. Which method is most practical?

A. Simple matching on all 12 variables B. Restriction to one level of every variable C. Randomization of air pollution exposure D. Blinding participants to their neighborhood E. Multivariable regression

Correct answer: E — Multivariable regression

A–E option analysis

A.Incorrect. Matching on many variables can be inefficient and difficult, especially with continuous covariates. Multivariable regression can estimate an exposure effect while statistically adjusting for multiple measured confounders, assuming the model is appropriately specified.
B.Incorrect. This would severely limit sample size and generalizability. Multivariable regression can estimate an exposure effect while statistically adjusting for multiple measured confounders, assuming the model is appropriately specified.
C.Incorrect. Assigning harmful pollution exposure is unethical and impractical. Multivariable regression can estimate an exposure effect while statistically adjusting for multiple measured confounders, assuming the model is appropriately specified.
D.Incorrect. Blinding cannot remove baseline confounding. Multivariable regression can estimate an exposure effect while statistically adjusting for multiple measured confounders, assuming the model is appropriately specified.
E.Correct. Multivariable regression can estimate an exposure effect while statistically adjusting for multiple measured confounders, assuming the model is appropriately specified.

Disease / clinical review: Air-pollution mortality studies must account for socioeconomic, smoking, geographic, and comorbidity differences.

Clinical safety note: Residual confounding should be acknowledged before converting an adjusted association into a strong causal claim.

MCCQE exam trap: Regression controls only measured and correctly modeled confounders.

Authoritative references:

Question 48 of 60

Which method can control both known and unknown confounders most effectively at the design stage when ethically feasible?

A. Randomization B. Restriction C. Matching D. Multivariable regression E. Stratification

Correct answer: A — Randomization

A–E option analysis

A.Correct. Randomization can balance both measured and unmeasured baseline factors in expectation and is therefore uniquely powerful against confounding.
B.Incorrect. Restriction controls only variables explicitly restricted. Randomization can balance both measured and unmeasured baseline factors in expectation and is therefore uniquely powerful against confounding.
C.Incorrect. Matching addresses only variables selected for matching. Randomization can balance both measured and unmeasured baseline factors in expectation and is therefore uniquely powerful against confounding.
D.Incorrect. Regression controls measured variables included in the model. Randomization can balance both measured and unmeasured baseline factors in expectation and is therefore uniquely powerful against confounding.
E.Incorrect. Stratification controls measured variables that are stratified. Randomization can balance both measured and unmeasured baseline factors in expectation and is therefore uniquely powerful against confounding.

Disease / clinical review: When a treatment question can be ethically randomized, an RCT often provides stronger causal evidence than observational adjustment.

Clinical safety note: Unknown confounding remains a major limitation of nonrandomized treatment comparisons.

MCCQE exam trap: Randomization reduces confounding; it does not eliminate every bias or guarantee perfect balance.

Authoritative references:

Question 49 of 60

A cohort study matches exposed and unexposed participants on age. Investigators later want to determine whether age itself is associated with the exposure. What problem can matching create?

A. Difficulty evaluating the matched variable as an exposure B. Recall bias C. Lead-time bias D. Attrition bias E. Publication bias

Correct answer: A — Difficulty evaluating the matched variable as an exposure

A–E option analysis

A.Correct. Because matching deliberately equalizes the distribution of age between groups, the study may no longer be able to estimate the natural exposure-age association.
B.Incorrect. Matching does not cause participants to remember differently. Because matching deliberately equalizes the distribution of age between groups, the study may no longer be able to estimate the natural exposure-age association.
C.Incorrect. No screening-related earlier diagnosis is involved. Because matching deliberately equalizes the distribution of age between groups, the study may no longer be able to estimate the natural exposure-age association.
D.Incorrect. Matching is performed at design, not during loss to follow-up. Because matching deliberately equalizes the distribution of age between groups, the study may no longer be able to estimate the natural exposure-age association.
E.Incorrect. This is unrelated to selective publication. Because matching deliberately equalizes the distribution of age between groups, the study may no longer be able to estimate the natural exposure-age association.

Disease / clinical review: Age is often a strong confounder, but unnecessary matching can reduce efficiency and limit what can be learned.

Clinical safety note: Overmatching can obscure clinically relevant differences and waste study resources.

MCCQE exam trap: Do not match on a variable unless there is a clear reason to control it.

Authoritative references:

Question 50 of 60

Researchers evaluating a prenatal exposure match cases and controls on a variable that is actually caused by the exposure and lies on the pathway to the outcome. What is the concern?

A. Improved control of confounding B. Reduced recall bias C. Elimination of selection bias D. Overmatching or inappropriate control of a mediator E. Guaranteed greater power

Correct answer: D — Overmatching or inappropriate control of a mediator

A–E option analysis

A.Incorrect. A mediator is not a confounder for the total effect and controlling it can block part of the causal pathway. Matching on a mediator can block part of the causal effect and create overmatching, leading to an attenuated or distorted estimate.
B.Incorrect. Matching does not correct differential memory. Matching on a mediator can block part of the causal effect and create overmatching, leading to an attenuated or distorted estimate.
C.Incorrect. Matching cannot eliminate all selection bias. Matching on a mediator can block part of the causal effect and create overmatching, leading to an attenuated or distorted estimate.
D.Correct. Matching on a mediator can block part of the causal effect and create overmatching, leading to an attenuated or distorted estimate.
E.Incorrect. Overmatching can actually reduce efficiency. Matching on a mediator can block part of the causal effect and create overmatching, leading to an attenuated or distorted estimate.

Disease / clinical review: Prenatal exposures may influence intermediate fetal or placental processes that then affect outcomes.

Clinical safety note: Incorrect causal adjustment can hide a genuine harmful or beneficial effect.

MCCQE exam trap: Confounders precede exposure; mediators are caused by exposure.

Authoritative references:

Question 51 of 60

A medication reduces stroke risk by 35% in adults younger than 65 but shows no reduction in adults 65 or older. The difference persists after appropriate control of confounding. What is the best interpretation?

A. Residual confounding necessarily B. Selection bias C. Effect modification by age D. Recall bias E. Non-differential misclassification

Correct answer: C — Effect modification by age

A–E option analysis

A.Incorrect. Residual confounding is possible but is not the best interpretation when a persistent true stratum difference is demonstrated. The treatment effect differs across age strata, which is effect modification or interaction by age.
B.Incorrect. The key finding is different treatment effects across age groups. The treatment effect differs across age strata, which is effect modification or interaction by age.
C.Correct. The treatment effect differs across age strata, which is effect modification or interaction by age.
D.Incorrect. The outcome is not based on differential memory. The treatment effect differs across age strata, which is effect modification or interaction by age.
E.Incorrect. Measurement error would not specifically explain a reproducible subgroup-specific effect. The treatment effect differs across age strata, which is effect modification or interaction by age.

Disease / clinical review: Stroke prevention benefits and harms can vary with age because baseline risk, competing risk, comorbidity, and pharmacology differ.

Clinical safety note: Subgroup-specific treatment effects can change the benefit-harm balance and may require age-tailored recommendations.

MCCQE exam trap: Effect modification should usually be reported, not 'adjusted away.'

Authoritative references:

Question 52 of 60

A study finds that smoking increases lung-cancer risk much more among workers exposed to asbestos than among workers without asbestos exposure. What phenomenon does this illustrate?

A. Effect modification or interaction B. Confounding only C. Recall bias D. Selection bias E. Regression to the mean

Correct answer: A — Effect modification or interaction

A–E option analysis

A.Correct. Asbestos exposure changes the magnitude of the smoking–lung-cancer association, indicating effect modification or interaction.
B.Incorrect. Confounding would distort a common effect; here the smoking effect itself differs by asbestos exposure. Asbestos exposure changes the magnitude of the smoking–lung-cancer association, indicating effect modification or interaction.
C.Incorrect. The scenario describes a difference in effect, not memory. Asbestos exposure changes the magnitude of the smoking–lung-cancer association, indicating effect modification or interaction.
D.Incorrect. No selection mechanism explains the subgroup-specific risk pattern. Asbestos exposure changes the magnitude of the smoking–lung-cancer association, indicating effect modification or interaction.
E.Incorrect. This is unrelated to repeated extreme measurements. Asbestos exposure changes the magnitude of the smoking–lung-cancer association, indicating effect modification or interaction.

Disease / clinical review: Smoking and asbestos both increase lung-cancer risk, and combined exposure can produce especially high risk.

Clinical safety note: Workers with combined exposures may require stronger prevention and surveillance strategies.

MCCQE exam trap: Ask whether stratum-specific effect estimates are meaningfully different, not merely whether the stratifying variable predicts outcome.

Authoritative references:

Question 53 of 60

A blood-pressure drug lowers systolic blood pressure by 12 mm Hg in patients with chronic kidney disease and by 11 mm Hg in patients without kidney disease. The confidence intervals substantially overlap. What is the most appropriate conclusion?

A. Chronic kidney disease is definitely an effect modifier B. The drug is ineffective in kidney disease C. Confounding has been proven D. There is no convincing evidence of effect modification E. Selection bias is certain

Correct answer: D — There is no convincing evidence of effect modification

A–E option analysis

A.Incorrect. The observed difference is small and may be compatible with chance. Similar effect estimates with overlapping uncertainty do not provide convincing evidence that chronic kidney disease modifies the drug's effect.
B.Incorrect. Both strata show blood-pressure lowering. Similar effect estimates with overlapping uncertainty do not provide convincing evidence that chronic kidney disease modifies the drug's effect.
C.Incorrect. A tiny subgroup difference does not establish confounding. Similar effect estimates with overlapping uncertainty do not provide convincing evidence that chronic kidney disease modifies the drug's effect.
D.Correct. Similar effect estimates with overlapping uncertainty do not provide convincing evidence that chronic kidney disease modifies the drug's effect.
E.Incorrect. No selection mechanism is described. Similar effect estimates with overlapping uncertainty do not provide convincing evidence that chronic kidney disease modifies the drug's effect.

Disease / clinical review: Blood-pressure response may vary with renal disease, volume status, medication class, and baseline pressure, but formal interaction evidence is needed.

Clinical safety note: False subgroup claims can lead to inappropriate withholding or preferential use of therapy.

MCCQE exam trap: Do not declare interaction from small numerical subgroup differences alone.

Authoritative references:

Question 54 of 60

A cohort finds that obesity is associated with diabetes with a relative risk of 2.0 in men and 4.5 in women, using adequately adjusted models. What is the next best interpretation?

A. Sex is necessarily a confounder B. The crude estimate should replace the stratified estimates C. Selection bias has been proven D. The findings must be due to recall bias E. Sex may be an effect modifier

Correct answer: E — Sex may be an effect modifier

A–E option analysis

A.Incorrect. Different stratum-specific effects suggest interaction, not merely confounding. Substantially different adjusted relative risks across sex strata suggest that the obesity–diabetes association may differ by sex.
B.Incorrect. Meaningful effect modification makes a single pooled estimate potentially misleading. Substantially different adjusted relative risks across sex strata suggest that the obesity–diabetes association may differ by sex.
C.Incorrect. No selection mechanism is specified. Substantially different adjusted relative risks across sex strata suggest that the obesity–diabetes association may differ by sex.
D.Incorrect. Neither exposure nor outcome measurement is described as differential by memory. Substantially different adjusted relative risks across sex strata suggest that the obesity–diabetes association may differ by sex.
E.Correct. Substantially different adjusted relative risks across sex strata suggest that the obesity–diabetes association may differ by sex.

Disease / clinical review: Obesity is a major risk factor for type 2 diabetes, and risk gradients can differ across biological and social subgroups.

Clinical safety note: Population prevention may need subgroup-sensitive absolute-risk communication rather than a single average effect.

MCCQE exam trap: When interaction is present, report stratum-specific effects rather than only one pooled estimate.

Authoritative references:

Question 55 of 60

In a trial, a new anticoagulant reduces thromboembolism more strongly in patients with atrial fibrillation than in patients without atrial fibrillation. What statistical approach directly evaluates whether the treatment effect differs between these groups?

A. Compare whether one subgroup p value is below 0.05 and the other is above 0.05 B. Adjust for atrial fibrillation only as a main effect C. Increase sample size without modeling interaction D. Use a paired t test E. Test an interaction term

Correct answer: E — Test an interaction term

A–E option analysis

A.Incorrect. A difference in separate significance levels does not prove the effects differ. An interaction term in an appropriate regression model tests whether the treatment effect differs according to subgroup status.
B.Incorrect. A main-effect term does not directly test whether treatment effects differ by atrial-fibrillation status. An interaction term in an appropriate regression model tests whether the treatment effect differs according to subgroup status.
C.Incorrect. A larger sample alone does not answer the interaction question. An interaction term in an appropriate regression model tests whether the treatment effect differs according to subgroup status.
D.Incorrect. A paired t test is not the appropriate method for subgroup interaction in this treatment-outcome setting. An interaction term in an appropriate regression model tests whether the treatment effect differs according to subgroup status.
E.Correct. An interaction term in an appropriate regression model tests whether the treatment effect differs according to subgroup status.

Disease / clinical review: Atrial fibrillation substantially changes thromboembolic risk and may alter absolute treatment benefit even when relative effects are similar.

Clinical safety note: Subgroup claims should be supported by proper interaction testing to avoid unsafe overinterpretation.

MCCQE exam trap: 'Significant in one subgroup, nonsignificant in another' is not itself a valid interaction test.

Authoritative references:

Question 56 of 60

A vaccine has a relative efficacy of 80% in both younger and older adults, but because baseline disease risk is much higher in older adults, the absolute risk reduction is greater in older adults. Does this prove effect modification on the relative scale?

A. Yes, because absolute benefit differs B. Yes, because older adults have higher baseline risk C. Yes, because number needed to treat differs D. No E. No, because vaccines can never have effect modification

Correct answer: D — No

A–E option analysis

A.Incorrect. Different absolute effects can occur purely because baseline risks differ even when the relative effect is constant. Effect modification depends on the effect scale. Equal relative efficacy with different baseline risks can produce different absolute risk reductions without relative-scale interaction.
B.Incorrect. Higher baseline risk changes absolute benefit but does not necessarily modify relative efficacy. Effect modification depends on the effect scale. Equal relative efficacy with different baseline risks can produce different absolute risk reductions without relative-scale interaction.
C.Incorrect. NNT depends on absolute risk and can differ without relative-scale interaction. Effect modification depends on the effect scale. Equal relative efficacy with different baseline risks can produce different absolute risk reductions without relative-scale interaction.
D.Correct. Effect modification depends on the effect scale. Equal relative efficacy with different baseline risks can produce different absolute risk reductions without relative-scale interaction.
E.Incorrect. Vaccines can have effect modification; it simply is not established on the relative scale here. Effect modification depends on the effect scale. Equal relative efficacy with different baseline risks can produce different absolute risk reductions without relative-scale interaction.

Disease / clinical review: Older adults often have higher baseline risk of severe infectious outcomes, which can increase absolute vaccine benefit even if relative efficacy is similar.

Clinical safety note: Clinical decisions often depend on absolute benefit, so absence of relative interaction does not mean equal clinical value across risk groups.

MCCQE exam trap: Always state the effect scale when discussing interaction.

Authoritative references:

Question 57 of 60

A study of hormone therapy finds an odds ratio for venous thromboembolism of 1.2 in women without inherited thrombophilia and 6.0 in women with thrombophilia. The difference is statistically supported. What should investigators report?

A. Only one overall pooled odds ratio B. Only the p value for the full cohort C. Stratum-specific effects and the interaction D. Remove thrombophilia from the analysis E. Treat thrombophilia solely as a confounder and adjust it away

Correct answer: C — Stratum-specific effects and the interaction

A–E option analysis

A.Incorrect. Pooling can obscure a clinically important subgroup difference. Supported effect modification should be presented with subgroup-specific effect estimates and appropriate interaction information.
B.Incorrect. A single p value does not describe effect magnitude by subgroup. Supported effect modification should be presented with subgroup-specific effect estimates and appropriate interaction information.
C.Correct. Supported effect modification should be presented with subgroup-specific effect estimates and appropriate interaction information.
D.Incorrect. The subgroup difference is clinically important and should not be discarded. Supported effect modification should be presented with subgroup-specific effect estimates and appropriate interaction information.
E.Incorrect. When the true exposure effect differs by thrombophilia status, it is an effect modifier and should be reported. Supported effect modification should be presented with subgroup-specific effect estimates and appropriate interaction information.

Disease / clinical review: Estrogen exposure and inherited thrombophilia can combine to produce substantially increased venous-thromboembolism risk.

Clinical safety note: Clinically important interaction may alter contraceptive or hormone-therapy counselling and safety decisions.

MCCQE exam trap: Confounding is controlled; effect modification is described.

Authoritative references:

Question 58 of 60

A pollutant is associated with asthma exacerbations only among children with a specific genetic variant, while no association is seen among children without the variant. Assuming valid measurement and control of confounding, what concept is illustrated?

A. Gene–environment interaction B. Recall bias C. Selection bias D. Non-differential misclassification E. Confounding only

Correct answer: A — Gene–environment interaction

A–E option analysis

A.Correct. The genetic variant modifies the effect of environmental exposure on asthma risk, an example of gene–environment interaction.
B.Incorrect. The difference is biological effect heterogeneity, not memory. The genetic variant modifies the effect of environmental exposure on asthma risk, an example of gene–environment interaction.
C.Incorrect. No selective enrollment mechanism is described. The genetic variant modifies the effect of environmental exposure on asthma risk, an example of gene–environment interaction.
D.Incorrect. Measurement error would not specifically explain a validated genotype-specific association. The genetic variant modifies the effect of environmental exposure on asthma risk, an example of gene–environment interaction.
E.Incorrect. The pollutant effect itself differs by genotype, which is interaction. The genetic variant modifies the effect of environmental exposure on asthma risk, an example of gene–environment interaction.

Disease / clinical review: Asthma results from interactions among genetic susceptibility, allergens, infections, tobacco smoke, pollution, and other exposures.

Clinical safety note: Genetic subgroup findings require careful validation before they are used to change care or policy.

MCCQE exam trap: Interaction can be biological, clinical, or statistical and may depend on the scale used.

Authoritative references:

Question 59 of 60

Investigators stratify a treatment effect by diabetes status. The stratum-specific relative risks are 0.60 in patients with diabetes and 0.61 in patients without diabetes, while the crude relative risk is 0.80. What pattern is most consistent with these data?

A. Strong effect modification B. No confounding C. Confounding without meaningful effect modification D. Recall bias E. Selection bias

Correct answer: C — Confounding without meaningful effect modification

A–E option analysis

A.Incorrect. The stratum-specific effects are essentially the same. Similar stratum-specific effects suggest little interaction, while the difference between crude and stratum-specific estimates suggests confounding by diabetes status or variables associated with it.
B.Incorrect. The crude estimate differs materially from the similar stratum-specific estimates. Similar stratum-specific effects suggest little interaction, while the difference between crude and stratum-specific estimates suggests confounding by diabetes status or variables associated with it.
C.Correct. Similar stratum-specific effects suggest little interaction, while the difference between crude and stratum-specific estimates suggests confounding by diabetes status or variables associated with it.
D.Incorrect. The pattern concerns crude versus stratified effect estimates, not memory. Similar stratum-specific effects suggest little interaction, while the difference between crude and stratum-specific estimates suggests confounding by diabetes status or variables associated with it.
E.Incorrect. No selection mechanism is specified. Similar stratum-specific effects suggest little interaction, while the difference between crude and stratum-specific estimates suggests confounding by diabetes status or variables associated with it.

Disease / clinical review: Diabetes strongly affects cardiovascular risk and may distort crude treatment-outcome associations if treatment use also differs by diabetes status.

Clinical safety note: Use adjusted or stratified estimates when confounding is present; a crude estimate may mislead treatment decisions.

MCCQE exam trap: Similar stratum effects + different crude effect points toward confounding.

Authoritative references:

Question 60 of 60

A treatment reduces mortality by a relative risk of 0.70 in every subgroup, but the absolute risk reduction ranges from 1% in low-risk patients to 10% in high-risk patients. Which statement is best?

A. Clinical benefit differs on the absolute scale despite a stable relative effect B. There is no subgroup difference of any clinical importance C. The treatment has contradictory results D. The data prove selection bias E. The treatment is ineffective in low-risk patients

Correct answer: A — Clinical benefit differs on the absolute scale despite a stable relative effect

A–E option analysis

A.Correct. A constant relative treatment effect applied to different baseline risks naturally produces different absolute risk reductions and NNTs.
B.Incorrect. Absolute benefit can differ substantially even when relative effect is constant. A constant relative treatment effect applied to different baseline risks naturally produces different absolute risk reductions and NNTs.
C.Incorrect. The relative and absolute findings are mathematically compatible. A constant relative treatment effect applied to different baseline risks naturally produces different absolute risk reductions and NNTs.
D.Incorrect. No selection mechanism is described. A constant relative treatment effect applied to different baseline risks naturally produces different absolute risk reductions and NNTs.
E.Incorrect. A 1% absolute risk reduction may still be beneficial depending on harms, cost, and preferences. A constant relative treatment effect applied to different baseline risks naturally produces different absolute risk reductions and NNTs.

Disease / clinical review: High-risk patients often gain greater absolute benefit from an equally effective preventive therapy.

Clinical safety note: Treatment decisions should weigh absolute benefit against harms rather than relying only on relative-risk reduction.

MCCQE exam trap: Relative-scale homogeneity does not imply equal absolute benefit.

Authoritative references:

Bias, Confounding & Effect Modification

Module 2 Result