EPID 621 Midterm

What is the counterfactual definition of a cause?

A cause is something that, if it had not occurred, the outcome would not have happened (at least not in the same way or time).

How does the counterfactual definition of a cause underlie our understanding of causal inference in a cohort study and an RCT?

In cohort studies, we compare exposed and unexposed groups to infer causality. In RCTs, randomization helps mimic the counterfactual by ensuring groups are comparable.

What is the difference between a predictor/risk factor and a cause? Why is this distinction important?

A predictor is associated with an outcome but may not be causal. A cause directly influences the outcome. This distinction is crucial for targeting interventions.

What are sufficient component causes? What are necessary component causes?

• Sufficient cause: A combination of factors that inevitably leads to disease.

• Necessary cause: A factor required for disease but not always sufficient on its own.

How does the sufficient component cause model explain differences in effect sizes (RR, IRR, OR) across populations?

Different populations have varying prevalence of co-factors, influencing how strongly an exposure leads to disease.

What is exchangeability? How does this concept relate to an RCT and a cohort study?

Exchangeability means that groups being compared are similar in all relevant aspects except exposure. RCTs achieve this through randomization, while cohort studies try to control confounding.

What is the difference between age, period, and cohort effects?

• Age effect: Changes due to aging.

• Period effect: Changes due to historical factors.

• Cohort effect: Differences among birth cohorts.

How can age, period, and cohort effects be graphically assessed?

Using age-period-cohort (APC) models and plotting trends over time.This allows researchers to visualize and separate the influences of age, historical events, and specific birth cohorts on health outcomes.

What are the advantages of an RCT compared to a cohort or case-control study?

RCTs reduce confounding and selection bias but may be expensive and unethical for some exposures.

What is the difference between intention-to-treat and per-protocol analysis in an RCT?

• Intention-to-treat: Includes all participants as randomized, reducing bias.

• Per-protocol: Analyzes only those who followed the protocol, which may introduce bias.

What are the benefits of blinding in RCTs and observational studies?

Blinding reduces bias from participants, researchers, and analysts.

What are the strengths and limitations of an RCT, cohort study, and case-control study?

• RCTs: Strong internal validity, expensive, ethical constraints. (measuring exposure without outcome)

• Cohort: Good for rare exposures, susceptible to loss to follow-up. (measuring exposure without outcome)

• Case-control: Efficient for rare diseases, prone to recall bias. (measuring exposure after outcome)

How do you calculate OR, RR, RD, and incidence proportion from a 2x2 table?

• Odds Ratio (OR): (a/c) / (b/d) = ad/bc

• Risk Ratio (RR): (a/(a+b)) / (c/(c+d))

• Risk Difference (RD): (a/(a+b)) - (c/(c+d))

• Incidence Proportion: a/(a+b)

What is the difference between an incidence rate and an incidence proportion?

• Incidence proportion: The probability of disease over a time period.

• Incidence rate: Cases per person-time.

What is the purpose of survival analysis?

To analyze time-to-event data while accounting for censored observations.

What are censored observations in survival analysis?

Individuals who do not experience the event before the study ends.

What are the differences between Kaplan-Meier and Lifetable methods?

Kaplan-Meier updates survival probability at each event, while Lifetable groups data into intervals.

Why are Kaplan-Meier and Lifetable approaches considered non-parametric?

They do not assume an underlying distribution of survival times.

What statistical tests compare survival curves?

Log-rank test and Wilcoxon test.

What are different sampling strategies in case-control studies?

• Traditional: Controls selected at end of study.

• Case-cohort: Controls selected at the start.

• Nested case-control: Controls selected at time cases occur.

What is the rule of thumb in selecting controls for a case-control study?

Controls should represent the exposure distribution of the source population.

What is precision, and how can it be increased in a study?

Precision refers to the reproducibility of results. It can be increased by larger sample size and better measurement techniques.

What is the difference between internal validity and generalizability (external validity)?

• Internal validity: Accuracy of causal inference.

• External validity: Applicability to other populations.

What is the correct interpretation of a p-value and a 95% confidence interval?

• P-value: Probability of obtaining observed results under the null hypothesis.

• 95% CI: Range where the true effect likely falls 95% of the time.

What factors affect sample size and power in a study?

Proportion (D+, E+), effect size, Beta level (Type 2 error), alpha level, and power.

What is selection bias, and how can it be reduced?

Systematic differences in who is included in a study; minimized through careful recruitment and follow-up.

What is information bias, and what is the difference between differential and non-differential misclassification?

• Information bias: Errors in measurement.

• Differential: Error differs by exposure or outcome status.

• Non-differential: Error is the same across groups.

What is confounding, and how is it controlled in RCTs, cohort studies, and case-control studies?

Confounding is a third variable distorting the exposure-disease relationship. Controlled by randomization (RCT), matching, stratification, and statistical adjustment.