21. Confounding

Question 1

We say that the study factor is associated with the study outcome when what happens in a cohort study? case control study?

Answer 1

cohort study: RR not equal to 1

case/control study: OR not equal to 1

Question 2

when you have an association in the sample but not the population, what happened?

Answer 2

a spurious association due to:
chance, bias

• lack of repeatability

Question 3

confounder

Answer 3

variable entangled w/ study factor that masks the true relationship b/w the study factor and outcome

Question 4

criteria for confounding

Answer 4

• confounder must differ by levels of exposure variable (associated with exposure)
• counfounder must be associated with study outcome

Question 5

what is the effect of confounding?

Answer 5

leads to wrong conclusions of the association of interest

Question 6

how can we adjust for confounding?

Answer 6

• during study design:
• –restrict (limit study to a narrow range)
• –match (force balance, used most often in case-control studies)
• –random allocation in RCT (spreads the unknown and known confounding variables equally across the two arms)
• after data are collected
• –stratify
• –adjust (combine stratified results into 1 estimate) (answers question: what would the association be if the groups did not differ by the confounder?)

Question 7

multivariate regression adjustment

Answer 7

statistically model relationships of exposure and confounders to disease, fix confounders

Question 8

when not to summarize or adjust for another variable?

Answer 8

intermediate variable: the variable to be adjusted lies between the study factor and outcome in a causal chain

• eg high cholesterol -> heart disease -> death
• thus if analyzing whether cholesterol is a RF for death, don’t control for the presence of heart disease
• this is not a confounding variable

Effect modification: the RR’s differ importantly across levels of stratification variable (also called interaction)

• effect is not homogenous
• i.e. degree of association depends on the level of the stratification variable
• eg in bald example, effect modification would exist if RRold was different than RRyoung

Question 9

bias vs chance in study design?

Answer 9

bias in study design is uncorrectable statistically and can lead to a statistically significant association but not a real or causal association

chance in study design can lead to a statistically significant association but not a real or causal association

Question 10

bias

Answer 10

systematic error in design/conduct/analysis of a study that results in a mistaken estimate of an exposure’s effect on the risk of the disease or outcome

Question 11

the association b/w estrogen use and endometrial cancer could be challenged on the basis of what type of bias?

Answer 11

selection bias

Question 12

if stratum specific estimates in a study are NOT the same, what is present?

Answer 12

effect modificaiton (report stratum-specific estimates)

Question 13

if stratum specific estimates in a study ARE the same but are DIFFERENT from the overall estimate, what is present?

Answer 13

confounding (adjust)

Question 14

if stratum specific estimates in a study ARE the same and ARE NOT DIFFERENT from the overall estimate, what is present?

Answer 14

no confounding, do not adjust