Question 1

Which two of the following methods cannot be used to prevent confounding from occurring in a study?&#10;A)Randomisation&#10;B)Restriction&#10;C)Stratification&#10;D)Matching&#10;E)Multivariable analysis

Accepted Answer

Stratification and multivariable modelling techniques can be used to
control confounding in the data analysis, i.e.after the fact, but they do not prevent
confounding from being present in the raw study data.Each of the other strategies
potentially limits the degree to which confounding can occur through promoting closer
exchangeability of the compared groups.

Question 2

In a study of illicit drug use and attempted suicide the crude relative risk is 2.1 overall but it is 3.4 for men and 1.3 for women when the sexes are considered separately.
This suggests that:

A)There is confounding by sex in these data
B)There is confounding by some unknown or unmeasured factor in these data
C)There is evidence of effect modification by sex in these data
D)The results have been adjusted for age and sex
E)The results are due to bias

Accepted Answer

There is evidence of effect modification by sex in these data.The crude
relative risk (RR) is 2.1 but when we stratify by sex the RR estimates are quite different
for men (RR=3.4) and women (RR=1.3).When the stratum‐specific estimates differ
substantially across strata, this is evidence of effect modification and may provide
additional aetiological perspectives on the association between an exposure and an
outcome.Confounding by sex (A) or other factors (B), and other biases (E) may or may
not be present, but we have no relevant information to assess this.The description of
the RRs as 'crude' indicates that they had not been adjusted for potential confounders
(D).

Question 3

When does confounding occur?&#10;A)When the characteristics of the participants in the exposed group differ in some important way from those in the unexposed group&#10;B)When the characteristics of the participants in the study sample differ in some important way from those in the source population&#10;C)When the controls are not matched to the cases in a case-control study&#10;D)When the people lost to follow&#8208;up in a cohort study differ from those who are not lost to follow&#8208;up&#10;E)It occurs in all epidemiological research

Accepted Answer

Confounding will occur when the exposed and unexposed differ in terms of
some other characteristic that is associated with the outcome of interest, i.e.the
groups are not exchangeable.While a lack of matching in a case‐control study may lead
to confounding this is not inevitable (C), while (B) and (D) are issues of selection bias.
And although confounding can occur in all observational epidemiological research, it
should not be a problem in large randomised controlled trials (E).

Question 4

What is the best single way to prevent confounding from affecting the results of a study?&#10;A)Randomisation&#10;B)Restriction&#10;C)Stratification&#10;D)Matching&#10;E)Multivariable analysis

Accepted Answer

Randomisation is the best way to prevent confounding as this will maximise
the chances that all confounders (known and unknown) are balanced across the study
groups, thereby maximising their exchangeability.While the other options offer
approaches to managing confounding that can be very effective in particular situations,
they can only account for known confounders.

Question 5

Which of the following statements best describes the potential effects of confounding?&#10;A)In the presence of confounding the adjusted RR will usually be lower than the crude RR&#10;B)In the presence of confounding the adjusted RR will usually be closer to the null than the crude RR&#10;C)In the presence of confounding the adjusted RR will usually be higher than the crude RR&#10;D)In the presence of confounding the adjusted RR will usually be further from the null than the crude RR&#10;E)In the presence of confounding the adjusted RR can be higher or lower than the crude RR

Accepted Answer

Confounding can bias an estimate upwards or downwards, towards or away
from the null value.It is because of this unpredictability that special care is taken to
control for confounders when conducting a research study.Common ways to control
for confounding include: restriction, matching or randomisation (in RCTs) at the study
design stage, and stratification, direct or indirect adjustments and/or multivariable
analysis at the data analysis stage.If when the results are adjusted for confounders the
adjusted risk estimates differ from the crude estimates then confounding is present and
the adjusted RRs should be presented as they will be more accurate than the crude RRs.
[Note: this assumes that the variables adjusted for are confounders and not e.g.
intermediaries and that there is no appreciable effect modification.]

Question 6

An intermediary variable is not a confounder because:&#10;A)It is not associated with the outcome&#10;B)It is not associated with the exposure&#10;C)It lies on the causal pathway between the exposure and the outcome&#10;D)None of the above

Accepted Answer

For something to be a confounder it must be a risk factor for disease and
also associated with the exposure of interest and it must not be an intermediary
between exposure and the outcome (i.e., it must not lie on the causal pathway).Thus,
by definition, an intermediary variable cannot be a confounder.The fact that an
intermediary lies on the causal pathway between exposure and outcome means that it
will be associated with both the exposure (B) and outcome (A).

Question 7

Using the directed acyclic graph below to identify which of the following statements are true? (Select all that apply)  &#10;A)C is a potential confounder of the relationship between B and D&#10;B)B is a potential confounder of the relationship between C and D&#10;C)C is a potential confounder of the relationship between A and D&#10;D)C is an intermediary between A and D&#10;E)Even if C was not a cause of B, it could still confound the relationship between B and D

Accepted Answer

C is a potential confounder of the relationship between B and D
because C is a cause of both B and D (Answer A).B is not a confounder of the
relationship between C and D (Answer B) because it is on the causal pathway from C to
D.Similarly, C is not a confounder of the relationship between A and D (Answer C)
because it is on the causal pathway i.e.an intermediary (Answer D).Because A is a
cause of both B and C, it does not matter whether C is a cause of B, it can still confound
the relationship between B and D (Answer E).As an example of the last situation,
smoking can confound the relation between alcohol and lung cancer even though
smoking does not cause someone to drink alcohol, because there are upstream factors
such as socioeconomic status that influence both smoking and alcohol consumption.

Question 8

Why might adjustment for age affect the magnitude of the association between comorbidities and ovarian cancer?&#10;A)Because women with ovarian cancer are likely to have other comorbidities as well.&#10;B)Because older women have more comorbidities.&#10;C)Because older women are at greater risk of dying.&#10;D)Because older women have more comorbidities and are also at greater risk of dying.&#10;E)Age is not a confounder of the association between comorbidities and ovarian cancer.

Accepted Answer

Because age is associated with both the number of comorbidities at
diagnosis and with ovarian cancer mortality - older women have more comorbidities
and are also at greater risk of dying - it confounds the relation between the two.

Question 9

In a study of alcohol and oral cancer the crude relative risk is 4.0 overall but it is 2.0 for men and 2.0 for women when the sexes are considered separately.This suggests
That:

A)There is confounding by sex in these data
B)There is confounding by some unknown or unmeasured factor in these data
C)There is evidence of effect modification by sex in these data
D)The results have been adjusted for age and sex
E)The results are due to bias

Accepted Answer

There is confounding by sex in these data.The crude relative risk (RR) is 4.0
but when we stratify by sex the RR is 2.0 for both men and women.When the stratum‐
specific estimates differ from the crude estimate this is evidence of confounding.Note
that the estimates are the same in men and women so there is no evidence of effect
modification by sex (C); these limited data can say nothing about confounding by other
factors (B) nor about bias in general (E).The description of the RRs as 'crude' indicates
that they had not been adjusted for potential confounders (D).

Question 10

The table below comes from a study of ovarian cancer survival in Australia (Anuradha et al., 2015).
Risk of mortality in a prognostic cohort study of women with invasive epithelial ovarian
Cancer according to the number of comorbidities at time of diagnosis (Australia, 2005) These data suggest that (select all that apply):

A)Compared to women with no comorbidity, having 3 or more co‐morbidities at diagnosis was associated with higher risk of mortality, but this association was
Partially confounded by age
B)Compared to women with no comorbidity, having 3 or more comorbidities at diagnosis was associated with higher risk of mortality and there was no confounding
By age
C)Compared to women with no co‐morbidity, having a greater number of comorbidities at diagnosis was not associated with a higher risk of mortality
D)The trend of increasing mortality with greater comorbidity seen in the unadjusted data disappeared after adjustment for age

Accepted Answer

Having 3 or more comorbidities at diagnosis was fairly clearly related
to higher risk of mortality in the crude analysis but this association became much
weaker after adjustment for age suggesting there was some confounding by age in
these data.There is a modest trend of increasing risk in the crude data, which
essentially disappears with age‐adjustment.

Deck 8: Muddied Waters: The Challenge of Confounding