Question 1

You want to assess the effects of physical activity on study habits so you advertise at your local university for people to take part in your study.What type of selection bias
Might this lead to?

A)Loss to follow‐up
B)Volunteer bias
C)Recall bias
D)Observer bias
E)None of the above

Volunteer bias

Accepted Answer

Volunteer bias.People who volunteer to participate in studies (i.e.they
spontaneously offer their involvement rather than being selected from a formal
sampling strategy) are often very different from those who do not volunteer.They are
more likely to be health conscious, which in this study might lead to over‐estimation of
physical activity levels.A is not a problem unless you intend to follow‐up your study
participants up over time and C and D are forms of information bias.

Question 2

What term best describes the misclassification in the previous question?&#10;A)Differential&#10;B)Non&#8208;differential&#10;C)Systematic&#10;D)None of the above

Accepted Answer

Non‐differential misclassification.That is, the misclassification in exposure
was 'not different' between cases and controls.Note that the OR of 1.51 is closer to the
null value (OR=1.0) than the 'true' estimate of OR = 1.76.Non‐differential
misclassification almost always biases associations toward the null value.

Question 3

To determine attack rates for a respiratory disease of unknown origin among people attending a conference, random samples of guests staying at four hotels were
Surveyed for subsequent illness.Because it was not feasible to survey all guests,
Random sampling provided the best information because:

A)It would identify all cases of disease
B)It would avoid selection bias
C)It would reduce measurement bias
D)It would eliminate sampling error
E)It would eliminate confounding

It would avoid selection bias

Accepted Answer

It would avoid selection bias.By sampling guests at random we are most
likely to get a group that is truly representative of the whole population.Because we
did not survey everyone at the hotels it is very unlikely that we would identify all those
who developed the disease (A); and it does not affect measurement (C).The fact that
we are taking a sample, random or otherwise, introduces the potential for sampling
error (D) rather than eliminating it; although the larger the sample we take, the less
error we will have.Confounding (E) is not be addressed by random sampling.
(Note the important distinction between random sampling of people from a population
where we do not control who is and who is not exposed to the factor of interest, and
randomisation where we select people (possibly at random) and then randomly assign
them to the exposure of interest or the control group.It is the process of randomly
assigning people to the exposure groups that controls for confounding.)

Question 4

In measuring an exposure in a prospective cohort study design, which biases are most readily avoidable?&#10;A)Misclassification&#10;B)Interviewer bias&#10;C)Recall bias&#10;D)Healthy worker effect&#10;E)None of the above

Accepted Answer

Recall bias cannot occur for measures of a participant's state at the
point of entry into a cohort study, as they are unaware of their future disease history so
cannot be influenced by it; and this is also true for an interviewer at baseline.In
contrast all measures may be misclassified (A) at any point in a study and the healthy
worker effect (D) influences the number of health events the cohort experiences.

Question 5

What would the OR have been if 10% of cases who were unexposed were misclassified as exposed, but all other groups were measured accurately?  &#10;A)1.33&#10;B)0.99&#10;C)2.34&#10;D)1.55&#10;

Accepted Answer

After redistribution as above, OR = (54 x 945) ÷ (85 x 257) = 2.34

Question 6

What term best describes the misclassification in the previous question?&#10;A)Differential&#10;B)Non&#8208;differential&#10;C)Random&#10;D)None of the above

Accepted Answer

Differential misclassification.That is, the misclassification in exposure
'differed' between cases and controls.In case‐control studies, differential
misclassification of past exposures can be a particular problem where cases (because
they are aware that they are a case) may over‐estimate their 'true' exposure status due
to recall bias.Note that differential error can change estimates in either direction,
depending on the circumstances.

Question 7

In a case&#8208;control study of the association between residence near a hazardous waste site and leukaemia, selection bias is likely to be a problem because:&#10;A)People with leukaemia may be more likely to participate than controls recruited from the general population&#10;B)People with leukaemia living near a hazardous waste site may be more likely to participate than people living further away.&#10;C)People without leukaemia living near a hazardous waste site may be more likely to participate than people living further away&#10;D)All of the above

Accepted Answer

All of the above.In case‐control studies, cases are often more likely to
participate than controls, perhaps for reasons of wanting to understand (or contribute
to an understanding of) the disease.Similarly, in this study, both cases and controls may
be more likely to have an interest in participating in a study about something that
relates directly to their neighbourhood, than those who live in other neighbourhoods.
These factors associated with participation appear to be associated with both the
exposure and the outcome, and so may lead to spurious associations.

Question 8

In a hypothetical case&#8208;control study with perfect measurement, the association between occupational resin exposure and lung cancer was OR=1.76.The data for this&#10;Association is provided in the Table below:   What would the OR be if 10% of all those exposed had been misclassified as&#10;Unexposed, and 10% of all those unexposed had been misclassified as exposed?&#10;A)1.33&#10;B)1.51&#10;C)0.99&#10;D)1.55

Accepted Answer

OR=1.51.Re‐allocation of 10% of unexposed cases to exposed cases (and
vice versa), and of unexposed controls to exposed controls (and vice versa) is given
below: 11ebbba8_3419_8738_bcf2_6f68ac1cabe8_TB8436_00 OR = (50 x 876) ÷ (89 x 326) = 1.51

Question 9

A study was conducted to determine the side effects of 4 anti‐malarial drugs in a group of overseas volunteers in Cameroon.A self‐report questionnaire was given to each
Participant and participation was voluntary.Each participant was asked to indicate
Which drugs they had used and the severity of any side effects they had experienced
Over the past two years.What kind(s) of bias would most likely be present in this
Study? (Select all that apply)

A)Interviewer bias
B)Recall bias
C)Observer bias
D)Misclassification
E)Volunteer bias

Accepted Answer

The fact that people were asked to think back two years will almost
certainly lead to general mismeasurement (D) of side effect frequencies in all groups.
Recall bias (differential recall among the different drug groups) is possible if the groups
differ in ways that could affect their recall (e.g.their disease experience, their jobs, their
qualifications).As there were no outside observers (C) or interviewers (A) those
answers are irrelevant.There would be no volunteer bias (E) in comparisons among the
drug groups, but the rates of side‐effects in the study group could differ from the true
rates in all volunteers if, for example, those who had more side effects were more likely
to participate.

Question 10

In prospective cohort studies, why is loss to follow&#8208;up a source of selection bias?&#10;A)The study will no longer be representative of the general population&#10;B)It reduces the size of the study sample&#10;C)Those lost to follow&#8208;up may have different characteristics to those who remain in the study&#10;D)Those who are lost to follow&#8208;up may recall things differently from those who remain in the study&#10;E)There is more likely to be ascertainment bias when measuring outcomes in those who remain in the study

Accepted Answer

Those lost to follow‐up may have different characteristics to those who
remain in the study.If these characteristics are related to disease outcomes, then this
differential selection out of the study will lead to biased effect estimates.Most cohort
studies are not directly representative of the general population anyway (A); the sample
will be smaller but this does not in itself introduce selection bias (B); recall bias is
irrelevant to follow‐up in a cohort study (D); and errors in outcome ascertainment
within the study will be unrelated to any losses of participants from it (E).

Deck 7: All That Glitters Is Not Gold: The Problem of Error