Systematic (non-random) error that results in an incorrect estimate of the association between exposure and risk of disease.
Can occur in all stages of a study
Not affected by study sample size
Difficult to adjust for afterwards, but can be reduced by adequate study design.
•Can never be totally avoided, but we must be aware of it and interpret our results accordingly
2. 22
Bias in epidemiological studies
•Systematic (non-random) error that results in an incorrect
estimate of the association between exposure and risk of
disease.
• Can occur in all stages of a study
• Not affected by study sample size
•Difficult to adjust for afterwards, but can be reduced by
adequate study design.
•Can never be totally avoided, but we must be aware of it and
interpret our results accordingly
•To what extent does our outcome measure correspond to the
“true value”?
3. 24
Consequences of bias
We make the wrong conclusion about the
relation between exposure and outcome in our
study
•Conclude there is no association while in
reality there is one
•Conclude there is an association while in
reality there is none
•Overestimate the strength of the association
•Underestimate the strength of the
association
4. 25
Bias in all stages of research
1. Research topic
2. Choice of population
3. Participation in a study
4. Inequality in study populations – cases-
controls / exposed and non-exposed
5. Assessment of exposure factor/disease
6. Measurement of follow-up
7. Analysis and interpretation of the data
8. Publication
9. Interpretation, judgment and action by
readers and listeners
5. 26
Two main classifications of bias
Selection bias : occurs when comparisons are
made between groups of patients that are not
comparable in determinants for the outcome
other than the one(s) under study.
Unequal Groups
Measurement / information bias: occurs when the
validity of measurement of Exposure and/or
Disease is dissimilar among groups of patients
not comparing the same exposure /
outcome parameters
6. 27
Selection bias
Systematic error that results from:
- the way subjects are selected into a study
Consequence: there is a difference between subjects
who are selected or participant, and those who are not.
Problem: we may make an incorrect conclusion about
the association between an exposure and an outcome.
7. Selection Bias
1. Are the characteristics of the study
population selected for the study different than
those who were not selected?
2. If yes, in what way / in which factor do they
differ?
3. Is this factor likely to affect the outcome
being studied?
8. 29
Measurement/ Information
bias
xposed
Systematic error that results from:
•the way information is obtained on one or more
variables (exposure, disease) that are
measured in the study.
•Differences in validity
of exposure data between cases and controls
of outcome data between exposed and une
Can occur in all types of studies
9. Information bias
1. Is there any possibility that the collection of
information has been influenced by :
1. Observer ?
2. Respondent?
3. Instrument?
2. Is there any possibility that the assessment in the
different groups has not been done equally similar,
appropriate, accurate and complete?
10. Why are we prone to bias?
Possible pitfalls in observational research
• Selectivity - selective nature of perception including of
selectivity of choice of topics
• Expectations -we only see what we expect to see
• Expert seeing - We only can see if we have the skills to see it
(e.g.microscope)
• The observer effect - Observing is already an intervention in
itself – the act of observation can sometimes affect what we
observe
11. We only see what we expect to see
Our ability to learn and to see is profoundly affected by:
1. Information to which we have previously been exposed
2. References and pathways we have built in our brain system
3. Emotional links_ emotionally charged events are
remembered better
4. Pleasant emotions are usually
remembered better than
unpleasant ones
12. Selectivity
Human mind has to filter information from outside
Perception is selective – we take what we feel is important
Risk of neglecting /overlooking important factors
The moment we label a person or a situation, we put on blinders to
all contradictory evidence
13. Direction of bias
The precise magnitude of bias can never really
be quantified, however, the direction of bias
can often be determined.
•The target parameter can be overestimated
(association measure (OR/RR) further away
from 1)
•The target parameter can be
underestimated (association measure
towards 1).
14. Examples of selection bias (1)
Healthy Worker Effect (HWE): The overall mortality experience
of an employed population is typically better than that of the
general population (in Western countries at least). Use of blood
donors as controls is a kind of HWE. Blood donors are self-selected
on the basis of better life styles.
Non-respondent bias: Non-respondents to a survey often differ
from respondents. This difference may be related to the subject of
interest
Volunteer bias:
Those who do participate in a study might be different than those
who do not participate. This difference is related to the subject of
interest
Attrition bias: Number of individuals lost to follow up may be
different between exposed/unexposed / intervention/ control
15. Non response bias in case control
study
• Case control study to study the effect of
passive smoking on the risk of heart attack
• Controls who were current smokers were less
likely to
participated in the study
• Smoking exposure in control group is likely
to be an underestimate of the true
proportion of smokers in the population
• Odds of smoking in control group is lower than
‘true’ odds
• Overestimate of the strength of the association
16. Selection bias in case-control studies
The major problem in case-control studies is choice of
controls.
How representative are hospitalised trauma patients of the
population which gave rise to the cases? Is OR under- or
overestimated?
Cases
liver cirrhosis
Controls A
trauma ward
80 40
20 60
Heavy alcohol use
OR=6
Light/no alcoholuse
17. Avoid selection bias – controls should arise
from same population as cases
Cases
liver cirrhosis
Controls A
trauma ward
Controls B
non-trauma
Heavy alcohol use
Light/no alcohol use
80 40 10
20 60 90
OR=6 OR=36
Hospital based case control studies more prone to
selection bias than population based case-control
studies
18. 42
Attrition bias in cohort studies
Systematic error that results from:
- the way we lose subjects during a study
(i.e., this is called loss to follow-up or
LTF)
Consequence: there is a difference between subjects
who are selected or participant, and those who are not
at the end of the study for analysis
Problem: we may make an incorrect conclusion about
the association between an exposure and an outcome
may lead to a over/ underestimation of effect
19. 43
Attrition bias -during follow up-
Problem:
Crossover or loss-to-follow up cause bias when
related to exposure
The exposure & non-exposure groups that were
comparable at the start of the study are no longer
comparable
20. Measurement/ information bias
Can be caused by:
• The observer (interviewer bias)
• The measurement instrument (diagnostic bias)
•The way the participant remembers and reports
events in that happened the past (recall bias)
45
21. Observer bias -Interviewer bias
Investigator gathers or interprets information in a different way for cases
and controls or for exposed and uneposed
Cases of
listeriosis
Controls
Eats soft cheese a b
Does not eat
soft cheese
c d
Investigator may probe listeriosis cases about consumption of soft cheese
• Overestimation of “a” overestimation of OR
22. 47
Diagnostic bias
When the exposure under study leads to intensified
diagnostic procedures and increased chance of
identifying the disease
Exposure and non-exposure groups not comparable
anymore
Example: effect of benign breast disease on breast
cancer. Women with benign disease undergo more
extensive diagnostic procedure and are more likely to
have cancer detected.
Bias : overestimation of effect
23. Can we trust what we
think we remember?
Recall bias
•The way the participant
remembers events in that
happened the past
24. Recall bias: what and when do we remember?
Children with
malformation
Controls
Took tobacco,
alcohol, drugs
a b
Did not take c d
Occurs when: Cases remember exposure differently than controls
Mothers of
Mothers of children with malformations will remember past exposures
better than mothers with healthy children. But maybe, they are ashamed
and will underreport their tobacco, alcohol and drug use
Overestimation or underestimation of “a”
overestimation/underestimation of OR
25. Recall bias can be minimized by:
• Interviewing cases and controls in a standard
way
• Making use of independent sources for
measuring exposure (or at least validate a
sample)
Minimizing recall bias
26. 54
Measurement/ information bias
Can lead to misclassification bias
Study subjects are classified in the wrong category
(disease /not diseased; exposed / non-exposed)
Non-differential misclassification
• Misclassification is equally divided among comparison groups
• Generally dilutes the exposure effect (toward to null effect)
Differential misclassification
• It is worse when the proportions of subjects misclassified
differ between the study groups
• Such a differential between groups may mask an association
or cause one when there is none.
• Effect is unpredictable
27. Exercise: alcohol consumption and colon
rectal cancer
You have conducted a case control study to examine
the association between high alcohol consumption and
colon rectal cancer.
In pairs:
List and discuss types of biases that may occur when
conducting a case control study to study this particular
question.
28. Exercise: Potential bias
Prone to selection bias: Selection of controls can introduce
bias
explain
• If controls do not represent the population from which the cases
come from
Prone to information bias: Recall bias
explain
• non-differential recall bias – difficult to remember for both cases
and controls how much alcohol one has consumed in past
• differential recall bias – cases may remember or report their
alcohol consumption more or less accurate than controls
29. Exercise: diabetes and stroke
You have conducted a cohort study looking at the association
between adult onset diabetes and risk of death from stroke,
during 10 years of follow-up.
In pairs:
List and discuss types of biases that may occur when
conducting a cohort study to study this particular question.
Titel
30. Exercise: potential biases
Prone to: Information bias: outcome assessment
Explain
• If the person determining stroke/no stroke had prior
knowledge of diabetic history they may be more or less
likely to classify the outcome as stroke
Prone to: Selection bias: attrition bias
Explain
• If in a prospective cohort study loss to follow up was
different between those with and without diabetes the
incidence rates of death from stroke is difficult to interpret
Titel
31. Study type and Bias
A study that suffers from bias lacks internal validity.
In case-control studies, recall bias (knowledge of disease
status influences the determination of exposure status) and
selection bias (knowledge of exposure status influences the
identification of diseased and non-diseased study subjects)
are most important.
In cohort studies, bias due to loss to follow-up (attrition)
would be the greatest danger.
The potential for misclassification is present in all types of
epidemiologic studies and in all stages of the study
32. Minimising selection bias
In case control studies:
• Clear definition of study population
• Explicit case and control definitions
• Cases and controls from same population
In cohort studies:
• Selection of exposed and non-exposed without
knowing disease status (retrospective cohort)
• Tracking procedures for loss to follow-up
33. Minimising information bias
Interviewer/observer bias
• Blinding of observer/interview for exposure/disease
• Development of a protocol for collection, measurement and
interpretation of information
• Standardize questionnaires, calibrated diagnostic tools
• Using trained and experienced interviewer
Attrition bias
• Minimise lost to follow up
Recall bias
• Blind participants to study hypothesis
• Objectively collect exposure data (e.g. work / medical
records)
Titel
34. Bias in randomised controlled trials
Gold-standard: randomised,
placebo-controlled, double-
blinded study
Least biased when:
• Exposure randomly
allocated to subjects -
minimises selection bias
• Masking of exposure
status in subjects and
study staff - minimises
information bias
Selectionbias
Performancebias
Attritionbias
Detectionbias
35. Reference population
Informed consent sought
Assignment by
randomization
Participants Non-participants
Follow for
outcome
Outcome
known
Outcome
unknown
Intervention
YES
Intervention
NO
Follow for outcome
Outcome
known
Outcome
unknown
Study population
36. Remember the consequences of bias
We make the wrong conclusion about the relation
between exposure and outcome in our study
• Conclude there is no association while in reality there
is one
• Conclude there is an association while in reality there
is none
• Overestimate the strength of the association
• Underestimate the strength of the association
37. How to deal with bias – think ahead
1. Design stage - minimize or avoid bias.
• Avoid selection bias by including/excluding eligible
subjects, by
• Choice of source population
• Choice of the comparison group
2. Analysis stage - determine presence or direction of
possible bias and also account for confounding in
analysis.
3. Publication stage - Potential biases typically described
in "Discussion" section. Provide judgment and possible
consequences of bias on results