Lecture 25: Selection Bias Flashcards
What is chance?
chance is random sampling error
How can we reduce random sampling error/chance?
by increasing the sample size
What are the four potential explanations for a study’s findings?
- it could be a true association
- it could be due to chance
- it could be due to bias
- it could be due to confounding
Define bias
any systematic error in an epidemiological study that results in an incorrect estimate of the association between exposure and risk of the disease
What is systematic error and how does it differ from chance?
it is error that is not random (like chance is) and it has a pattern (unlike chance does)
Why does bias matter?
because it can result in an incorrect estimate of the association between exposure and risk of disease
Systematic error can mean that the findings could be (3)?
- over estimated
- under estimated
- not affected
What are the three potential sources of bias?
- selection bias
- information bias
- publication bias
When can we control for selection and information bias?
in the design and data collection stages
What can the study investigators do in the planning stages of the study?
- identify the potential sources of bias in the proposed study
- identify possible ways to minimise these potential sources of bias
When does selection bias occur?
it occurs when there is a systematic difference between the people who are included in the study and those who are not, or when the study and comparison groups are selected inappropriately or using different criteria
What three things do we need to think about when considering selection bias?
- how do we recruit people into epidemiological studies?
- does everyone agree to participate?
- does everyone remain in the study?
Why do we need to think about how we recruit people into the study? How should we recruit people?
- if we ask for volunteers, the people that volunteer might be different from those who don’t (eg. they might be healthier of have an interest in the condition)
- we should randomly select people from a defined population in order to be more representative
How does bias relate to “does everyone agree to participate”?
not everyone is going to agree to participate and if people who are agree are different from the people that do not agree then bias can occur and this can affect the findings of the study
How does bias relate to “does everyone remain in the study”?
because if the reason why people stop taking part is due to the exposure or outcomes, then bias can occur
Why would people stop taking part in the study (5)?
because the study goes for a long time, people could lose interest, they could move away, they could be too sick or could die
We want to reduce loss to follow up as much as possible. What are some ways to do this?
- having alternative contact details
- maintaining regular contact with patients
- doing multiple attempts to get a response
How can selection bias affect a cross sectional study?
because CSS only look at exposures and outcomes at one point in time, bias associated with loss to follow up is not a problem but who enters the study is still relevant
What three things do we need to ask when we are considering selection bias in a CSS?
- who entered the study?
- is the sample representative of the source population?
- what is the response rate?
If the sample was not representative of the source population, we need to consider whether the prevalence is like to be (3)
- over estimated
- under estimated
- not affected
One of the big limitations of case-control studies is that they can be susceptible to selection bias. How can this be?
Case control studies occur when you recruit people with the outcome as your cases and people without the outcome as the controls and trace backwards to see if either group was exposed to the outcome. Because selection bias can occur when study or comparison groups are selected inappropriately or using different criteria, it is very susceptible to selection bias
How can selection bias occur in case-control studies?
We know that participants’ selection is automatically related to their outcome because that is the basis on which they are selected. But if their selection based on outcome (either the way they are recruited of whether they actually participated) also related to the exposure status, then bias can occur
In CCS, selection of cases and control must be independent of the ________ _________
exposure status
What is the function of the control group in a case-control study?
to provide an estimate of the odds of exposure in the population from which the cases came
As well as considering the selection of cases in a CCS, we also need to consider the selection of ________
controls
What are three potential biases in the selection of controls in a CCS?
- if the controls are not representative of the population which gave rise to the cases (in place or time)
- if the inclusion/exclusion criteria differs between cases and controls
- if there is non-participation of eligible controls
What can we do to eliminate “if the controls are not representative of the population which gave rise to the cases (in place or time)” as a potential source of bias in CCS?
ensure the controls are from the same defined population as the cases over the same time period
What can we do to eliminate “if the inclusion/exclusion criteria differs between cases and controls)” as a potential source of bias in CCS?
have the same inclusion and exclusion criteria for both cases and controls
What can we do to eliminate “if there is non-participation of eligible controls” as a potential source of bias in CCS?
ensure there is high participation
How do we calculate the odds ratio of a CCS?
the number of cases in the exposed group divided by the number of cases in the comparison group all divided by the number of controls (no cases) in the exposed group divided by the number of controls (no cases) in the comparison group (no exposure)
(a/c)/(b/d)
What are the three steps involved in the general approach to the effects of bias?
- who does it affect?
- how does it affect them?
- what effect does this have on the measure of association - is it under or overestimated?
For a harmful factor, is the relative risk above or below the null value?
above
For a protective factor, is the relative risk above or below the null value?
below
What can we say if the MOA is underestimated for a harmful factor?
that the MOA is biased numerically downwards, or it’s biased towards the null.
What can we say if the MOA is underestimated for a protective factor?
it is biased numerically upwards, or it’s biased towards the null
One of the big limitations of cohort studies is that they can be susceptible to selection bias. How can this be?
Cohort studies occur when you recruit people that have had an exposure and then you follow up over time to see if they develop the outcome or not. This means that there is a huge potential for loss to follow up in a cohort study and therefore they are susceptible to selection bias
How can loss to follow up cause selection bias in a cohort study?
if the loss to follow up is related to both the exposure and the outcome, this can lead to bias
How does the healthy worker affect relate to bias in a cohort study?
In a cohort study, one group is selected and they are classified on whether they have the outcome or not. However, sometimes the comparison group can be selected separately from the exposed group. If they are selected separately, this can lead to bias. HWE is an example of this:
if the workers are the exposed group then people who do not work are the comparison group and they may have different characteristics compared to the working population because the non-working population includes people that are too unwell to work
One of the big limitations of RCTs is that they can be susceptible to selection bias. How can this be?
In an RCT, participants are randomly assigned to an intervention or control group and then they are followed up over time to see if they developed the outcome. Therefore there is the potential for loss to follow up which is a factor in selection bias
