Lecture 27: Confounding | Flashcards
What is confounding?
A mixing or muddling of effects when the relationship we are interested in is confused by the effect of something else - the confounder
What are the three properties of a potential confounder?
- independently associated with the outcome
- independently associated with the exposure
- not on the causal pathways
What is meant by the confounder being independently associated with the outcome? Give an example
A potential confounder is a risk or protective factor for the outcome regardless of the exposure status.
For example in a study looking at the link between childhood antibiotics and obesity at age 5, SES is a potential confounder. By it being independently associated with the outcome, it means that SES is a risk factor for obesity (low SES means more likely to be obese) regardless of whether their child had antibiotics.
What is meant by the confounder being independently associated with the exposure? Give an example
There needs to be different proportions of people with potential confounders across the exposure groups otherwise it won’t lead to a different outcome.
For example in a study looking at the link between childhood antibiotics and obesity at age 5, SES is a potential confounder. By it being independently associated with the exposure, it means that SES is a risk factor for childhood antibiotics (low SES means more likely to have antibiotics) regardless of whether their child is obese.
What is meant by the confounder not being on the causal pathway? Give an example
The confounder is not the mechanism which the exposure affects the risk of the outcome.
For example in a study looking at the link between childhood antibiotics and obesity at age 5, SES is a potential confounder. By it being on the causal pathway, it means that whether a child has had antibiotics, it does not determine their SES and whether they are obese or not
What four things can confounding do?
- can cause an overestimation of a true association
- can cause an underestimation of a true association
- can cause a change of direction of a true association
- can give an appearance of an association when there really isn’t one
What three things can we do to control for confounding?
- randomisation
- restriction
- matching
Describe the process of randomisation
In an RCT, of the sample, people are randomly allocated into treatment group and control group
For a harmful exposure, if the study RR is greater than the true RR, what has happened?
confounding has resulted in an overestimation of the true harmful effect of the exposure (the association appears stronger than it really is, the RR is further away from the null)
For a harmful exposure, if the study RR is less than the true RR, what has happened?
Confounding has resulted in an underestimation of the true harmful effect of the exposure (the associated appears weaker than it really is, the RR is closer to the null)
For a beneficial exposure, if the study RR is less than the true RR, what has happened?
confounding has resulted in an over-estimation of the true protective effect of the exposure (the association appears stronger than it really is, the RR is further from the null)
For a beneficial exposure, if the study RR is greater than the true RR, what has happened?
Confounding has resulted in an underestimation of the true protective effect of the exposure (the association appears weaker than it really is, the RR is closer to the null)
How can you identify potential confounders? Give an example
you have to plan ahead and collet information on all potential confounders and then you can look for an imbalance in potential confounders between groups eg. is there a big difference in age or SES between the two groups
When can you control for confounding?
in the study design
in the analysis phase
What are three ways to control for confounding in the study design?
- randomisation
- restriction
- matching
What is the purpose of trying to control for potential confounders in the study design?
to try and make the groups being compared alike with regard to potential confounders
What is randomisation? What is the advantage of of it and what are some potential issues?
- when people are randomly allocated into treatment or control groups
- the advantage to this is that is controls for potential confounders because if people are randomly allocated, there should be an even distributed of both known and unknown confounders between the two groups
- it works best with large sample size, it requires equipoise and an intention-to-treat analysis
What is restriction? Give an example
when you restrict the sample to only one stratum of potential confounder eg. when looking at a study involving breast cancer, you may choose to exclude men
What are some “buts” of restriction?
- it can reduce generalisability
- it reduces the number of potential participants
- there is the potential for residual confounding with imprecisely measured (or broadly defined) confounders
- you can usually only do it with one confounder
What is matching and when is it done?
When you choose people for the comparison (control) group who have the same values of the potential confounding factor(s) as people in the exposed group (cohort studies) or case group (case-control studies)
What are the two ways to do matching?
individual or frequency matching
What is individual matching?
each case matched with one or more controls having the same confounding variable characteristics
What is frequency matching?
when you match at an aggregated level
What are some advantages and disadvantages of matching?
Positives:
- useful for difficult to measure/complex potential confounders
- can improve efficiency of case-control studies with small numbers
Negatives:
- individual matching can be difficult and limit the number of potential participants
What are some disadvantages to controlling for confounding in the study design?
you can’t assess the association between the potential confounder and the outcome and you can’t assess whether the confounder is actually a confounder
