1/22
Looks like no tags are added yet.
What is confounding
mixinf of effects between exposure and ourcome.
where theres a 3rd variab;e
distorts relationship between exposure and outcome.
Counterfactual ideal (Time machine)
ideal comparison groups in a cohort study are the same individuals in the exposed groups had they not been exposed.
ie you want groups thar are as similar as possible.
How is confounding related to the counterfactual ideal?
Confounding = failiurs of counterfactual
Unexposed group no longer represents the counterfactual experience of the exposed group.
Risk factors for outcome = no longer equally distributed.
Describe the diabeties vs dementia example
Here, RR = 3.5
Particp. with diabeties were older than those withouit
age = risk factor for dementia
confounding occured bc age was associated with both diabeties and dementia.
What defines a confounder.
Must cause or predict outcome in both exposed and unexposed (for example, age in dementia example.
Must be associated with the exposure in the population that produced cases
Must not be on the causal pathway between exposure and outcome.
Steps to seeing if something is a confounder
Association is oresent → Confounding is possible
Evaluate association between confounder and exposure.
Association present → Confounding = present.
Brigning back Is age associated with dementia back to check if this is a confounder
in those without diabeties: RR = 2.0
In those with diabeties: RR = 2.0
Seeing if age is associated with diabeties: we see a RR of 9.0
Confounding formula:
RR crude - RR adjusted/ RR adjusted
confounding does one of the 5 things:
Distorts association measure by a small amount
Distorts the association measure by a large amount
Reverses the associaton direction
Removes an association when one is there
Creates an association when none is there
Describe the myopia light exposure example:
Full room of lights: OR = 5.5
Night lights: OR = 3.4
why was this confounded
failed to control for mypoic status of parents
Found that myopia was higher in children of 2 myopic parents.
Myopic parents = more likely to keep lights on.
3rd variable → Exposure → Outcome
Confounding by indication + Example
Reflects that treatements are given to particular disease
Mixing between treatmen, disease and outcome under study.
Comparison to those w/o disease is confounded
Example Antidepressents + Memory defecits
Depresseion → indication for antidepressants
Depression → associated with memory defecits.
Solution: Compare different antidepressets in people with depression.
Confounding by severity + Example
disease + Severity are both indications for the exposure
Makes treatment look less effective
Ex: Combination therapy + Protease inhibitor for HIV during pregnancy
found this combination was assoc. with severe HIV
more compromized system, higher viral load.
ART more likely to be given to women with severe disease.
Solution: Experimental study
Controlling for confounding (Design phase)
Randomization → Balances know + unknown confounders.
Design phase: restriction
• Entry into the study is confined to individuals who fall into
a specific category of the confounder.
Reduces or eliminates variability induced by the
confounder
Difficult to find participants, limits generalizability.
Design Phase: Matching
making distribution of matching factors identical across study groups.
individual matching: moth cases and controls must be 55 y.o. men.
Frequency matching: 20% of cases + controls are aged 40-59
Matching downfall
• Can’t assess association of matching factor with outcome
• Difficult and expensive to find matches
• “overmatching” (associated with exposure but not
disease)
• Good for neighbourhood control studie
Analysis phase: Standardization
indirect or direct standardization
Analysis phase: stratification
evaluate and control for confounding
Assess association of exposure and outcome within groups defined by level of confounder (age sex etc)
explain the stratification example from the slides
DDE and breast cancer
Overall OR = 1.9
aged below 50: OR = 1
Aged 50 or older OR = 1
1/1 = 1 meaning that within each group, DDE exposure is not associated with increased odds or breast cancer.
Multivariable regression
mathematical model to describe association between exposure, disease and confounders.
model association of exposure with disease, adjusted for confounders.
Propensity Scores
summarize likelyhood of participant being exposed vs unexposed based on observed confounders.
Can be used to match participants, stratify participants, or included confounders in regression to adjust for confounders.
regression to identify predictors of exposure → Propensity score representing probabiloity of exposure → Matching stratification, or regression.
Residual confounding.
Confounding that remains after confounding variables were controlled
Data not collected
Differences in risk within category or confounder
missclassified.
3 things to remember from lecture
Confounders are variables that distors the effect of an exposure or outcome.
Confounders are associated with the outcome, are assocuated with the expopsure and are not on the causal pathway between exposure and outcome.
effect modifiers are varuables tht change the relationship and exposure outcome.