Lecture 4: Confounding & Effect Modification Flashcards
the best population to compare with, is themselves minus the event
counterfactual theory
counterfactual theory requires the assumption of?
exchangeability
lack of exchangeability in a population results in ?
confounding or
we are unable to compare
when we are assessing the associations of an exposure and outcome between two groups, what question are we asking ourselves regarding a true association?
are there alternative explanations to the association?
researchers evaluate 3 aspects of their study, before declaring a true association
- check for confounding variables
- check for bias
- check for statistical significance
define ‘confounding variables’
a 3rd variable that distorts an association between the exposure and outcome
an alternative explanation
what associations do confounding variables effect?
OR
HR
RR
simple definition of confounding
another explanation that appears to be the cause of the outcome, but it is not
in order for a variable to be considered a confounder, it must have 3 things
- independent association with exposure
- independent association with outcome
- not directly in the causal-pathway
what does it mean to ‘not directly be in the causal pathway’ ?
the confounding variable is not a stepping stone between exposure and outcome
it is independent of both
a measure of association that ignores all other factors
CRUDE measure of association
or
unadjusted
DAG
direct acyclic graph
if confounding is present, how do you go about knowing if it is present?
you must test and calculate for confounding
step 1 in testing for confounding
calculate crude RR
also called unadjusted RR
step 2 in testing for confounding
calculate OR for each individual strata of the 3rd variable
step 3 in testing for confounding
compare the crude vs. adjusted measures
when comparing calculations, what tells you if the 3rd variable is a confounder?
compare crude vs. adjusted
if >15% different === confounder
3 steps in testing for confounding
- find crude RR
- fund adjusted RR for each strata
- compare crude vs. adjusted
what are the 2 main impacts of confounding variables?
magnitude of association
direction of association
what is the purpose of controlling for confounders?
to get a more accurate measure of association
at what 2 points in a research study can you control for confounding?
study design stage
analysis of data stage
what are 3 ways to control for confounding in the study design stage?
randomization
restriction of population
matching of participants in each group
what are 2 ways to control for confounding in the analysis of data stage?
this is after data has been collected
- stratification
- multivariate statistical analysis
2 important weaknesses of using randomization to control for confounders
must have a large sample size
practical only for interventional studies
weaknesses of using restriction for confounder controlling.
narrow restriction criteria limits who can be a subject
can negatively impact generalizeability
strengths of randomization
will large N, it is successful
stratified version precisely assures equal-ness
strengths of restriction
convenient, cheap
does not negatively impact internal validity
weaknesses associated with matching
difficult to do, time, expensive
only controls for confounders that are matched for
strengths of matching
greater analytic efficiency
strengths of stratification
straight forward, enhances understanding of data
weakness of stratification
impractical for controlling multiple confounders, especially ones with multiple strata
statistical analysis of data, evaluating associations between exposure and outcome within the various strata of a confounding variable
stratification
data stage confounder control mechanism
statistical analysis of data by mathematically factoring out the effects of the confounding variables
multivariate analyses
regression stats
multivariate weaknesses
requires researchers who really understand math
time consuming for statistician
multivariate strengths
control for multiple confounders at once
ORs can be obtained and interpreted
if you are interpreting an adjusted ratio, you must also include…..?
in the interpretation you must include what has been adjusted for
a 3rd variable capable of changing the magnitude of effect, of a true association, by changing it within different strata of that variable
effect modification
compare effect modification and confounding: relate to association between talc use and cancer in women:
crude = 1.3 adjusted for race = 1.3
race is not a confounder (<15%)
but if we check each race strata…..
we find that not every race has the same talc use
so race is an effect modification
effect modification is vital detail that we want to ____ within a study.
describe
3 steps in testing for effect modification
- calculate crude
- calc strata ratios
- compare each of the strata
step 1 in testing for effect modification
calculate crude RR between exposure and outcome
step 2 in testing for effect modification
calculate RR for each specific strata within a variable
ex.
race – white, black, latino, asian
step 3 in testing for effect modification
compare the highest and lowest RR of the strata
>15% difference = effect modification is present
