Lecture 8 - Confounding and effect Flashcards
define confounding
the mixing of the effect of exposure with effect of another factor
define confounder
disease risk factor thats independent of exposure, associated with exposure but not a consequence of exposure and not in the causal pathway
define stratified analysis
test association (OR) btw exposure and outcome separately within each level of confounder
what are the design aspects of managing confounders
randomization, restriction, matching
define restriction
limit entrance to subjects who fall in one category of confounder
what are the problems of randomization
cant randomize in an observational study
what are the problems of restriction
reduce @ of participants, residual confounding, cant study effect of other counfounding category
what are the analysis aspects of managing confounders
multivariable regression analysis or stratification
what is multivariable regression analysis
method for studying relationship btw 1 dependent variable (criterion) and 2 independent variables (predictor variable)
what is the design of a multivariable regression analysis
10x as many cases as independent variables
what is the purpose of multivariable regression analysis
theory building, prediction, explanation
What are the 4 things a confounder can do?
- wipe out association 2. create an invalid association 3. reverse direction of association 4. change degree of association
define effect modification
3rd variable that changes relationship btw exposure and outcome
define effect modifier
exposure-disease relationship differs at diff levels of effect modifier
confounder effect on exposure-disease relationship ______ at diff levels of confounder
stays the same
what is the equation for stratified analysis
manzel haenzsel method
What does mantel haenszel method provide
a pooled odds ratio across the strata of fourfold table
what does mantel haenszel estimator measure
measure of the average effect of exposure across all strata
mantel haenszel accounts for what
confounding and effect modifier
where does the mantel haenszel measure lie
btw OR/RR strat 1 and OR/RR strat 2
if crude OR/RR falls outside range of stratum specific OR/RR, what can u interpret
crude does not equal average of data and shouldnt be used
a third variable can be ….
confounder, EM, both or neither
it is a confounder if
crude does not equal MH, and strat 1 equals strat 2
it is an effect modifier if
crude equals MH, and strat 1 does not equal strat 2
it is both an effect modifier and confounder if
crude does not equal MH, and strat 1 does not equal strat 2
it is neither an effect modifier and confounder if
everything is equal (crude, MH, strat 1, and strat 2)