Flashcards in Confounding and Effect Modification Deck (14):
What thre aspects much be studied before declaring a statistical association?
Counfounding or effect modification, bias, or statistical significance
Purpose of Controlling for confounders
To get a more precise estimaste of the true association btw the exposure
Ways to control for confounding
Randimization, restriction, and matching
Analysis of data stage
stratification and multivariate or matched statistical analysis
Randomization in confounding
Allocates an equal number of subjects with the known and unkown confounders into each intervention group
Strengh: likely to be sucessful with lots of ppl
weakness: sample size may mnot be large enough...
What kind of studies is randomization practical for?
Restriction in confounding
Study participation is restricted to only subjects who do not fall within prespecified categories of confoudner..
strenth- straight fwd, does not negatively impact internal validity,
weakness- narrows criteara, harder to enroll ppl, negatively impacts generalizability
Matching in Confounding
Study subjects seleted in matched pairs related to the confounding variable to eqaully distribute confouder among each of the study groups.
stregth= intuitive... some thing greater analytic efficency
weakness- difficult to accomplish, time consuming and expensive. ... only controls for matched cofounders.
stratification in confounding
statistical anaylsis of the data by evaluating the association within various layers
weaknes- can't work for multiple co founders.
Multivariate analysis for confounding
Statistical anaylsis of the data by mathematically factoring out the effects of the confounding variables
strengths, can control for mualtiple variables
Weakness. process can cause some individuals to not clearly understand the data
linear and logistic regression
cox proportional hazards modeling
A 3rd vaiable that when present modifies the magnitude of effect of an assocaition by varying it within different levels... unlike confounding an effect modifying variable shoudl be described and reported at each level of the variable rather than controlled for.
How to check for effect modification?
Compare the strata to one another.
Two main ways control fro confounding
Study design stage and analysis of data stage