Class 11-12: Confounding Flashcards
Counterfactual theory
Counterfactual outcomes based on individuals most similar to test subjects..but can’t take back the exposure so you can’t use same person to test effects of an exposure. So, you try to compare to a subject you can exchange w/. This subject.
Exchangeability
Comparability
Lack of exchangeability
Confounding
3 aspects of evaluating internal validity of study
Confounding or effect modification
Bias
Statistical significance
Watch CBS before claiming internal validity
Confounding variable
A 3rd variable that distorts an asso. Btw exposure and outcome- makes groups NOT exchangeable in terms of asso.
Direct Acyclic Graph
Exposure------>Outcome
^. ^
I. I
I I
ConfounderOther ways to define confounding
A mixing of effects; an asso. Distorted d/t them (exp. and outcome) being mixed.
Confounding is a confusion of effects. Exp. is distorted b.c the effect of a 3rd factor is mistaken for effect of exposure
3 requirements of con founders
- Asso. w/ Exposure
- Asso. w/ outcome
- Not directly in the causal-pathway linking exposure to outcome
How test for confounding [3 steps]:
1-calculate crude outcome measure (OR/RR)-unadjusted asso.
2-Calculate outcome measure of asso. (OR/RR)- for all layers of the 3rd variable
3- Compare [divide by difference of unadjusted & adjusted the unadjusted vs. adjusted measures of outcome: if >15% confounding is present
2 main impacts of confounders:
1-Magnitude of asso.
2-direction of asso.
What must be included in the adjusted OR
Must state what was accounted for in the adjustments
Purpose for controlling for confounders:
To get more precise estimate of the measure of asso.
Ways to control for confounding:
- study design stage[during study]
- Analysis of data stage [after research completed]
3 study design stage [during research]: ways to control for confounding
Randomization, restriction, and matching
2 study design stage [after research]: ways to control for confounding
Stratification (w/ weighting)
Multivariate statistical analysis [regression analyses]
Randomization
HOPEFULLY Allocates an equal # of subjects w/ the known confounders into each intervention grp.
Not for sure..only can be used for interventional studies
Ex: hoping–flipping a coin
Restriction
Don’t allow certain subjects to control for confounders….doesn’t effect internal validity but it can effect EXTERNAL validity (generalizability)
MAtching
Study subjects in matched pairs r/t confounding variable to equally distribute confounder among study group.
Difficult to accomplish/time consuming. Can over match…not control for other factors besides the confounder.
Stratification
Descriptive and statistical data analysis between exposure and outcome…ex: 2x2 table young vs. old
This can be impractical
Multivariate analysis
This is the most common: Statistical analysis of data by mathematically factoring efforts of the confounding variable…ex: crude vs. adjusted
Time consuming.
Effect modification
A 3rd variable that modifies the magnitude of effect of a true asso. By varying it w/in different levels of a 3rd variable-modifies the effect across the strata
What’s the different about effect modification and confounding?
An effect modifying variable should be described and reported at each level of the variable….rather than controlled for
Ex of effect modification in class: live birth weight and mortality:
Unadjusted OR:1.06 Adjusted OR:1.02 = No confounding present, BUT When looking at strata layers of birth weights...the mortality OR varied greatly
Steps for testing Effect modification:
1: Calculate crude measure of asso. Btw exposure and outcome
2: Calculate measure asso. Btw exposure and outcome for each strata
3: Compare each of the stratum-specific measures of asso. [OR] btw each other while referencing adjusted measure….>15% = effect modification present
