# cueFlash

## Glossary of Epidemiology: 9) Confounding

### Deck Info

#### Other Decks By This User

Basic confounding Def'n
Apparent effect of E is actually due to other differences between E+ and E- groups.
Derives from roots meaning "to mix together"
3 major requirements to be a confounder
1) It is a risk factor for D
2) It is correlated with E
3) Can NOT be on the causal pathway between E and D (nor effect of E)
Confounding Diagram
(9-3)
Magnitude of Confounding (eqn)
w = (RRc*Q1+(1-Q1))/(RRc*Qo+(1-Qo))
RRc = risk ratio (or rate) for confounder
Q1 = Confounder prev in E+ group
Qo = Confounder prev in E- group
Apprent risk ratio from confounding
RR(true E) * w(risk confurred by confounding)
3 ways to manage confounding during design
1) Randomize (RCT)
2) Matching
3) Restriction
3 ways to manage confounding in analysis
1) Regression modeling
2) Rate standardization
3) Stratified analysis
Matching in cohort studies (def'n)
Equalize the distribution of C in the E+ and E- groups
Eg. of classical matched studies
1) Twin studies (genetics)
2) Related controls (eg. siblings)
3) Mathced on other chars
Propensity scores
Generalized matching on single or matching factors. Calculated probability that a particular subject is E+
1) Increases comparability of E+/E- groups.
2) If similar pairs made: increases statistical efficiency.
3) Reduces data collection costs (esp when E+ group is small)
1) Equalization of characteristics is often assumed but difficult to quantify.
2) If does not produce similar pairs it can decrease statistical efficiency (sample size).
3) Can require specialized stats analysis.
Restriction
Restrict cohort to a single value of the confounder. (Eg. in hip fracture study men and women separately)
Adv: good if etiology of disease is different according to potential confounder values.
Disadv: loss of power and selectio bias
Factors in the causal pathway between E and D
It has been shown that High levels of HDL protect against MI. A study suggests that ETOH can protect against MI. However, if you control for HDL, ETOH has no beneficial effect. You CANNOT control for HDL because ETOH modifies HDL
Direct effects of the exposure
No need to treat yellow-fingers as a confounder.
In cohort study it can reduce statistical power but not distort etimates.
If matching in case-control it can introduce bias.
Confounders from Design problems
1) Quality of information
2) Insensitive diagnostics
3) Length of follow-up
Misclassifying confounders
Does NOT bias towards the NULL; has same effect as not controlling for confounding