Epidemiology: 9) Confounding
Terms
undefined, object
copy deck
- 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+
- 3 advantages of matching
-
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) - 4 disadvantages of matching
-
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 and Disadv of restriction
-
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