EPI 201 Summary 2
[Confounding]
A source of bias that distorts the true measure of association and is something that we aim to eliminate
from our study.
[Impact of confounding on validity]
In the presence of confounding, the point estimate is biased (Confounding introduce bias into point
estimates)
=> The results of statistical inferential procedures such as p values and confidence intervals are biased.
=> p-value: too small or too large,
CI: centered (bias in the point estimate) or width (bias in the estimate of the variance of the point
estimate)
The magnitude of the bias due to confounding in an epidemiologic study depends on:
・
The association between the confounder and the outcome
・
The ratio of prevalence of the confounder in the exposed group divided by the prevalence of the
confounder in the unexposed group
・
The prevalence of the confounder
Downward bias: the association appears more protective
∴ after adjustment, the association will be higher on the absolute value, closer to the null and it may even
indicate a harmful association

[Example of confounding]
Crude Odds Ratio
Smoking and MI
Prevalence Ratio
Crude OR is biased toward upward direction.
DM
No DM
Total
MI
122
1490
1612
No MI
31
1581
1612
Total
153
3071
3224
Smoking
No
Smoking
Total
MI
500
990
1490
No MI
155
1426
1581
Total
655
2416
3071
DM
No DM
Total
Smoker
15
155
170
No
Smoker
16
1426
1442
Total
31
1581
1612

[Matching on Case-control study]
・
Matching on case-control study leads to confounding (selection bias)
・
Matching on factors that are likely to be confounders can improve the efficiency of the study.
・
Once the data are stratified on the confounding factor, there will be balance between the number of
cases and controls within each stratum.
[Alteration for the association between the confounder and the exposure of interest in the study]
・
Matching in a cohort study
・
Restriction: (restriction to women => no confounder about gender)
・
Selection criteria
・
Randomization
[Matching on confounders in a cohort study]
1: Prevalence of the matching factors (confounders) is the same in the exposed and unexposed.
2: eliminate the confounfer-exposure association
× matching in a case control study (lead to selection bias)

In a case-control study, the controls are used to estimate the exposure distribution on the study base.
Matching may cause the exposure distribution among the controls to no longer represent the exposure
distribution in the study base. Thus one must also adjust for the matching factors in the analysis stage.
○ un-matching, but selection
[Selection criteria]
[95% CI]
The data are consistent with odds ratios between A and B, with 95% confidence.

[Effect measure modification]
・
Effect measure modifier
−
In the presence of effect measure modification, the magnitude of the association between exposure and
disease varies according to a third factor, called an effect modifier.

#### You've reached the end of your free preview.

Want to read all 15 pages?

- Fall '18