WEEK 5: confounding Flashcards
What is confounding?
biased exposure –> outcome relation, when an effect of an extaneous factor is mistaken for/mixed with actual exposure effect. Threat for internal validity
(e.g. drinking -> diabetes, but people who drink tend do smoke a lot as well)
What is effect measure modification?
Measure modification: association which differs across levels of a third factor (e.g. sex is an effect modifier) -> treat for external validity
= fact of nature, life, society or culture
How to avoid confounding in observational studies?
- Prevention in the design
● Restriction
● Matching - During data-analysis
● Stratification
● Multivariable logistic regression analysis (statistical technique)
How does confounding restriction in the case-control design work? Limitations?
Do your study in participants with or without the bias only, not mixed
E.g. in a study of drinking -> disease with confounder: smoking, choose to do your study in non-smokers or smokers only
Limitation: difficulty in attaining required sample size
How does confounding matching in the case- control design work? Limitations?
Match cases and controls on the confounding habit/confounder
>
Select study subjects in such a way that major known confounders are equally distributed across cases and controls
Limitation: may be very difficult to find individuals that match multiple characteristics
How can you select study subjects in such a way that major known confounders are equally distributed across cases and controls?
- On individual basis, with one or more controls matched to each case (e.g. same sex, birth date) one-to-one pair matching
- Frequency matching: distribution of confounders is equal in the group of cases and controls
- Select a population of controls such that the overall characteristics of the group match the overall characteristics of the cases.
● e.g. if 15% of cases are under age 20, 15% of the controls are also
What variables should you not match? Why should you be critical?
do not match on variables that could be in the causal pathway of the exposure-disease association
If you match on a variable you cannot study its impact anymore
What is stratification? Limitation?
When both smokers and non-smokers are in your study population and no matching was performed, you adjust for the confounder by dividing into two groups: smokers vs non-smokers.
Odds ratio from whole group is different from small groups
Limitation:
With multiple stratification variables, there will be many empty strata (stratification tables?)
What is the “intention-to-treat” principle?
Effect of a treatment is assessed on the basis of the planned treatment rather than the actual treatment given → All randomized subjects are included in the data-analysis
Why would you use the “intention-to-treat” principle?
Eliminating data from non-complint/lost subjects from analysis may yield biased results
What is a drawback of “intention-to-treat”?
Including non-compliant subjects may weaken the effect of treatment
What is the ‘per protocol’ method?
Compares those who actually received treatment with those who did not
Why use the ‘per protocol’ method?
Then you look directly at the effect of treatment
Drawback of the per protocol treatment?
Susceptible to bias if those who adhere to treatment differ from those who do not
When is double blinding not possible?
e.g. weight loss trials, dietary interventions