Confounding

Question 1

What is confounding

Answer 1

• Different explanations in the data cannot be differentiated
  o 3rd factor has a relationship with exposure and outcome – ie people who drink are likely to smoke, and smoking actually causes the lunger cancer not alcohol

Question 2

How is confounding fixed

Answer 2

• Can be adjusted for in the analysis (unlike bias)

Question 3

Characteristics of confounding

Answer 3

o Must be associated with both the exposure and the outcome
o But must not be a path variable
o It can increase or decrease any observed effect

Question 4

Possibilies for the relationship between exposure and outcome

Answer 4

• No association
• E is directly associated with O
• E is inversely associated with O
• Third variable partly explains the association

Question 5

How do you identify confounders

Answer 5

• Literature search of similar studies
• Biological plausibility
• Univariate analysis to test association with exposure/outcome
  o Usually table 1, sociodemographic comparison table

Question 6

How do you control for confounders in design phase

Answer 6

o Restriction
 Restricted inclusion criteria – homogenous study for potential confounder

o Randomisation

o Matching
 Limited use outside of case control studies

Question 7

How do you control for confounders in the analysis phase

Answer 7

o Stratification
 Assess exposure-outcome relationship independently for different confounder strata

o Standardisation
 Model exposure-outcome relationship weighted by the potential confounder

o Adjustment
 Most commonly used procedure
 Model relationship while controlling for potential confounders (multivariate modelling)

Question 8

What is stratification

Answer 8

 Assess exposure-outcome relationship independently for different confounder strata

Question 9

What is standardisation

Answer 9

 Model exposure-outcome relationship weighted by the potential confounder

Question 10

What is adjustment

Answer 10

 Model relationship while controlling for potential confounders (multivariate modelling)

Question 11

What is residual confounding

Answer 11

Residual confounding = distortion between association between exposure and outcome even after controlling for confounding in design and analysis

Question 12

Disadvantages of restriction

Answer 12

Limits generalisability
Difficult if more than 1 confounder

Question 13

Disadvantages matching

Answer 13

Increasingly difficult with more confounders
Cannot examine effect of matched variables

Question 14

Disadvantages randomisation

Answer 14

Limited use in observational studies