L21: Association VS Causation

Question 1

How do you know if an observed association is valid?

Answer 1

Need to rule out 3 alternative explanations:

• chance
• bias
• confounding

Question 2

How is chance an alternative explanation?

Answer 2

• Observed result can be due to chance to random sampling variability.
• can be due to inadequate sample size –> large chance variability
• p-value (smaller the p value, the stronger the evidence against H0)
• 95% CI

Question 3

How is bias an alternative explanation?

Answer 3

• bias is any systematic error in the design, conduct or analysis of a study that results in a mistaken estimate of an exposure’s effect on the risk of an outcome
• once bias is introduced, cannot be fixed `

Question 4

What are the two main types of bias?

Answer 4

Selection bias and information bias

Question 5

What is selection bias?

Answer 5

• absence of comparability between groups studied
• if the way the exposed and unexposed (for cohort studies) and cases and controls (for case control study) were selected such that an apparent association is observed even if in reality exposure and outcome are not associated. the apparent association is due to selection bias

Question 6

What is information bias?

Answer 6

• incorrect determination of exposure, outcome or both –> misclassification
• reluctant to report past exposure (reporting bias)
• differential misclassification in the diff groups can lead to bias in either direction
• non-differential misclassification will bias the result to the null value

Question 7

What is confounding?

Answer 7

Third variable that is associated with the exposure, and influences the outcome.

Question 8

What are the three criteria for a confounder?

Answer 8

NEED TO MEET ALL 3 criteria:

• associated with the exposure
• risk factor for the outcome
• not an intermediary step in the causal pathway from exposure to outcome

Question 9

How do we control for confounding in study design?

Answer 9

• restriction

- matching (eg cases and controls can be matched by smoking status)

Question 10

How do we control for confounding in data analysis?

Answer 10

• stratified analysis (analyse separately)
• multivariable analysis (eg multivariable logistic regression) -> determine crude RR or OR first, then the adjusted RR or OR –> interpret associations based on adjusted RR or OR

Question 11

How do we know if an observed association is causal?

Answer 11

Using the Bradford Hill Criteria:
ACCESS PTB

A - Analogy (is an association established for a similar exposure?)
C - Consistency (in diff studies and diff populations)
C - Coherence (coherent with other data or knowledge)
E - Experiment (experimental evidence?)
S - Strength of association (stronger, more likely causal)
S - Specificity (not impt)
P - Plausibility (coherent with current body of biologic knowledge?)
T - Temporality (exposure occurred before outcome?)
B - Biological gradient (dose response rs?)