Association vs Causation Flashcards
Exposure -> Outcome considerations
- validity of association observed
- if valid, is the observed association causal?
Assessment of validity of association between exposure & outcome (3)
Need to rule out 3 alternative explanations
- chance
- bias
- confounding
Chance definition
Observed result may be due to chance
Relationship between
- inadequate sample size
- variability
- chance
smaller the sample size -> larger the chance of variability -> higher probability of observed association is due to chance
p-value definition
Probability that the observed result or extreme result is by chance assuming Ho is true
95% CI
- suggests statistical significance & precision
- affected by 3 factors (sample size, SD & confidence level)
Bias definition
- any systematic error in the design, conduct or analysis of a study, resulting in a mistaken estimate of an exposure’s effect on the risk of an outcome
Can bias be fixed once it has been introduced?
Can confounding be fixed if present?
Bias cannot be fixed once it has been introduced.
Confounding can be accounted for by various method (in-study design & in-data analysis)
2 major types of bias
- Selection bias
2. Information bias
Selection bias definition
- groups in the study are not comparable due to the biased nature of the selection of groups (case & control, exposure & non-exposure)
**
eg cohort study investigating effects of health screening on overall health outcome
((exposure))
case : individuals voluntarily participate in annual health screening
control : individuals who do not participate in annual health screening
Examples of selection bias **
eg cohort study investigating effects of health screening on overall health outcome
((exposure))
case : individuals voluntarily participate in annual health screening
control : individuals who do not participate in annual health screening
eg case-control study on potential protection from use of NSAIDs against colorectal cancer
((outcome))
case : patients with colorectal cancer
control : patients without colorectal cancer
eg patients with arthritis / peptic ulcer
Reason for chance affecting the validity of observed association (4)
- random sampling variability
- inadequate sample size
- p-value
- 95% CI
Relationship between
- p-value
- chance
- Ho
smaller p-value -> lower probability that the observed association is due to chance -> stronger evidence against Ho
Why selection bias affect validity of the observed association?
- results in an apparent association between exposure & outcome is observed even if in reality there is no association
- reduce validity of the observed association
Information bias definition
Stems from incorrect determination of - exposure - outcome - both , resulting in misclassification
Examples of information bias (3)
- Recall bias
- quality of recall - Reporting bias
- reluctant to report information accurately - Poor sensitivity of test kits
Types of misclassification (2)
& its effect
- Differential misclassification
- misclassification rate differ in different study groups
- lead to bias in both direction
eg association observed but in reality there is no association & vice versa - Non-differential misclassification
- similar misclassification rate in the study groups
- lead to results supporting null hypothesis & diluting association
Confounding definition
- a third variable that is associated with the exposure & can influence the outcome
Factors of a confounder (3)
- Associated with the exposure
- Risk factor for the outcome
- Not an intermediary step in the causal process of exposure to outcome
Methods to control confounders (2)
- In-study design
2. In-data analysis
In-study design (2)
- restriction of confounder
eg pick participants without confounder - matching
eg match participants from both groups with confounders
In-data analysis (2)
- stratified analysis
- multivariable analysis
eg multiple regression
After controlling for confounders, how to assess for association? (2)
- compare crude RR/OR & adjusted RR/OR
- interpret association based on adjusted RR/OR
How to assess if the observed association is causal?
Bradford Hill Criteria
- do not need to have all 9 factors
Bradford Hill Criteria
9 factors
ACCESS PTB
Analogy Coherence Consistency Experiment Specificity Strength Plausibility Temporality Biological gradient
Analogy
- any similar exposure to support observed association?
Coherence
- is the relationship between exposure & outcome coherent with the general knowledge?
eg natural history & biology of the outcome
Consistency
- is it consistent with other studies?
Experiment
- presence of experimental evidence to support the causal relationship
Specificity of association (2)
- weakest factor
- single exposure lead to a specific outcome
Strength
- strength of association
Plausibility
- biological system mechanism logical & possible?
Temporality
- factor occur first then outcome
Biological gradient
- is there a dose response relationship?
