Lecture 11: Bias Part 2 Flashcards
What is the difference between association and causation?
Association:
-Occurs when theres an identifiable relationship between an E and O
-Implies that E might cause disease
Causation
-Implies that there is a true mechanism that leads from E to O (something that ensures lung cancer will happen)
How do we determine causality?
3 Ways:
- Statistical association
- Epidemiological association
- Casual inference/understanding
(can also relate to each other not independent)
What is a statistical association?
-Associations: 2 events must occur together more or less frequently than would be expected due to chance alone
-Formal statistical tests only test whether chance could produce observed difference (important but not everything need to consider systemic bias ie ones last lecture)
-Can use P value = statistically significant
if p< 0.05 then sat sig and can reject the null
if p>0.05 then not sat sig and therefore dont have evidence to reject the null
What are the Epi measures of association
-Relationship b/w and exposure and its outcome (odds ratios, risk ratios)
-The stronger the relationship & bigger numerical value of measure of association the more evidence it provides that it could potentially be causal
-ASK: does a significant and strong associated exists b/w the exposure and the outcome?
What is a causal inference: Bradford- Hill criteria?
- Temporal sequence (most important) where E precedes O
- Strength of Association (Risk ratio RR or odds ratio OR very high or very low–> less likely to be due to chance or bias)
- Dose-response (risk of disease increases directly wit the level of exposure)
-Changes in E are related to a trend in association - Consistency/Replication of findings
-Multiple studies with the same results (different populations, different circumstances, different study designs) - Coherence/biological plausibility
-But : new findings/changing knowledge
-Also consider that we sometimes know little about the underlying biology( remember the black box) - Not as relevant: Specificity
-Easier to support causation when the associations are specific –> but often multiple effects per cause and vice vera ex smoking - not as relevant: Analogy
-similar results in other species
-But may be specific to a particular species - Experimental (best if done with target species as close to ‘natural conditions’ as possible
What are the “additions” to Hills Criteria?
-Consideration of alternate explanations (ex cofounding, anything else that can explain this)
-Cessation of exposure
-If factor is a cause of disease, expect risk of disease to decline when reduce exposure but in some cases, pathogenic process may be too far gone or not reversible
-Consistency with other knowledge
What is the causal inference: Component-Cause Model?
-Allows us to take what er have learned from the stats and epi associations and try to infer causation
Necessary cause
-Precedes the disease
-If not present, disease can’t occur
-It will always be present if the disease is present
Sufficient cause
-Precedes the disease
-If present the disease always occurs
What are component causes?
-Multifactorial is broken down into component causes, which are individual factors that can be combined to create a single sufficient cause
-One of a number of factors that in combination, constitute a sufficient cause
-Removing one component of a sufficient cause will render in insufficient
ex sufficient cause of heart attacks
1. Smoking
2. Hypertension
3. Obesity
4.Lack of exercise
What is the importance of causal factors?
-Epidemiologist & public health are often interested in how much disease could be completely removed from the pop by removing the exposure of interest (pop attribute fraction)
-Population attribute fraction: if the proportion of disease in the pop that is attributable to the exposure (and thus could be eliminated if the exposure was eliminated)
-Can add up to more than 100% bc some components are involved in more than 1 sufficient cause, means we double count the role in each sufficient cause
What is a recap of component-casue model?
Necessary cause: Precedes the disease/outcome and MUST be present for the disease to occur (ie its NECESSARY) ex infectious (covid or influenza)
Sufficient cause: Precedes the disease/outcome and ALWAYS produces the disease/outcome typically multifactorial ex radiation exposure in thyroid occurs, radiation is only 1 way to get cancer, but not everyone who has radiation will get cancer
Component cause: One of a number of factors that, in combination, constitutes a sufficient cause
What is the web of causation?
-no single cause of disease- often an interplay of many factors
-Causes are interacting
-Causal webs help illustrate the inter-connectedness of possible casues
What is the bottom line of this lecture?
-Associations dont necessarily = causation –> need to assess study design, statistics, epidemiological associations etc in light of criteria for causation
(various criteria/models of causation can help)
-using bradford-hill criteria, component-casue model, web of causation
