Lecture 6 Flashcards
For a binary outcome, what are 2 ways of Quantifying the strength of association?
1. Difference
• risk diff
- "their risk has ↑ by X"
• rate diff
2. Ratio
• risk ratio (relative risk)
-"their risk ↑ X fold"
• rate ratio
• odds ratiowhat is an indicator of precision?
Confidence interval
- do ration btwn low vs high end of interval
- higher the number the less precise
what is an indicator of statistical significance?
p-value
What is the range used for difference?
What implies no association?
Neg infinity to Positive infinity
0 = no assoc
• farther away from zero, stronger assoc
What is the range used for Ratio?
What implies no association?
ZERO to indinity
1 = no assoc
> 1 = + assoc
< 1 = - assoc
Odd ratio
Odds
Odds ratio
• relative measure based on “odds”
Odds
• ratio of probability of an event occurring : the probability of it not occurring
• Range 0-infinity
What do you want to asses if you are trying to see if there is a causal relationship?
ABCs • Is the observed ASSOCIATION due to: • Bias • Confounding • Chance • Cause
How do you determine if there is an association?
- randomized controlled trial
* difference OR ratio of effect size btwn Tx groups
Bias
Aka Systematic Error
• Selection
• Information
• Specification
- Not removable by statistics!
- STUDY DESIGN – reduces bias
Selection bias
- loss to follow-up
- control selection is dependent of exposure of interest
Information bias
- misclassification
- recall bias
- not blinded to as Tx
Specification bias
- wrong statistical models
Confounding
- Lack of comparability btwn study groups
- results in bias in estimation of the effect of interest
- adjustable statistically
- ↓ w/ study design
Which can be adjusted w/ statistical methods… Confounding or Bias?
Confounding
What must a confounding variable be?
- A cause of the outcome
- (independent of exposure) - Correlated w/ exposure in study population
- Not affected by the exposure or outcome
Chance
aka Random Error
• statistics quantify probability that association is due to chance
• assumes random sampling & randomization
Measures
• P-value – small sample –> ↑ ↑ p-value
• confidence interval – Wide = non-precise
Valid result
= unbiased result
• measuring exactly what we WANT to be measuring
Precise result
= small variability
Factors affecting validity
- bias
* confounding
Factors affecting precision
- Random variation/error
- sample size
- Study design (efficiency)
Hill’s guideline
Used to determine causal vs non-causal associations
- Strength of association
- Consistency
- Specificity
- Temporality <—-MUST be there.
- Biological gradient (dose-response)
- Plausibility
- Analogy
Can a weak association be causal?
Yes, but a strong association is more likely causal
Does a strong association indicate cause?
NO! (think literbox example)
Consistency
Repeated observations of an association in diff populations under diff circumstances
• meta-analysis = good for assessing consistency
Does lack of consistency R/O causal association?
- No, some effects are produced under unusual circumstances
Specificity
One exposure –> one effect (not multiple)
• not often the case
Temporality
ONLY necessary criteria
• cause precedes effect
Biological gradient
Presence of monotonic dose-response curve (unidirectional)
Analogy
Finding similar results elsewhere
Multifactoral etiology
various factors interact to result in disease
• think of dz as being an ECOLOGICAL problem
