Measuring association and making causal inference Flashcards
What is the epidemiologic approach?
Is there an association between a factor and the disease?
If yes, is it a causal relationship?
What are two ways to quantify the strength of an association (for a binary outcome)?
- difference
2. ratio
What is a binary outcome?
dependent variable we are looking at in a model
ill/well
case/control
alive/dead
If you had to translate the incidence ratio into words what would you say based on this example:
Tx 1 cumulative incidence = 0.5
Tx 2 cumulative incidence = 0.3
incidence ratio = (0.5/0.3) = 1.7
Treatment one increases the cumulative incidence of remission by 1.7 fold comparing to treatment 2
What type of difference measurements can be used for measures of association - binary outcomes?
Risk difference (proportion) Rate difference
What type of ratio measurements can be used for measures of association - binary outcomes?
Risk ratio (proportion)
Rate ratio
Odds ratio
What range do proportions, risks, cumulative incidence have to be in? Rates?
Risk = 0-1 Rate = > 0
What does a difference of 0 imply?
no association
What do difference measures range from?
- infinity to + infinity
What do ratio measures range from?
0 to infinity
What does a ratio of 1 imply? > 1? < 1?
1 – No association
> 1 – exposure is positively associated with disease
< 1 – exposure is negatively associated with disease
By looking at ratio, how will you decide which has the strongest association?
ratio further away from 1
How do you calculate a risk ratio?
risk proportion 1 / risk proportion 2
How do you calculate risk rate?
diseased / population at time
How do you find the risk?
diseased / population at risk
Define odds
the ratio of the probability an event will occur to the probability it will not occur
Is odds itself proportion, ratio, or probability?
ratio
What can odds range to if we calculate it?
0- infinity
What will the odds ratio tell us in words?
based on the conditions you are looking at factor x with increase the outcome odds by ___ fold compared to factor y
Calculate the odds.
of dz animals/ # of non dz animals
When thinking about causal association what 4 things could the association be due to?
Bias
Confounding
Chance
Cause
What are the three main types of bias?
- Selection
- Information
- Specification
T/F
Bias might be reduced or controlled in study design
True
What else do we call bias?
systematic error
What do you look at when deciding if something was associated by chance?
statistical significance.
What is a selection bias?
The group used for a study is not a good representative of the population. They have been selected to fit certain criteria.
What is an information bias?
The information in the study was inaccurately collected. The question was worded wrong, the recipient may have knowingly acted one way
What is a specification bias?
Using a wrong statistical model that may make assumptions. It may make the wrong assumption
What is confounding?
a lack of comparability between study groups resulting in a bias in the estimation of the effect of interest.
What must a confounder be?
- cause of the outcome, independent the exposure
- correlated with exposure in the study population
- Not part of the causal pathway
T/F
A confounder is affected by exposure or outcome
False
What is chance also called?
random error
What is a valid result?
This is an unbiased one. We are measuring exactly what we want to measure
What is a precise result?
Has small variability
Use confidence interval to look at precision
What are 4 major threats to a study’s validity?
selection bias
information bias
specification bias
confounding
What are 3 factors that may affect precision of a study?
Random variation (error) Sample size Study design (efficiency)
What are Hill’s guidelines for causal inference?
- strength of association
- consistency
- specificity
- temporality (cause must occur first)
- biological gradient (dose-response)
- plausibility
- analogy
What do Hill’s guidelines for causal inference help us do?
Helps us decide if a situation is more likely to be causal or not
What is more likely to be causal, strong or weak association?
strong, however, a weak association can still be causal
Why can’t we rule out causal association if there is poor consistency?
some effects are produced by their causes only under unusual circumstances.
What does specificity suggest? Is this a good recommendation to assess causal relationship?
requires that a cause leads to a single effect, not multiple.
Not a good recommendation – most causes can lead to different effects
What is a must for causal relationship according to Hill’s guidelines?
Temporality
refers to the necessity that the cause precedes the effect in time
What is the analogy part of Hill’s guidelines?
Similar results can be found elsewhere
T/F
One cause leads to one disease
False
Diseases have a multifactorial etiology – various factors interact to result in disease
Disease is an ecological problem
What are the A-B-Cs?
Is the observed Association due to: Bias Confounding Chance Cause