Bias and Confounding Flashcards
What kind of error is due to bias?
Systematic
Is there a formal way to deal with systematic error?
NO
What is validity in epidemiological context?
Relates to the absence of systematic error in a study result
*a valid measure of association in a study will have the same value as the true measure in the source population, except for error due to random variation
What is bias in the context of epidemiological studies?
The extent to which a measure of association from a study differs from the true measure of association in the source population
**only for differences due to systematic error
T/F: Systematic error can make a study result biased
True
systematic error
causes bias
Bias causes a study to be _______
Invalid
T/F: Random error can make a result biased
false
just makes the study wrong
What does it mean when a study value is accurate?
The study was free from bias and is valid
Can you determine whether a study has a systematic error when results are inaccurate the first time the study is run?
NO
systemic errors are repeatable - so on the first go at the study - you can’t determine whether your error is random or systematic
Will random errors result in inaccurate values that are far off from expected or close to expected?
Far off or close
Random errors can be large or small: the key is - each value will differ greatly (making it non repeatable and unbiased)
When you perform a study multiple times, and get values that are far from expected, but similar values each time, what kind of error(s) do you have?
Systematic - since the study repeatedly failed in the same way
*random error (in this case small) because the values were not exactly identical
What will ensure validity in epidemiological studies?
Proper study design
What is internal validity?
The study result is valid with respect to the population under study
What is external validity?
The study result is valid to a wider population.. beyond the study population and or source population
aka generalizability
Example study: assessing associations between risk factors and lymphoma in a sample of GR without lymphoma in Maine..
The study may be ____ valid if the study association was true
But may note be ____ invalid for other breeds, or for dogs in different populations/locations
internally valid (within the study population)
externally invalid (within all populations)
What is the study population, source population, and other populations, in the context of epi studies?
Study pop = the subjects in the study
Source pop = population from which the subjects were drawn
Other pop = target populations —> populations to which we may want to generalize our results
____ causes the measure of association estimated from the study population to be different from the true measure of association in the source population
Bias
**bias causes the study results to be different from those that truly exist in the source population
What is a non-differential bias?
A bias that equally affects the study groups
- diseased and non-diseased are equally biased
- exposed and non-exposed are equally biased
What is a differential bias?
A bias that affects one group more than another
Data from a study:
In the source population: OR = 2
- the true proportion of the dz cases with exposure was 20%
- the true proportion of controls with exposure was 10%
In the sample collected from the study: )R = 2
- the dz cases had 10% with the exposure
- the controls has 5% with exposure
Is this differential bias or non-differential bias?
Non differential
The bias was a 50% reduction in the proportion exposed in both the dz cases and the control populations
Two general sources of bias are ____ bias and _____ bis
Selection bias: error in selection of study subjects (the sample is different from the population)
Information bias: error in measurement (measure things with error or incorrectly) aka misclassification bias
Besides bias, what can contribute to systematic errors?
Confounding - an unknown factor distorts the relationship between the exposure and outcome
What does selection bias usually result from?
From groups being selected for the study not coming from the complete source population
T/F: Most veterinary observational studies have selection bias
TRUE
This is because samples are collected from convenient locations that provide easy access to animals
Ex: veterinary clinics, abattoirs, shelters (these don’t represent entire animal populations)
How do you reduce selection bias in a cross sectional study?
Random samples: identify all the individuals in the target population and randomly sample them
How do you reduce selection bias in case control studies?
Matching - ensure that controls have similar demographics, age, breed, gender, etc as cases
How do you reduce selection biased in cohort studies?
Matching: ensure that exposed and non exposed have similar demographics
What are some examples of selection bias?
- self selection bias - studies based upon volunteers
- Healthy worker effect: studies or working people or working animals -who are healthier than the population as a whole
- Diagnostic bias: diagnosis of a dz may be influenced by their knowledge of the exposure and dz (this can be reduced by having a clear, well defined case definition)
What will reduce information bias in cross sectional descriptive studies?
evaluate the accuracy of measuring tools and adjust your estimates to reflect the error
**error in measurement of dz in the sample may result in the prevalence estimate in the sample being different than the true prevalence of in the target population
If there is informational bias in a case control study, what will be affected?
The association will be bias
Ex: in a case control study, owners of sick pets (cases) are more likely to remember exposures than owners of healthy pets (controls)
What are some examples of information bias?
- observer variation (this is minimized by using standardized protocols)
- Deficiency of tools and technical errors
- Recall bias (those affected by dz have a greater sensitivity for recalling exposure)
- Reporting bias
The confounding factor influences the exposure, the outcome, or both? This results in distorting the measure of association
BOTH the exposure and the outcome
the distortion can be large and lead to over estimation or under estimation of an association
it can even change the apparent direction of an association
What three criteria must a third factor meet to be a confounder?
- associated with the exposure
- associated with the outcome
- Not in the casual pathway between exposure and outcome
What three ways can you reduce confounding variables in a study?
Before the study starts: if you predict that there is a confounder:
- Match the study - select cases and controls so that the confounding factor is equally represented in both groups
- Restriction - do not enroll any animals that have the confounding factor
- Randomization - will reduce confounding
What do you do if you realize there is a confounding variable after the study is completed?
Deal with the confounding factor in the analysis by stratifying
Stratify: partition the results based on the confounding factor
Ex: if the confounding factor is sex - split the data into males and females, then analyze both groups separately
