Bias Flashcards
A systematic error that causes the effect estimate to deviate from the truth
Bias
What are the three types of biases?
Selection bias, Information bias, and Confounding
an error resulting in systematic differences in characteristics between participants enrolled
in a study and potential participants who do not (or who are not asked to participate)
Selection bias
Both experimental and observational studies are subject to ________ and ______ biases
Selection and Information
Error in the measurement of exposure or outcome data that results in different levels of
accuracy for information collected from the comparison groups
Information bias
Selection bias is related to _____ validity
External
Information bias is related to ______ validity
Internal
Erroneous classification of the individual into a category other than that which he or she
should have been assigned
Misclassification
a bias resulting from the distortion of the exposure – outcome association by a third factor, i.e. a mixing of effects of the confounder and the exposure on the outcome
Confounding
Confounder ____ causes (can be a proxy) the outcome
independently
Confounder is associated with the ______
exposure
The confounder is ___________ across the levels of the exposure but does not result from the exposure
unevenly distributed
Observational studies particularly prone to ______
Confounders
What are the ways to prevent confounding?
Randomization, Restriction, and Matching
Theoretically balances confounders between arms of study
Randomization
Restricts study population to one level of potential confounder
Restriction
Match exposed and unexposed on potential confounding factor
Matching
What are the ways to assess and control confounding?
Stratification and Statistical modeling
Stratify the data by the ________ and examine the crude and stratum-specific estimates
potential confounder
If the stratum-specific estimates are similar (homogeneity of effect), pool the estimates with what test?
Mantel-Haenszel OR
If the ORMH differs from the ORcrude by more than X% (%-change criterion an arbitrary
value set a priori), then report ___________
the pooled estimate
Examine effect estimate from model with the exposure (may contain other variables)
Reduced
Examine effect estimate from model that differs by only the potentially confounding variable
Full
If the adjusted estimate differs from the reduced estimate by more than X% (%-change
criterion an arbitrary value set a priori), then report___
the adjusted estimate
