Confounding and Bias Flashcards
if we survey NYC and top baseball team is Yankees, can we see it is the US’s fave team? Why not?
No, bc of selection bias
what 2 studies have forward directionality?
cohort
RCT
selection bias in cohort/RCT
ppl dropout or are lost to follow up (may be unequal in each group)
selection bias in case control
hospital controls may have higher alcohol intake (or a diff characteristic) than community controls
internal validity
proper group selection + lack of error in measurement
for internal validity, you must have accurate measurement of __, __, and __
for internal validity, you must have accurate measurement of exposure, outcome, and association btwn the two
external validity aka
generalizability
external validity
ability to generalize beyond a set of observations to a universal statement
confounding
an association btwn exposure and outcome is observed, BUT only as a result of a third variable
crude and adjusted measures of effect are not equal when __ is present
crude and adjusted measures of effect are not equal when a confounder is present
to be a confounder, the extraneous factor must:
1) be a risk factor of the disease
2) be associated with the exposure
3) not be an intermediate step in causal pathway btwn exposure and disease
3 ways to prevent confounding
1) randomization
2) restriction
3) matching
randomization attempts to have equal __ of __ in groups
randomization attempts to have equal distribution of confounders in groups
pros of randomization
- convenient
- cheap
- straightforward data analysis
cons of randomization
- need control over exposure and ability to assign subjects to study groups
- need large samples
randomization: how many conditions of confounding are not met?
1 of 2
restriction aims to prevent __ of __ in study groups
restriction aims to prevent variation of confounder in study groups
restriction provides complete control over __
restriction provides complete control over KNOWN confounders
restriction pros
- conceptually easy
- handles difficult to quantify variables
- can be used in analysis phase
restriction cons
- limits eligible subjects
- inefficient to screen and then not use people
- residual confounding if categories are not narrow enough
- limits generalizability
- can’t assess interaction (can’t stratify by age when age is narrow)
restriction: how many conditions of confounding are not met?
2
restriction is good for when confounder is __, but __
restriction is good for when confounder is quantitative, but hard to measure
restriction exmaple
sexual behavior and HIV
injectable drugs is a confounder, quantitative, but hard to measure
SO, just don’t include individuals who use injectable drugs
matching 2 types
- frequency matching
- individual matching
frequency matching
number of cases with match characteristics are equalized
individual matching
pairing of 1 or more controls to each case based on sex, race, etc.
cohort study matching
unexposed black matched to exposed black
case-control matching
control age 50 to case age 50
matching pros
- less ppl required than unmatched studies with same hypothesis
matching cons
- expensive b/c extensive searching and record keeping is required
matching cons
- expensive b/c extensive searching and record keeping is required
2 types of analysis to control confounding
- stratification
- multivariate techniques
stratification holds ___ constant
stratification holds confounder constant
pros of stratification
- direct and logical strategy
- minimum assumptions required
- straightforward
- you may find an interaction
cons of stratification
- strata may not have equal numbers
- multiple ways to form strata when using continuous variables
- hard with multiple confounders
- categorization causes information loss
cons of stratification
- strata may not have equal numbers
- multiple ways to form strata when using continuous variables
- hard with multiple confounders
- categorization causes information loss
multivariate statistics uses __ to estimate what association would be if __ wasn’t associated with __
multivariate statistics uses stats to estimate what association would be if confounder wasn’t associated with exposure
pros of multivariate statistics
- continuous variables to do not need to be converted to categorical
- can control for multiple exposure variables at same time
cons of multivariate statistics
- potential for misuse
confounding issues
- can conclude a relationship when there’s not one
- can conclude there’s not a relationship when there is
confounding is NOT __
confounding is NOT an error
is an error to __
is an error to NOT control for confounding
is an error to __
is an error to NOT control for confounding
bias =
systematic design error that mistakes exposure’s effect on disease
what kind of studies is bias a problem in?
ALL studies
selection bias is flaws in
selecting participants
information bias is flaws in
gathering info
selection bias: relationship btwn disease and exposure is different from in those who
selection bias: relationship btwn disease and exposure is different from in those who were eligible but didn’t
in study looking at cancer and coffee, the study sample over-represents…
in study looking at cancer and coffee, the study sample over-represents controls who don’t drink coffee (d)
cohort study selection bias: lung cancer and abestos
who is over-represented?
unexposed with disease
lots of unexposed sick people at risk for lung cancers
exclusion bias
if you apply different eligibility criteria to cases and controls
EX: Reserpine and breast cancer
non-response bias is a type of __ bias
non-response bias is a type of selection bias
non-response bias
responses may be higher in diseased or exposed
could show associated when tehre is none
how to limit non-response bias?
- get people to respond
- characterize non-respondents as much as possible (they may be more likely to have certain traits)
types of information bias
interview bias
surveillance bias
recall bias
reporting bias
misclassification bias
misclassify unexposed as exposed is more common in cases
misclassification bias dilutes __ and __
misclassification bias dilutes RR and OR
interviewer/abstractor bias
interviewer probes cases more thoroughly for an exposure
prevarication (lying) bias
participants have ulterior motives
similar misclassification in cases and controls pushses association towards
1
5 ways to prevent bias
- careful attention to sampling
- minimize non-reponse
- standardize measurements
- training and quality control
- blinding
