Exam 2 Flashcards
Define case-control study
A study in which cases of disease are identified, and then the sample of source population that produced the cases is identified
Purpose of case-control study
assess whether exposure is disproportionately distributed between the cases and control
Strength of case-control study
less time and less expensive
small sample size
compare multiple exposures
useful for rare exposures
Limitation of case-control study
can’t determine incidence, prevalence or causality
recall and selection bias
not useful for rare exposure
criteria for selecting controls in a case-control study
- controls must come from the same source population as the cases
- controls must be selected independently of the exposure
Calculate Odds
Pr[Y=1]/Pr[Y=0]
Odds of disease among exposed
Pr[Y=1IA=1]/Pr[Y=0IA=1]
a/b
odds of disease among unexposed
Pr[Y=1IA=0]/Pr[Y=0IA=0]
c/d
interpretation of odds ratio
the odds of death in those living in a high pollution city is 1.65 times higher than the odds of those in a low pollution city over the 15 years of follow up
Calculate a 95% confidence interval of an odds ratio
write out
Define a randomized control trial
must have a large enough population……
the control group and the active treatment group will have similar characteristics at the time of random treatment assignment, the only difference between the groups at baseline is the treatment assignment
Purpose of randomized control trial
ensures the exposed and unexposed groups are exchangeable at time of randomization in terms of measured and unmeasured variables
strengths of randomized controlled trials
Reduces sources of bias and/or internal threats to validity concerns
Can determine cause and effect
Limitations of randomized controlled trials
costly and time consuming
treatment not well defined
noncompliance
participants and investigators may not be blinded
Kaplan-Meier curve
help look at risk over smaller times chunks, which mitigate issues with estimating risk when the population has loss to follow up
Intention to treat analysis
compare the incidence of outcome in those randomly assigned to treatment vs control, regardless of the treatment they completed or received
Benefits of intention to treat analysis
gives real-world estimate on treatment effectiveness under practical conditions where people do not always comply with their treatment assignment
As treated analysis
compare the incidence of the outcome in those ACTUALLY treated with A=1 and in those actually treated WITHOUT treatment A=0
Benefits of as treated analysis
provides a more realistic estimate of the effect of treatment in a real world setting
limitation of as treated analysis
the treatment group is no longer exchangeable because randomization is not preserved
could introduce bias
efficacy analysis
Compare incidence of disease among those actually treated with treatment who were assigned to treatment with those who were not treated with treatment and who were assigned to no treatment
benefit of efficacy analysis
reduce dilution of treatment effect
reflective of efficacy in ideal conditions
limitations of efficacy analysis
treatment groups are no longer exchangeable because randomization is broken
often over estimate the benefit of the treatment
define confounding
systematic differences between the exposure groups being compared that distort the true association between exposure and disease because a 3rd variable is:
1) risk factor for the disease
2)is unevenly distributed across exposure levels
3)AND is not consequence of the exposure
Define exchangeability
treatment group and control group are functionally the same, so if you which them in your experiment you will get the same result
criteria for confounding
associated with the disease among the unexposed
associated with the exposure in the source population
not a consequence of the exposure
DAG
A: Exposure
Y: Outcome
L: Confounder
—————–>
L. —-> A. Y
What method exists to control for confounding in the design stage
Randomization
Restriction
Matching
What method exists to control for confounding in the analysis stage
Standardization
Stratified Analysis
-Mantel Haenszel
Multivariate regression analysis
IRRmh
write out
RRmh
write out
ORmh
write out
Interpret an adjusted measure of association
the (measure of association) in the exposed group was (magnitude) times (lower/higher) than the (measure of association) in the unexposed group after adjusting for (confounder) categories
Magnitude of confounding equation
(RRcrude-RRadjusted/RRadjusted) x 100%
Positive Confounding
biased away from the null
exaggerating the association
Negative Confounding
biased toward the null
masking the association
Assumptions of the Mantel-Haenszel approach
the association is constant across strata
no residual confounding within strata
Limitation of the Mantel-Haenszel approach
computationally rigorous
need a very large sample size to have sufficient information in each RxC cell if we adjusted for multiple confounders
residual confounding
confounding that remains even after many confounding variables have been controlled
strengths of regression modeling
allows to adjust for multiple exposures
quantifies the relationship (strength and direction)
clear interpretability of covariates in the model
Regression used for continuous outcome
linear regression
Regression used for binary outcome
logistic regression
Assumptions of linear regression
Linearity: constant slope
Independent outcomes
Normally distributed residuals
Constant variance