EPI 201 Midterm Flashcards
Relationship between CI and IR
CI = 1 - e^(-IR*t)
Which interval does a censored person go in?
The one that ends soonest AFTER they are censored
What do you need to include with a CI?
Time period
What do you need to include with an IR?
Person-time
What do you need to include with a prevalence?
Point in time
Measures of association
Cumulative incidence ratios/differences, prevalence ratios/differences, incidence rate ratios/differences
Traditional definition of a confounder
- Must be associated with exposure in the study base
- Must be a cause or correlate of a cause of the outcome among the unexposed
- Cannot be a downstream consequence of exposure or outcome
Definition of a prospective cohort study
Exposures measured for research purposes before the OUTCOME occcurs
Study base types
Primary, secondary
Types of case-control sampling
Density - IRR, risk-set (subset of density) - IRR, case-crossover - IRR, cumulative incidence - CI (under rare disease assumption), case-cohort - IRR
Measures of association for a closed cohort study
IRs, CIs, ORs…
Hill Criteria
Strength, consistency, specificity, temporality, biologic gradient, plausibility, coherence, experimental evidence, analogy
Popper postulates
Knowledge accumulates by falsification
Scientific theories can only be falsified, never proven
Scientific hypotheses must have empirical content and be falsifiable
Problems with random digit dialing
Probability of having a phone may relate to exposure
Probability of answering may relate to exposure
Different ratios of people to phone
Area codes no longer relate to geography
Key point with hospital matched controls
The control condition must be UNRELATED to the exposure of interest
Attributable risk
CID = CI_E+ - CI_E-
Population attributable risk percent
Pr[A=1|Y=1] = (CI_tot = CI_E-)/CI_tot
Number needed to treat
1/CID
Population attributable risk
CI_tot - CI_E-
Attributable rate
IRD
Attributable risk percent
(CIR-1)/CIR * 100
Measures of association for an open cohort study
Odds, IR, only approx CIs using exponential formula
Definition of D-separation
Two variables in a DAG are D-separated if all paths between them are blocked
What happens if you condition on a descendant of a collider?
This is just like conditioning on the collider itself – opens a path
Problems with ecological studies
Often have poor measure of exposure
Often have no info on who is exposed
No info on if exposed are the ones getting the outcome
No data on individual-level confounding
In what kind of study can you directly measure CI?
Closed cohort study (however, you would often use IR in a closed cohort situation due to LTFU, changes in exposure over time, etc). In an open cohort, you cannot measure CI directly, but you can estimate it using the exponential formula.
Problems with CI
Tends towards 1
No info on when event occurred
Does not allow for changing population size
Does not allow for time-varying exposures
Events may not be observable due to censoring
Probability tree notes
L=0 (0.33) 20
Non-treatment variables in circles
Relationship between prevalence and IR
prevalence odds = IR * duration (under steady state)
Difference between confounding and selection bias
Confounding occurs when there is a lack of exchangeability between exposed and unexposed groups. Selection bias occurs when we condition on a collider.
When is it appropriate to use the NON-exponential formula?
To approximate CI using the IR, you can use the NON-exponential formula C=I*t when the product of I and t is small.
