Basic Epi Flashcards
How do you calculate prevalence?
of existing cases/# of individuals in study population at one point in time
How do you calculate prevalence if you only have incidence rate?
multiple by disease duration and you get prevalence (product)
What is a major drawback of using prevalence as disease measure? Give an example of this with a health condition/disease.
It only gives one snapshot in time/is cross sectional…
Essentially, this could make treatment look harmful if treatment prolongs living with a condition like HIV. Those who being treated are still alive while those who were not treated perhaps have died… there will be a higher prevalence of treated HIV cases v. untreated.
How do you calculate cumulative incidence?
calculation: # of incident cases in a time period (t0 - t1)/ # of individuals at risk *at the beginning of the time period**(at t0)
What is a synonym for cumulative incidence?
risk
When an outcome is rare, what does prevalence equal?
prevalence odds
What is the range of prevalence?
0-1
What is the range of cumulative incidence?
0-1
What are the conditions under which you would use cumulative incidence as a measure of disease?
Best for observing new cases over a short period of time in a population assumed to be closed and exposure what assumed to be at the same time (e..g, food-borne illness, cruises, etc.)
What is always required for interpretation of cumulative incidence>
time-frame
Ex. “over the study period” (i.e., how many new cases observed during a discrete period of time)
What are some drawbacks of using cumulative incidence as a measure of disease?
- proportion not a rate… does not incorporate dimensions of recovery or exposures that change over time –> tends to 1 over time
- requires that everyone be followed the entire time (cannot have loss to follow up, competing events, etc. otherwise you cannot determine the numerator)
How do you calculate odds? What is this effectively comparing?
Odds = p/(1-p)
OR:
Odds = Pr(Y=1)/Pr(Y=0)
*effectively comparing occurrence (numerator) to non-occurrence (denominator)
What is a major drawback of odds as a measure of disease?
Difficult to interpret
high prevalence = ____ duration OR ___ incidence rate
long duration or high incidence rate
how do you calculate incidence rate?
of new cases during a time period/total person-time accumulated during that time period **amongst those at risk**
What does incidence rate mean in lay words and a metaphor for it?
measure of the “speed” at which people develop disease (“force of morbidity”)
What is a synonym for incidence rate?
Hazard rate
What does person-time account for?
The fact that during a time period, people can change exposure groups (e.g., vaccinated v. not) or if they have contracted a disease/condition, they are no longer “at risk”
What is the difference between closed and open cohorts? Give an example of each in terms of Harvard students.
closed cohorts are defined by an event (like a threshold walking through) while open cohorts are defined by possessing a set of characteristics
EX. individuals who were Harvard PhD students (closed) admitted in 2020 v. currently enrolled Harvard PhD students (open)
What are the main calculations used for survival/Kaplan Meier curves? How do you determine a risk set?
- conditional cumulative incidences at each time point when a case occurred by: # events/risk set
- risk set: # in prior risk set - # events - # censored
How do you calculate cumulative survival proportion?
calculate cumulative survival proportion by multiplying ALL conditional survival proportions (1-conditional CI) up until time point of interest
What are some reasons for censoring?
administrative: study ends before some people have outcome of interest
competing risk: event that removes someone from having an outcome (EX. someone dies from heart disease before getting cancer - what we’re interested in)
loss to follow up: dropped out
What are the three types of censoring?
right: individuals censored at end of time interval in study
interval: individuals censored at midpoint or some point in the interval between visits or study contact points
left: people enter study with condition (e.g., some 10 year old girls already had puberty before the start of our study)
Is incidence rate a probability?
NO!!! it’s a rate that incorporates time dimension, not probability…
What is a drawback of using incidence rate?
person-time is not an intuitive way to interpret public health impact
How can you calculate cumulative incidence if you have incidence rates?
1-e*integral of IR over each time interval
What is a synonym for cumulative incidence ratio?
Risk ratio
How do you calculate CIR/RR?
cumulative incidence in the treatment group/cumulative incidence in the untreated group
(Pr[Y=1|A=1]) / (Pr[Y=1|A=0])
When outcome is rare in a CLOSED cohort, odds estimates _______.
cumulative incidence
What is the implication of using CIR (RR) v. risk difference?
- using the ratio tells the degree of risk between unexposed/exposed (EX. those in the exposed group have X% greater risk of death compared to those unexposed)
- using difference gives an interpretation of the public health impact (EX. Smokers had 11 excess cases of lung cancer per 100 individuals compared with non-smokers.)
What is exchangeability and why do we need it for causal inference?
exchangeability — assumption that the exposed and unexposed groups would have experienced the same risk of outcome had they received the same treatment level (i.e., assumed to be identical with exception of the treatment variable)
*we need it for causal inference because it is impossible to observe the same population (the exact same individuals) in one time period under exposed and unexposed conditions —> must compare exposed to group to assumed exchangeable unexposed group
What are the 4 assumptions needed for causal inference/
- exchangeability
- consistency/SUTVA
- positivity
- well-defined interventions & outcomes
Define consistency with respect to counterfactual theory.
consistency — the assumption that a participant’s counterfactual outcome under their observed exposure is equal to their observed outcome
What does SUTVA stand for? What is the assumption?
stable unit treatment value assumption
assumption that there is no interference or heterogeneity in treatment
What is interference with regards to SUTVA and what are some examples?
condition where treatment status of a participant may affect potential outcomes of other units
Ex. spillover effects, herd immunity with vaccination as treatment
What is the average causal null hypothesis?
That the probability of the outcome would be the same if the population were exposed v. if they were not exposed.
What is the difference between the average causal null and sharp causal null? What is an example?
sharp causal null –> counterfactual outcomes under exposed v. unexposed conditions would be the same for ALL participants
average causal null –> *the probability* of the outcome would be the same for the population was exposed v. unexposed but individual counterfactual outcomes under exposed v. unexposed conditions could vary
Ex. if people given BP medication…
sharp causal null–> everyone’s BP having been given the medication = BP having not been given the med
average causal null –> average of everyone’s BP having been given the medication = BP having not been given the medication BUT for some, BP without medication may have been higher while for others, BP without meds would have been lower…
What is the difference between association and causation? What is needed to claim association = causation?
association –> difference in risk amongst two disjoint subsets of population
causation –> difference in risk in the same population under 2 exposure values
**need exchangeability
Why does ATT (average txt effect among the treated) need weaker assumptions than other causal inference?
Because you are making comparisons about the observed outcome in a treated group compared to their baseline untreated outcome.
