cohort studies Flashcards
start with
cohort studies
environmental exposure
then find association with disease or outcome
temporality
timing of info about cause and effect
we want info about cause and effect gathered at the same time
which studies have no temporality
observational studies:
cohort
cross-sectional
case-control
which studies have temporality
randomized controlled trial
limits of cross-sectional studies
hard to distinguish exposures and outcomes
limits of case-control studies
case’s recall of past exposures may be different than that of controls
start with
cohort studies
exposed and not exposed group (neither group can have disease)
folow up and categorize by disease or no disease
2 observation points
cohort studies
- determine exposure status and eligibility
- determine incident (new cases
variable of interest
cohort studies
incidence
we can find risk!
exposure-based cohort studies are good for
cohort studies
rare exposures
occupational groups may have more exposure to certain things
compare to non-exposed group of similar demographics
are cohort studies randomized?
no!
no randomization means
unclear whether exposure caused disease
or whether a confounder is involved
prospective cohort study
cohort studies
start with exposure status in present disease free population, then follow forward to future
retrospective cohort study
cohort studies
start with exposure status from past and track forward to present
follow-up has already taken place
reconstructed from records
prospective cohort pros
cohort studies
- exposure assessed before disease occurs
- can get incidence rate, cum. incidence, RR
- can study several outcomes to 1 cause
- NO RECALL BIAS
prospective cohort cons
cohort studies
- slow, expensive
- large samples (unless risk period is long: since we look at person-years)
- loss to follow-up
- not for rare disease
- not easy to reproduce
retrospective cohort pros
cohort studies
rare exposures
cheaper
occupational studies or rare events
may be quick
retrospective cohort cons
cohort studies
bad for rare disease
hard to get outcome
hard to get exposure
need to determine duration and intensity of exposure
combined retro and pro cohort
cohort studies
exposure from past objective records, may gather more info overtime and follow forward into future
design: assembling the cohort
cohort studies
all must be at risk
exclude ppl with history
maybe limit to higher risk (certain age group)
collab with other areas to increase sample size and generalizability
determine exposure status
cohort studies
questionnaires
lab tests
physical measurements
special procedures
hawthrone effect
hawthorne effect
cohort studies
be careful about asking about diet every visit, they may begin to eat differently
study begins to influence behaviors
measurement of exposures
cohort studies
must be comparable for exposed and unexposed
performed by someone blinded to exposure
examine at predetermined intervals
establish diagnostic critera BEFORE study begins
incidence in exposed
cohort studies
a/a+b
incidence in non-exposed
cohort studies
c/c+d
cumulative incidence
cohort studies
avg probability an individual in the cohort will develop disease over follow-up period
cum incidence = all incident cases over follow-up / all ppl at risk at baseline
when should we not use cumulative incidence?
cohort studies
when we lose a lot of people to follow-up
incidence rate
cohort studies
incidence rate = all incident cases over follow-up / person-years
measures of association
cohort studies
relative risk
rate ratio
attributable risk
relative risk
cohort studies
cum incidence in exposed / cum incidence in nonexposed
relative risk = 3 means
cohort studies
disease in exposed is 3 times that in unexposed
rate ratio
cohort studies
incidence in exposed / incidence in nonexposed
RR =
cohort studies
(a/a+b) / (c/c+d)
RR < 1
cohort studies
risk in exposed < risk in nonexposed
neg association
may be protective
RR = 1
cohort studies
no association, no risk
RR > 1
cohort studies
risk in exposed > risk in nonexposed
pos association
may be causal
attributable risk aka
cohort studies
risk difference
attributable risk
cohort studies
cum incidence exposed - cum incidence in nonexposed
shows total increase in incidence
absolute measures
greater public health importance bc they tell us incidence and risk difference
relative measures
determine causality
how strong association is
study population
actual study members
target population
population you want to generalize your results to
generalizability aka
external validity
good generalizability =
can generalize results to other populations
internal validity
is association btwn exposure and disease a good estimate of real life
what is internal validity impacted by
confounding, bias, chance
exchangeability
even distribution of confounders
unexposed and exposed should be same aside from exposure
lower exchangeability = more confounders
confounding issue in cohort studies
subjects choose their exposure
also smoking
other bias in cohort studies
outcome ascertainment bias
outcome ascertainment bias
cohort studies
no blinding so clinician may be influenced
information bias
quality of info obtained from each group must be equal
hard with retrospective
loss to follow up
we hope loss from all groups is equal
if loss to followup from group a is greater,
relationship is underestimated
if loss to followup from group c is greater,
relationship is overestimated
unequal loss leads to
bias of RR and AR
analytic bias
investigators and statisticians have preconceptions about findings
dangers of big sample size
more likely to see association even if it doesn’t exist
power
probability of finding a statistically significant association in data, given that association exists in population
survival analysis
product limit analysis
product limit analysis
1 - probability of not developing disease in any time interval
product limit analysis = 0.395
in absence of withdrawal, death, loss to followup, 39.5% of cohort would develop disease by time t