|
Research Currents
Observational
Study Designs: Cohort Study
Imagine that a new drug, mungafenil, has become
widely used for erectile dysfunction (ED). A VA
clinical researcher suspects that mungafenil may
sometimes precipitate atrial fibrillation (AF).
How could this suspicion be formally tested?
In principle, a randomized trial would yield the
most definitive answer, comparing AF incidence
between ED patients assigned at random to receive
mungafenil or not. But often it may be impossible,
impractical, or unethical to manipulate patients’ exposure
to a drug or other agent for research purposes.
Instead, an observational study may be the next
best alternative. Here the investigator would not
try to influence anyone's use of mungafenil. Instead,
he/she would merely examine the association between
mungafenil use and AF among ED patients in routine
care.
Among various observational study designs, a cohort
study most closely resembles a randomized trial.
It involves first identifying two groups of patients:
one group of mungafenil users (the exposed group),
and a comparison group of mungafenil non-users
(the unexposed group). These groups might, for
example, be identified through medical records
or pharmacy data. Then the frequency of AF would
be monitored and compared between groups over a
defined time period, possibly through medical-record
surveillance.
A cohort study could be either prospective or retrospective:
•
In a prospective cohort study, the AF
episodes of interest would occur in the future relative
to when the study is initiated. This option lets
the researcher collect additional data concurrently
for research purposes that may not be captured
in medical records.
•
In a retrospective cohort study, the AF episodes
of interest would already have occurred before
the study actually got underway. Pre-existing medical
record data would be used to ”reconstruct” a
comparison between users and non-users over a follow-up
period already past. When sufficiently complete
and detailed archival data permit this option,
it can yield an answer relatively quickly and efficiently.
A key concern in any observational study is
the possibility of confounding. Without randomization
to balance the groups, mungafenil users may
differ
from non-users in ways that also influence
their risk of developing AF—for example, pre-existing
heart disease. If so, the observed mungafenil-AF
association could be distorted, mixing a possible
true effect of mungafenil with spurious differences
due to the confounding factors. To avoid bias,
the researcher would need to identify, measure,
and control for relevant confounding factors using
techniques discussed later in this series.
References
Koepsell TD, Weiss NS. Cohort Studies. Chapter
14 in: Koepsell TD, Weiss NS. Epidemiologic Methods:
Studying the Occurrence of Illness. New York: Oxford
University Press, 2003.
Gordis L. Epidemiology (3rd edition). Philadelphia:
W.B. Saunders, 2004.
Samet JM, Munoz A. Evolution of the cohort study.
Epidemiol Rev 1998; 20:1-14.
Prentice RL. Design issues in cohort studies.
Stat Methods Med Res 1995; 4:273-92.
|
