Fundamentals of epidemiology - an evolving text - Are you looking ...

Selection bias in cohort studies
Selection bias is generally regarded as a greater danger in case-control than in cohort studies. The
reason is that in cohort studies the investigator generally knows how many and which participants
were lost to follow-up, so that s/he can assess the potential extent of bias. The investigator can also
often examine baseline characteristics of participants who are lost to follow-up for indications that
attrition is uniformly distributed and therefore less likely to result in selection bias.
Population cohort attrition
There is, however, a type of attrition that affects both cohort and case-control studies but which is
unseen and difficult to categorize. The problem relates to the representativeness of people eligible
for study. For simplicity we explain the situation in relation to cohort studies, but since a casecontrol
study is simply an efficient method for studying the same phenomena as a cohort study of
the study base, the problem is effectively the same in case-control studies.
A cohort consists of people who are alive at a point or period in calendar time or in relation to some
event they undergo (e.g., graduating from college, joining a workforce, undergoing a surgical
procedure) and then followed forward in time. The investigators' attention is for the most part
directed at what happens after the cohort has been formed, but it is conceivable that mortality and
migration occurring before that point have influenced who is available to enroll and thereby
influenced what will be observed. If these early selection factors are related to an exposure under
study they may diminish an observed effect.
Some examples:
If a cohort of HIV-infected persons is recruited by enrolling HIV seropositive persons
identified through a serosurvey, those who have progressed to AIDS more quickly will
be underrepresented as will persons involved in risk behaviors (e.g., injection drug use)
that are associated with high mortality. Progression to AIDS in such a cohort will
appear different than what would be observed if people were recruited at the time of
initial HIV infection.
A study of the effect of hypertension in a cohort of elderly participants cannot enroll
persons whose hypertension caused their death prior to the entrance age of the cohort.
If those who died earlier had characteristics that made them more vulnerable to endorgan
damage from hypertension, then the cohort study may observe less morbidity and
mortality associated with hypertension than would be observed if the study had enrolled
younger participants.
Even a cohort study in a population of newborns can enroll only infants from conceptions
that result in a live birth. If environmental tobacco smoke (ETS) increases the rate of
early fetal losses, possibly undetected, there may be differences between the fetuses who
die and those who survive to birth. If fetuses who survive are more resistant to harm
_____________________________________________________________________________________________
www.epidemiolog.net © Victor J. Schoenbach 2001 10. Sources of error - 305
with Joellen Schildkraut and Wayne Rosamond, rev. 5/11/2001, 5/16/2001