|
Selection bias occurs when the subjects studied are not representative
of the target population about which conclusions are drawn.
Target
population › Study population › Study sample
For
example, if you wish to estimate the prevalence of pain among
advanced cancer patients (the target population), the choice of
the study population and of the study sample can be crucial in
influencing the results. Table 1 demonstrates how different results
have been obtained from studies of heterogeneous populations.
Table
12.1 Prevalence of Pain in Some Different Studies
|
|---|
| Author
| N
| %
with pain
| Stage
| Source
|
|---|
| Foley
1979 | 397
| 38
| all
stages
| cancer
hospital
|
| Pannuti
et al 1980 | 291
| 64
| advanced
| oncology
dept
|
| Trotter
et al 1981 | 237
| 72
| advanced
| oncology
outpatients
|
| Foley
1979 | 39
| 60
| far
advanced
| hospice
|
| Haram
1978 | 607
| 66
| far
advanced
| hospice
|
| Twycross
1974 | 500
| 84
| far
advanced
| hospice
|
| Wilkes
1974 | 300
| 58
| far
advanced
| hospice
|
|
In
symptom studies it is usually impossible to obtain complete assessments
for the entire study sample, and the analyses are performed on
the subgroup of patients for whom symptoms were evaluated. This
can lead to bias if the distribution of symptoms is different
among the subgroup of non-responders compared with responders.
For example, a study assessing patients with HIV/AIDS in a palliative
care service found more severe symptoms among those patients who
were unable to self-complete questionnaires
(Butters, Higginson, et
al., 1992; Butters, Higginson,
et al., 1993).
Note that
an acceptable level of response (the compliance rate) does not
exist -- it depends on the expected range of error that we can
accept as not being relevant for the research question.
|