This
happens when the diagnostic approach is related to knowledge of the
subject's prior exposure to a putative cause (e.g. taking a certain
drug, being exposed in an outbreak etc [2]).
Example: let's suppose that a case-control study
is conducted to test if oral contraceptives (OC) are a risk factor for
endometrial cancer. A group of cases and an equal number of controls are
selected. Cases are selected at GP (family doctor) surgeries. Cases who
use OC may be more likely to be offered screening for endometrial
cancer either systematically or because of a side-effect of OC
(breakthrough bleeding). The chance of undertaking detection of
endometrial cancer is therefore higher among OC users than among other
cases i.e. the use of OC may cause the search for endometrial cancer (by
causing symptomless patients to bleed) rather than causing the cancer
itself. The result is that a higher proportion of cases report using OC,
with an overestimation of 'a', leading to an overestimation of the odds ratio.
Exposure |
Cases of endometrial cancer
|
Controls
|
OR |
Uses OC |
a↑ |
b |
OR↑ |
Doesn't use OC |
c |
d |
reference |
Total |
|
|
|
Sackett [2]
describes this example, where an innocent exposure may become a
suspect, if, rather than causing a disease, it causes a sign or symptom
which precipitates a search for a disease, as 'unmasking (detection signal) bias'.