Dear all
I am interested in how one might test for sensitivity to loss (or exclusions
from analysis) in studies of diagnostic tests. But I cannot find any
information on how one might go about it. Ideally of course you would look
for other papers, but that is always not the real world.
It seems to me that applying the worst case scenario could be appropriate in
much the same way it can be done with loss in RCTs.
Below are details from on a case-finding instrument
Enrolled: 543
Analysed: 536 (7 excluded b/c data missing)
Cases identified by reference test: 97
Sensitivity: 96%
Specificity: 57%
LR+: 2.2
LR-: 0.07
Including the loss as false positives is likely to have little impact, but
including the participants as false negatives changes the likelihood ratios
to 2.0 and 0.18 and is thus the worst case.
Any thoughts would be greatly appreciated
regards
Andrew Jull
|