In article <[log in to unmask]>, Dr Nilesh Patel
<[log in to unmask]> writes
>Dear All
>
>Is the following reasoning correct?
>
>When evaluting the validity of trial about a therapy, ensuring that all
>patients are analysed in an intention to treat manner is important since
>the percentage of compliers versus non-compliers may be different in the
>placebo and intervention groups. In this way we are testing whether the
>offering of a specific treatment is more beneficial than offering placebo.
>
>Now when analysing whether patients would benefit from a series of services
>organised in a certain manner (e.g. stroke unit treatment) compared to
>routine care, how do you analyse patients who only at the end of (or
>during) the trial are found not to meet the inclusion criteria (e.g. not a
>stroke but a cerebral metastasis)?
Include them in the ITT analysis. You can do subset analyses if you
like, but these will be distinctly secondary to the main one.
>
>Surely because the entry criteria are not met, the patients should be
>excluded BEFORE the analysis.
Well, maybe they should have been. But they weren't. And the trial
therefore tests the hypothesis of interest to practical clinicians,
namely does the new treatment offer benefit to patients in a previously-
defined group, say those clinically diagnosed as having a stroke. In the
real world, any such group will include people who turn out to have had
some other problem, both in the trial and in routine practice to which
the trial is meant to be a guide. Since I can't exclude such patients
fro my routine practice, I want to know what my new unit will do for the
entire group who will be going in there, not just of a subset. Though if
it works overall, of course it's interesting to do some retrospective
analyses.
There was a nice BMJ article a few years ago on the benefits of DICE
therapy; how to extract plausible patterns from random data. It's easy
to do with subgroup analyses. And a tangential example is one of the
objections raised to Semmelweis's hand-washing; he wasn't taking account
of the detailed pathological findings in the deaths from puerperal
sepsis, which enabled their classification into a number of different
entities. That *didn't* affect the fact that his new procedure worked
for all of them, but it did distract attention.
>This is not the same as someone being
>allocated to the stroke unit and decides not to be admitted for any reason
>(e.g. the unit may be at another hospital). The latter would be analysed
>including the patient as "a patient who the trial intended to treat".
Indeed.
>
>Is my reasoning correct?
Up to a point, I think.
>
>Dr Nilesh Patel
>Director: Clinical Services
>The Brenthurst Centre for Rehabilitation Medicine
>Johannesburg
>South Africa
>
>
--
Richard Keatinge
homepage http://www.keatinge.demon.co.uk
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|