on 18/05/2004 16:50, Tero T Kivela at [email protected] wrote:
> Test for trend of survivor functions
>
> chi2(1) = 36.8827
> Pr>chi2 = 0.0000
>
> Intuitively, the classification which gives
> chi2(1) = 38.2276
> divides the cases better accoridng to survival than the one which gives
> chi2(1) = 36.8827
You have to look at the context. A test is not better because it has a
statistically significantly better association with survival. A test is
better because it has a greater information yield at a particular stage in
the diagnostic process.
Consider these questions:
1. Do the two tests give independent information? If so, then perhaps you
would be best to do both. If not, then one may have no prognostic
significance once the other one is known - it may not carry independent
information.
2. If they carry the same information, then there are other factors to
consider too. One test may be more expensive than another, or more invasive,
and its superior performance may not be justified in terms of these costs.
The better test may be the test that is most easily and readily done on the
greatest proportion of people. For instance, in Europe you need a glucose
tolerance test to make a diagnosis of diabetes, while in the US it's based
on fasting glucose. The European standard is 'better' in that the persons
with European diabetes are at higher risk than their American counterparts,
but the American diagnosis can be made simply and readily, allowing more
people to be tested. We consider that the superior performance of the
European diagnostic criteria are worth the money. As far as I know, the US
is debating this right now, having decided some years ago that they
preferred the simpler system.
I am always reluctant to be drawn into making a definitive pronouncement on
a clinical problem. Statistics can inform the choice of test, but cannot
dictate it.
Ronan M Conroy ([email protected])
Lecturer in Biostatistics
Royal College of Surgeons
Dublin 2, Ireland
+353 1 402 2431 (fax 2764)
--------------------
Just say no to drug reps
http://www.nofreelunch.org/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
