Dear Allstat,
I am currently involved in a study were patients outcome is assessed using
10 correlated endpoints. I want to investigate the treatment on these 10
endpoints and make a global statement about the significance of the effect.
To do this I plan to use the GLS statistic proposed by O'Brien in
Biometrics, 1984.
The papers I have consulted on the use of the GLS statistic discuss its
application properties in the two sample case, where the treatment effect
may be investigated using t-tests for the individual endpoints.
The study I am involved in does two treatments, however, a centre effect and
a covariate are also important in modeling the data. Hence, investigation
of the individual end points would require an ANCOVA model rather than a
simple two sample t-test.
My question is two fold. Firstly am I correct is saying that the GLS test
statistic can only be used to combine asymptotically normal test statistics,
such a t-stats and cannot be used in this way with F statistics produced by
the ANCOVA model? Secondly, if this is the case is there any extension of
the GLS statistics that can be used in this situation.
Thanks in advance
Lynne Holton
__________________________________________________________________
Astra Clinical Research Unit
10 Logie Mill, Logie Green Road, Edinburgh, EH7 4HG, Scotland
Tel: +44 (0)131 557 4929 Fax: +44 (0)131 556 6237
[log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|