Hi all,
I was wondering if anyone had tried to replicate the mixed models method
of analysing crossover studies with period-dependent baseline
measurements as described in Kenward & Jones 2010 [1], in SPSS?
I have the blood pressure data of their first example (taken from [2]),
and wish to use SPSS to replicate their results (treatment effect
described briefly on page 9). They provide SAS syntax in the Appendix.
My results are different enough to make me think I'm not doing this
quite right:
Kenward & Jones: treatment effect = 6.67, se=3.04, p=.04
Me: Type(1)*Group(1) parameter estimate = 7.06, se=2.50, p=.018
I wonder if I have (a) set up the data correctly (the data used in the
example is a subset of the full data, and I am not sure how 'Period' has
been coded in the example once only two periods are used); and (b)
converted the syntax correctly to SPSS:
MIXED SBP BY Pernum Type Group
/FIXED=Pernum*Type Type*Group | SSTYPE(3)
/METHOD=REML
/PRINT=SOLUTION
/RANDOM=Type | SUBJECT(Subjid) COVTYPE(UN)
/REPEATED=Type | SUBJECT(Subjid*Pernum) COVTYPE(UN).
SAS code given in [1]:
proc mixed data = RM;
class Subjid Pernum Group Type;
model Z = Pernum*Type Group*Type /solution ddfm = kr;
random Type /subject = Subjid Type = UN;
repeated Type /Subject = Subjid*Pernum Type = UN;
run;
If anyone else has looked at this (or is interested enough to try it out
now!), then I'd be keen to compare notes. I can provide the dataset
off-list if anyone is interested.
Thanks,
Kylie.
[1] MG Kenward & JH Roger. The use of baseline covariates in crossover
studies. Biostatistics (2010), 11(1):1-17.
http://biostatistics.oxfordjournals.org/content/11/1/1.full.pdf+html[2] B Jones & MG Kenward. Design and analysis of cross-over trials, 2nd
ed, pg 244.
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
