I am not going to address the first question, leaving it for Robin or
Dale. I will suggest that if you have multiple visits of the same
patient through time, the CS error structure is unlikely to provide the
best fit. Look into the time series (i.e., AR, ARH) and spatial power
(i.e., SP(POW)) error constructs.
As for the second question, look at the SLICE option or the
ESTIMATE/CONTRAST statement. Both should allow you to conduct the test
you are interested in.
From: mkhah@JSSRESEARCH.COM [mailto:mkhah@JSSRESEARCH.COM]
Sent: Monday, February 16, 2009 12:26 PM
Subject: proc mixed model
For my first experience in proc mixed , I want t o run mixed model for
change in serum creatinin in different visit time, I consider it as
repeated measure case .
I have some variable such as pt_id sex renal_stratification visit
treatment_type period (difference between visit date and baseline
date) serum_creatinine (baseline) serum_creatinine( in each visit) and
dif_serum_creatinine(change of serum creatinine at each visit after
baseline). I have to show the actual value of serum_creatinine in each
visit, change of serume creatinine compare to baseline and percentage
of serum creatinine change in two treatment group (TX group and
1. whether the statement that I am using is correct.
PROC MIXED DATA=data file METHOD=reml;
CLASS pt_id sex renal_stratification visit treatment_type
MODEL dif_seume_creatinine=seume_creatinine_baseline sex
renal_stratification visit treatment_type
REPEATED period/sub=pt_id R RCORR TYPE=CS;
2. How can I find whether there is statistically significant
difference between seume_creatinine change compare with baseline in
particular visit, let say I have 6 equal visit points after baseline?
Thank you for your help,