[R] Repeated Measures design using lme
nortonsm at yahoo.com
Mon Apr 9 04:55:27 CEST 2007
I have what I believe is a repeated-measures dataset that I'm trying to analyze using lme(). This is *not* homework, but an exercise in my trying to self-teach myself repeated-measure ANOVA for other *real* datasets that I have and that are extremely similar to the following design.
I'm fairly sure the dataset described below would work with lme() -- but it'd be great if anybody can confirm that after I describe the dataset below)
The study involves measuring the effect of a drug on blood pressure. There were 16 patients in all and 6 replicate measures per patient of their blood pressure on one week (one measure per day). Two weeks later, a drug was introduced to 8 randomly selected patients in such a way that I had equal representation of the 4 age groups among the two treatment groups. Then, another two weeks later, 6 replicate measures per patient (per day) of blood pressure was retaken. So each patient had 12 total measures whether they were in the treatment group or in the control group (6 reads (R1-R6) in the baseline-week and another 6 reads (R1-R6) in the post-treatment week).
Background: 16 patients
Response measure: Blood pressure
Fixed Factor: 4 Age groups
Fixed Factor: Drug vs. NoDrug
Random factor: Day of the read (i.e. 6 replicate reads (R1-R6) at the baseline time, and 6 replicate reads (R1-R6) after the drug has had time to take effect)
Random Factor: Subjects 1-16
Patient AgeGroup BP(Blood Pressure) Read (replicate reads) Pre/PostTreatmentWeek Group
1 20-29 83 R1 pre Treat
2 20-29 81 R1 pre Control
3 20-29 74 R1 pre Treat
4 20-29 85 R1 pre Control
5 30-39 82 R1 pre Treat
… … … … … …
3 20-29 74 R2 pre Treat
… … … … … …
1 20-29 83 R1 post Treat
2 20-29 82 R1 post Control
3 20-29 86 R1 post Treat
4 20-29 84 R1 post Control
… … … … … …
I'm trying to do an analysis of variance to decide whether there is a measurable change in blood pressure between the Treat and Control groups.
Another issue is that some of the 16 patients didn't get all 6 replicate reads in their pre/post treatment weeks, so I need to include the na.omit function.
What I think I'm having the most trouble with is the repeated reads (R1 through R6) in the pre/post treatment weeks. I'm fairly sure this is a random variable -- their order or identify (R1 in pre-treatment week has no relation to R1 in the post-treatment week, etc). By placing Read as a random variable, am I covering myself there?
If I execute:
> summary(lme(BP ~ Group, random = ~ 1 | Patient, data = bloodpress, na.action=na.omit))
I get a result, but I'm not sure it's correct -- do I need to tell the model about the Read factor (R1-R6 in pre/post weeks)?
I'm really trying to set the right form of the lme() function call to decide
1) if there is a statistical difference between the Treat/Control groups and,
2) if one takes into account AgeGroup, is there a statistical difference between Treat/Control Groups, and finally
3) if I don't see a statistical difference, can someone recommend an R function that might solve the supplemental question, "given the noise in day-to-day blood pressure reads, and given that I wanted to have enough statistical power to observe a say, a 5% benefit in blood pressure, how many additional reads or patients I would need."
Basically, is lme() the proper function, and can someone offer any pointers on what my call to this function should look like to make the above to determinations?
More information about the R-help