Calculating intrasubject variability

[placebo]

I'm wondering which is the proper approach for an assessment of intrasubject variabilty.

I have a series of patients of whom I have taken a parameter (contineous variable x) on 4 different days. How to check for the intrasubject variability?

What additionally complicates the matter is, by the virtue of experimental design, I already expect a uniform trend between the 4 readings. For example Expt1, 2, 3 and 4 should show a steady decrease of the variable x. The results are haywire now and what I expect is the increased intrasubject variabilty is causing this. How to find it out?

Thank you in advance.

I have heard of mixed Model ANOVA which could be used here. But not sure of how to go about it

[TheAnalysisFactor]

If you just want an estimate of intrasubject variability, not accounting for other variables, just put in subject as a random factor. That would be your only independent variable. Your DV is whatever you want the variability for.

You'll have to use a mixed models procedure--Proc MIXED in SAS, Mixed Models in SPSS, etc.

If you tell me which software you're using, I can tell you how to do it (depending on which one you use).

Karen

[placebo]

Thank you Karen.

I use SAS.

I have the same variable reading taken on four different days. Now I have to check the intrasubject variability.

In SAS I could test only variables on two days. Can I check the intrasubject variabilty considering all the four resdings?

What is the exact meaning of intrasubject CV .. I am getting something like 93% . What does that actually mean? (My stats knowledge is limited, as actually i am a physician)

Thank you very much

[placebo]

Hi Karen,

Can you also tell me how to do it in SAS? That would be a great help.

[TheAnalysisFactor]

Hi Placebo,

I don't have SAS in front of me, and honestly, it's been a while since I've done one of these, so this is off the top of my head (maybe someone could confirm this for me?).

PROC Mixed;
Model DV= ;
RANDOM Subid;
run;

Where Subid is the subject ID number and DV is the variable you want to measure the intrasubject variance of.

I don't know what you mean by you could only test the variance on two days. There is no reason to not include all four, as long as all four day's measurements are in the same column.

Karen

[placebo]

Thank you Karen,

We could calculate the intrasubject CV.

There is another doubt.

For the calculation of intrasubject CV on four different days, the assumption is that the experimental set up is identical on all the four days. But in my case I'm doing four experiments on 4 different days over six months. By the virtue of the design of experiment, I have to expect a downward trend of the variable from day 1 - day 4. Is there any correction in Mixed model ANOVA, to account for this already expected downward trend. Downward trend can be calculated by taking means of all the 100 patients.

Thank you

Placebo.
www.clinpharm.********.com

[TheAnalysisFactor]

Thank you Karen,

We could calculate the intrasubject CV.

There is another doubt.

For the calculation of intrasubject CV on four different days, the assumption is that the experimental set up is identical on all the four days. But in my case I'm doing four experiments on 4 different days over six months. By the virtue of the design of experiment, I have to expect a downward trend of the variable from day 1 - day 4. Is there any correction in Mixed model ANOVA, to account for this already expected downward trend. Downward trend can be calculated by taking means of all the 100 patients.

Thank you

Placebo.
www.clinpharm.********.com

Hi Placebo,

I have to say, I'm at the point of giving advice without being sure I exactly understand the experiment.

That said, if what you're saying is that you want to measure the intrasubject variability of a variable, which I will call DV, after accounting for the fixed and expected effect of Day, you would enter Day as a fixed effect in the model.

PROC Mixed;
CLASS Subid Day;
Model DV= Day;
RANDOM Subid;
run;

I realized I forgot the class statement in the previous code.

If you expect the effect of Day to be a linear downward trend, you would leave it out of the class statement. If you leave it out, it will be treated as continuous and the model will fit a slope. If you put it in the class statement, it will be treated as categorical, and the model will estimate the difference in means on the four days.

Can anyone else confirm my code?

Karen