Share your ideas for the JMP Scripting Unsession at Discovery Summit by September 17th. We hope to see you there!
Choose Language Hide Translation Bar
Highlighted
Thierry_S
Level VI

Repeated Measure Mixed Linear Model with an Additional "Nested" Parameter?

Hi JMP Community,

 

I need help with a relatively complex Repeated Measure Mixed Linear Model:

Context:

I have collected data from subjects exposed to different treatment, at Baseline, Week 4, and Week 12 (repeated measures; see small excerpt attached). I have analyzed this data set using the Fit Model REML platform with the following model effects:

  • ARM
  • TIME POINT
  • ARM*TIME POINT
  • Patient ID[ARM] & Random
  • Patient ID*TIME POINT[ARM] & Random

Issue:

I now need to incorporate an additional effect  (i.e. PLATE ID) that needs to be nested with Patient ID too (correct me if I use the term incorrectly).

Assuming that this can be done with JMP 14.1 (not Pro), what would be the correct structure of the model effects? 

Would it be correct to only include a new term = Patient ID[PLATE ID] & Random  ?

 

Thank you for your help.

 

TS

 

 

Thierry R. Sornasse
6 REPLIES 6
Highlighted
statman
Level VII

Re: Repeated Measure Mixed Linear Model with an Additional "Nested" Parameter?

 I can give you my thoughts regarding your data set:

1. You have 4 Plate_ID's, 7 Patient ID and 5 ARM's and then have measures over 3 time periods.  If I read it correctly, you have only 7 unique combinations. The data looks like the picture below.

statman_0-1591126080727.png

 

2. In some cases, Patient ID and Plate_ID are confounded.  ARM is confounded  with Patient ID, except for when ARM=E.

3. There is no context for the DATA?  How much of change in this response variable is worthy of consideration (is of practical significance)?

4. You indicate you want a linear model, but the "factors" are more than 2 levels.

So how can you handle the time series data:

1. You could take a quick look at the consistency of the time series with a range chart.

statman_1-1591126651583.png

 

2. You could calculate a statistic to summarize that effect (average and standard deviation, attached JMP file) although the data looks to be inconsistent.

3. What are you looking for in the time series?  Would the slope of the line be of value?

Highlighted
Thierry_S
Level VI

Re: Repeated Measure Mixed Linear Model with an Additional "Nested" Parameter?

Hi Statman,

Thank you for taking the time to look at the small subset of the data I shared. Obviously, I have many more patients in this data set (n = 115) with 10 - 25 patients per arm. Unfortunately, I cannot share the entire data set hence the description.

While I truly appreciate your constructive approach, I'm already clear of what I want to achieve: LSMEANs contrasts between arm A and all other at each time point. I expect that a significant difference of +/- 0.2 will be meaningful biologically.
Where I actually need help is to clarify the correct model structure to account for a potential confounding factor linked to the assay platform (i.e. PLATE ID). Any thoughts?

Thanks

TS
Thierry R. Sornasse
Highlighted
statman
Level VII

Re: Repeated Measure Mixed Linear Model with an Additional "Nested" Parameter?

Sorry, I can only work with what you give me.  You made no mention this was a subset of your data set...

 

Regarding your question "Where I actually need help is to clarify the correct model structure to account for a potential confounding factor linked to the assay platform (i.e. PLATE ID)."

 

Simply enough, you can't separate confounded factors, so you must be careful drawing conclusions about said factors.

Highlighted
Thierry_S
Level VI

Re: Repeated Measure Mixed Linear Model with an Additional "Nested" Parameter?

Hi Statman,

Thank you for your efforts to help me. Sorry for giving misleading information: in my initial post, I wrote "see small excerpt attached". I understand that it may not have been clear that this as a subset of my data. I should have been clear that this is subset of data was only included to give a snapshot of the data organization and not of the entire data set.

Regarding my question, I realize that the use of the term "confounding" might have very specific connotations that I did not intend. Maybe, a better way to describe what I'm looking for is a recommendation on how to best account for the possible contribution of a batch effect (i.e. PLATE ID nested within PATIENT ID) in general when exploring data organized as shown in the small table subset I shared.

I know how difficult it is to provide feedback on an issue when the full data set cannot be shared (I have been in the same situation trying to help other users)

Best ,

TS
Thierry R. Sornasse
Highlighted
statman
Level VII

Re: Repeated Measure Mixed Linear Model with an Additional "Nested" Parameter?

No worries, I wish I could be more helpful.  I really don't understand what you mean by Plate ID is nested within Patient.  From the small example data set you included, there was no obvious nesting.  For a variable to be nested, it must be contingent upon another variable.  So for your example of batches, within batch samples are nested within batch.  I just don't understand your situation to provide better guidance.  I don't see how batches would be nested within Patient, but this is due to not understanding your particular situation.  If you do have a variable (A) and it is nested in variable (B), then you would add the term A[B] to the model.  Of course, this would not allow for interaction effects to be estimated.

Highlighted
Thierry_S
Level VI

Re: Repeated Measure Mixed Linear Model with an Additional "Nested" Parameter?

Hi Statman,

 

First, I screwed up: I got the Nesting relationship description inverted - i.e. Patient_ID is nested in PLATE_ID - not the other way around.

Second, I created an anonymized data set that will give the JMP community a much better look at the complete data set.

Third, I'm thinking of using the following fit model structure with REML personality:

  • ARM
  • TIME POINT
  • ARM*TIME POINT
  • Patient ID[ARM] & Random
  • Patient ID*TIME POINT[ARM] & Random
  • Patient ID[PLATE ID] & Random

Is this approach appropriate to explore difference between the effect of ARM and TIME POINT for these repeated measures (DATA) while accounting for PLATE_ID as a possible source of variation? FYI, a quick visual survey of the data suggests that the DATA is influenced by PLATE_ID, ARM, and TIME POINT.

 

Best,

 

TS

Thierry R. Sornasse
Article Labels