PROC MIXED with nested/clustered repeated measures

[tka726]

Hello -

I’m running a repeated measures mixed model with lab data.   For each patient - lab value is measured at 4 timepoints, but at each timepoint the lab data value is measured by 2 separate devices. Each patient was also randomized to one of 2 treatments. I am interested in testing the difference in lab values by treatment, timepoint, and device.

Each patient has results at 4 timepoints, for 2 devices (8 records per patient). I am unsure how to specify the nested/clustered nature of device and patient. In the repeated statement, I included subject = ptno(device) -  is this correct?

Thanks!!

PROC MIXED DATA=labdat;

  CLASS ptno device time trt;

  MODEL labval = device time trt device*time trt*device trt*time trt*device*time;

  REPEATED time / subject=ptno(device) type=cs;

  LSMeans device*time*trt/ cl DIFF;

run;

Each patient has 8 records (2 devices, 4 timepoints each):

ptno  trt  device time labval

1 1 1 1 99

1 1 1 2 105

1 1 1 3 520

1 1 1 4 467

1 1 2 1 267

1 1 2 2 325

1 1 2 3 261

1 1 2 4 313

....

20  2  1  1  119

20  2  1  2  127

20  2  1  3  618

20  2  1  4  529

20  2  2  1  267

20 2  2  2  320

20 2  2  3  290

20 2  2  4  400

@SteveDenham, any suggestions?

[SteveDenham]

This looks as though it should work.  Is there something in the log or listing that indicates that there is an issue?

SteveDenham

[jiltao · Accepted solution]

There are different ways to model the correlations in this data. One approach is to use the kronecker product covariance structure --

repeated device time / subject=patno type=un@ar(1);  *** or un@un or un@cs;

Other ways are also possible, including using both RANDOM and REPEATED statements. See my paper below (pages 4 and 5 are relevant here).

https://support.sas.com/resources/papers/proceedings15/SAS1919-2015.pdf

Hope this helps,

Jill

[SteveDenham]

Follow @jiltao 's advice. She is who I would turn to if I had any questions.

SteveDenham

[tka726]

Thank you, both.

This paper perfectly answers my questions!