topic Re: proc mixed with repeated measures outcome fixed baseline exposure and features in Statistical Procedures

proc mixed with repeated measures outcome fixed baseline exposure and features

Jest — Fri, 22 Sep 2017 21:32:23 GMT

Hello,

I have a longitudinal dataset with a continuous score outcome and baseline features which I am trying to model. The outcome varies between visits/years but not the exposure or the other variables.

I have been using proc mixed but keep getting that the model stopped because of infinite likelihood. I then tried putting ID*exposure in the subject statement and the convergence criteria was met. I am unsure why it would only work as such and if I can trust the model as such. I am pasting below how my data is setup and the code.

ID Year Outcome Exposure Age v12 v13 v14 v15

1 0 96 0 50

1 1 97 0 50

1 2 98 0 50

1 3 94 0 50

2 0 99 1 44

2 2 93 1 44

2 4 92 1 44

2 6 95 1 44

3 0 91 0 57

3 2 98 0 57

3 3 94 0 57

3 4 96 0 57

This is the code I am using:

proc mixed data=data.fa;

class ID year exposure;

model outcome = exposure year outcome*year age v12 v13 v14 v15;

repeated/ subject=id*exposure type=cs;

run;

Any ideas why the convergence is only met when putting id*exposure in the subject statment in the proc mixed model above? Is this correct? Is the code above reading the data correctly? Do I need to format the data differently to make it read a longitudinal outcome with baseline fixed features?

Thank you!

Re: proc mixed with repeated measures outcome fixed baseline exposure and features

sld — Sat, 23 Sep 2017 20:07:02 GMT

Quick thoughts:

1. Having "outcome*year" in the MODEL statement as a predictor is just wrong when outcome is the response. You probably mean "exposure*year".

2. Does ID have a unique value for each subject, or are you using the same value of ID with each level of EXPOSURE? Regardless, using subject=id*exposure is a good idea, and it's absolutely necessary if you have ID=1, etc. for multiple levels of EXPOSURE.

3. It's a good idea to specify the repeated measures factor in the REPEATED statement. Otherwise, the dataset must be sorted correctly.

repeated year / subject=id*exposure type=cs;

4. There could be issues with v12 v13 v14 v15.

HTH

Re: proc mixed with repeated measures outcome fixed baseline exposure and features

Jest — Mon, 25 Sep 2017 15:50:42 GMT

Hi HTH,

Thank you for the response.

1) Yes you are correct it was exposure*year and not outcome*year...That was a typo

2) ID does have a unique value for each subject, however it is repeated for each measurement for the outcome and hence the baseline values as the data I described...

3) I added the repeated year in the statement and it worked. I got the same results