topic Re: longitudinal proc mixed in Statistical Procedures

longitudinal proc mixed

akonamone — Sat, 25 Jun 2016 15:26:15 GMT

Hi - I am using PROC MIXED to estimate a longitudinal model. Below is a sample code

PROC MIXED DATA=TEST3 COVTEST NOCLPRINT NOITPRINT METHOD=ML;

CLASS SUBID RACE;

MODEL=TMA PRAC CAGE RACE|CAGE/SOLUTION DDFM=BETWITHIN OUTP=P ;

RANDOM INTERCEPT CAGE/SOLUTION TYPE=UN SUBJECT=SUBID;

RUN;

TMA is a continous outcome variable, PRAC is a cumumlative continous exposure variable (cumulative exposure to a toxic drug) and CAGE is a grand mean centered age variable, RACE is a three categroical level race variable. My main predictor of interest is PRAC. My question is in order to estimate the association between PRAC and TMA - over duration of study, I include PRAC|CAGE interaction in the model?

Thank you for your help

Re: longitudinal proc mixed

SteveDenham — Tue, 28 Jun 2016 14:47:51 GMT

it would certainly let you know if the relationship between PRAC and TMA differed due to CAGE. I did notice what is probably a typo in your model statement--the equals sign should follow TMA, not precede it.

Steve Denham

Re: longitudinal proc mixed

akonamone — Tue, 05 Jul 2016 18:46:11 GMT

Thank you for the correction.

Re: longitudinal proc mixed

akonamone — Tue, 05 Jul 2016 18:51:15 GMT

I have another related question for the cumumlative continous exposure variable PRAC. How can I query PROC MIXED to produce estimates of slopes at various points on the association curve between PRAC and the outcome. For instance, if I wanted to know the slopes of the outcome when PRAC = 50?

Thank you

Re: longitudinal proc mixed

SteveDenham — Mon, 11 Jul 2016 17:03:32 GMT

This looks like you need to use an ESTIMATE statement with a SUBJECT option to set up random-effect contrasts.

Or use the LSMEANS statement with an AT option. However, that wouldn't give you a slope, but rather a point estimate at, for instance, PRAC = 50.

The thing here is that the random intercept and slope are independent of the value of PRAC, so maybe I am completely misinterpreting your question.

Steve Denham