output effect estimate from effectplot of a variable modelled by spline term

sasbee — Thu, 15 Aug 2019 16:39:18 GMT

Hi,

I am running a linear mixed model with random intercept using glimmix to estimate the blood pressure over time of participants in a randomized control trial. I use a spline function to account for the sharp changes in blood pressure I observed over a treatment period, and my study question is if the blood pressure (bp) trajectory differ by treatmentgroup and exposureofinterest.

Here is my SAS code:

proc glimmix data=data order=formatted;
class visit release_id year treatmentgroup covariate1 covariate2 covariate3;
effect spl=spline(year/basis=bspline degree=3 KNOTMETHOD=list(1,4) details); 
model bp= spl|treatmentgroup|exposureofinterest
covariate1 covariate2 covariate3
/solution;
random intercept/type=un sub=release_id;
store model;
run;



proc plm source=model;
effectplot slicefit(x=y sliceby=exposureofinterest plotby=treatmentgroup)/clm; 
run;

I can visualize the effect using the effectplot function, however, I wonder if there's a way to output the effect estimate at each cross-section of time, ie. year=1 year=2? (is it possible to use the estimate and contrast function? but since year is included in the spline but not as a variable in the model, how do I specify time?

I also tried using the score function to output one unit of change in exposureofinterst by treatmentgroup, hoping to obtain just the effect estimate using the following code. But I am not getting any predicted lclm uclm. Is there anything wrong with my code?

data template; 
do intercept=0; 
do &x._0_L2=1;
do SEX=0;
do B_AGEGROUP=0;
do	RACE=0;
do	B_edu=0;
do	B_income=0;
do	B_married=0;
do	B_ALC_DAY=0;
do	B_smoke=0 ;
do	DASH=0;
	do assign="Lifestyle";
	do y=0 to 10 by 0.5; 
	output; 
	end; 
	end; 
	
	do assign="Placebo"; 
	do y=0 to 10 by 0.5; 
	output; 
	end; 
	end; 
end; 
end; 
end; 
end;
end; 
end;
end; 
end;
end; 
end;
end;
run; 

data template; 
set template; 


*calculate the predicated values for each value of t;
proc plm restore=plot2;
  score data=template out=estimate predicted=predicted lclm=lclm uclm=uclm;
run;

Thanks.

Re: output effect estimate from effectplot of a variable modelled by spline term

sld — Sun, 18 Aug 2019 18:56:15 GMT

This example

Example 38.15 Comparing Multiple B-Splines

appears to do what you want.

topic Re: output effect estimate from effectplot of a variable modelled by spline term in Statistical Procedures

output effect estimate from effectplot of a variable modelled by spline term

Re: output effect estimate from effectplot of a variable modelled by spline term