How to get P for trend

[ChuksManuel]

proc surveylogistic data =newnsch3a;
class   sex (ref = '0') race(Ref = '0') FPL (ref= '0')   smokes(Ref= '0') adulteducate (ref='0')
family (ref='0') sleep (ref='0') comp (ref='0') age3_1718 (ref='2')  MOLD (ref='2')
exercise (ref='0') pest (ref='0')/param=glm ;
strata Fipsst;
cluster hhid;
weight fwc;
Model ast (event ='1') = age3_1718 sex race FPL smokes  comp adulteducate 
MOLD exercise pest family sleep smokes*comp;
lsmeans smokes*comp/oddsratio cl diff;
slice smokes*comp/ sliceby=comp diff=control('0' '0') oddsratio cl; 
run;

Hello statisticians,

I am interested in finding the P for trend for the Odd Ratios (OR) that i got from the interaction between smoking and a composite variable that i created. Composite is a variable that has 4 level (0,1,2,3) while smokes have 2 levels (0,1). The interaction between composite and smokes was significant. So i have 4 ORs. My problem is that i would like to get a P for trend on these 4 ORs. Is there any other syntax that i should add to my code (above) to enable me get a P-value for the trend?

[SteveDenham · Accepted solution]

You can get a trend on the log-odds values working from the LSMEANS statement.  I think the lsmeans will have the log odds for the 4 levels of comp at smokes=0, followed by the log odds for the four levels of comp at smokes=1.  If that is the case, the following LSMESTIMATE statement will give the tests in the LOG ODDS space.

lsmestimate smokes*comp 'Trend due to comp for smokes=0' -3 -1 1 3 0 0 0 0,
                        'Trend due to comp for smokes=1' 0 0 0 0 -3 -1 1 3,
                        'Difference in trends'           -3 -1 1 3 3 1 -1 -3;

Of course if the ordering of the smokes*comp lsmeans is different than what I supposed, these will have to be adjusted.

SteveDenham