Mianalyze and nonparametric test for trends with ordinal variables

[Giampaolo]

Dear SAS users,

I would like to test the trend for the severity of cancer prognosis (continuous) vs. an ordinal variable (e.g. education). 

With a single dataset I would use a non parametric test such as jonkeree terpstra but having performed multiple imputation  I need also  to pool the results.

I  used GLM with the estimate statement to obtain a standard error I can use with mianalyze, but I am not sure if I can do this for nonparametric data. I am also  not sure what approach would be appropriate with an ordinal variable. Could you please advise on this problem?

Thank you

title 'GLM by &catvar';

proc glm data=MI_FCS2 ;
by  _imputation_ ;
class &catvar;
model Prognosis = &catvar / solution;

estimate 'edu3' intercept 1 &catvar 0 0 1;
estimate 'edu2' intercept 1 &catvar 0 1 0;
estimate 'edu1' intercept 1 &catvar 1 0 0;
estimate 'edu1-edu2'         &catvar 1 -1 0;
estimate 'edu2-edu3'         &catvar 0  1 -1;
estimate '(edu1,edu2)-edu3' &catvar 0.5 0.5 -1;
ods output estimates=est_ds;
run;

[SAS_Rob]

There isn't a good option here for combining the results from the JT test directly from Proc FREQ since it only reports J*.  One option might be that, since the JT statistic follows a standard normal distribution, you could hand-calculate the J and Var(J) in the formula.  Once you had those you could feed them into Proc MIANALYZE using the MODELEFFECTS and STDERR statements.  I can't think of any reason why this wouldn't be valid given the distribution, but you may want to dig a little deeper in the literature to confirm that. 

[Giampaolo]

Thank you so much!