multiple strata factors in CMH test

[fengyuwuzu]

I have three strata factors (age group, disease stage, sex) and when I want to do CMH test, someone told me that I have to put all three strata factors together, which give insignificant p value:

proc freq data=mydata;
    table agegr * sex * dis_stag* trt_arm * resp / cmh;
run;

but when I put the strata factors one by one, I found two of them (age group and disease stage) gave significant p value for a certain subgroup. 

table agegr * trt_arm * resp / cmh;
or 
table  dis_stag* trt_arm * resp / cmh;

Now the question is, can I analyze the data using only one strata, knowing that both the experiment was designed and subjects were randomized based on all 3 strata factors?  Is using all three strata factors in CMH more defend-able than using only one strata factors?  

[HB]

I think CMH tests are more about the same experiment done multiple times or on multiple sets of data.

http://www.biostathandbook.com/cmh.html

But the force is weak in this one and for sure I cannot say. 

It sounds like you've stratified a group into different buckets and want to know about the differences in the buckets.  I don't know if that is a CMH thing. If you're just looking at difference between buckets, I would think you could collapse buckets down and practically be in Chi square country.   

But a number of years since I did any serious stats it has been.