Conditional logistic regression

[rob]

Hello

I am analyzing a matched subjects case control study where some strata have one case with
multiple controls and other strata have multiple cases and multiple controls. I have read that
COX Regression would work for single case strata but not so well for strata with more than
one case because the method used on COX Reg for handling tied survival times, Breslow's
method, will not produce accurate results... My question is, is there some technique that can
be implemented that can handle this type of design within SPSS? Any help is greatly
appreciated. Thanks

Rob

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[MaxJasper]

Cox easily handles {case, control} = {1,1}, {1,n}

{m,n} can perhaps be broken into m*{1,n}

 

 

 

Hello

I am analyzing a matched subjects case control study where some strata have one case with
multiple controls and other strata have multiple cases and multiple controls. I have read that
COX Regression would work for single case strata but not so well for strata with more than
one case because the method used on COX Reg for handling tied survival times, Breslow's
method, will not produce accurate results... My question is, is there some technique that can
be implemented that can handle this type of design within SPSS? Any help is greatly
appreciated. Thanks

Rob

 

[Bruce Weaver]

Alternatively, one could use generalized estimating equations (GEE) to account for the correlated nature of the data.  See the GENLIN procedure.  (IIRC, I used both conditional logistic regression and GEE with 1:3 matching, and found that the results were very similar.)  For an introduction to GEE, see the article by James Hanley et al. in Am J Epidemiol 2003;157:364–375.

HTH.

maxjasper wrote

Cox easily handles {case, control} = {1,1}, {1,n}
{m,n} can perhaps be broken into m*{1,n}





Hello

I am analyzing a matched subjects case control study where some strata have
one case with
multiple controls and other strata have multiple cases and multiple
controls. I have read that
COX Regression would work for single case strata but not so well for strata
with more than
one case because the method used on COX Reg for handling tied survival
times, Breslow's
method, will not produce accurate results... My question is, is there some
technique that can
be implemented that can handle this type of design within SPSS? Any help is
greatly
appreciated. Thanks

Rob

[rob]

Thanks for the thoughts on how to best deal with this. Question - maxjasper indicated "Cox easily handles {case, control} = {1,1}, {1,n} {m,n} can perhaps be broken into m*{1,n}". Have you tried this approach? Would this approach require the computation of additonal variables? Sorry for my ignorance but this is a relatively new area for me.

rob

[MaxJasper]

{case, control} = {1,1} and {1,n} can easily be implemented using Cox...However, {m,n} data I break them into m*{1,n} but here is an advice by Marija Norusis:

 

"...you should not analyze data with multiple cases and multiple controls in a set with the Cox Regression procedure, given the method implemented for handling tied survival times within a stratum." [Marija Norusis: Advanced Stat Proc Companion]

 

Max.

 

 

Thanks for the thoughts on how to best deal with this. Question - maxjasper indicated "Cox easily handles {case, control} = {1,1}, {1,n} {m,n} can perhaps be broken into m*{1,n}". Have you tried this approach? Would this approach require the computation of additonal variables? Sorry for my ignorance but this is a relatively new area for me.

rob