Learning SAS? Welcome to the exclusive online community for all SAS learners.

Cochran-Mantel-Haenszel Odds Ratios

Reply
New Contributor
Posts: 3

Cochran-Mantel-Haenszel Odds Ratios

I am completing a meta-analysis for my graduate thesis in Epidemiology and am attempting to find a CMH odds ratio for the relationship between Schizophrenia and 3q29 deletions.  I have written the code as:

 

proc freq data=schizophrenia3q292;
    tables study*case*deldupcat / relrisk CMH;
    title "3q29 Deletions and Duplications";
run;

 

with data in the form of:

 

studycat $ case x n

 

with case=(0=control, 1=case), x=number of individuals with deletion and n=total number of individuals (cases or controls).

 

I followed that up with:

 

data schizophrenia1p362(drop=i);
    length studycat $20;
    set schizophrenia1p36;
            do i=1 to n;
            if i<= x then deldupcat=1;
            if i > x then deldupcat=0;
        status=2-deldupcat;
        output;
        end;
run;

 

in order to create an observation for each individual (a pseudo-"studyid") in order to avoid using a weight.

 

While this is only one of several CNV's I am studying, I know, from a prior meta-analysis that the unadjusted odds ratio should be around 54 (mine is 55) and the CMH odds ratio should be around 41 (mine is 284!).  What could be causing my adjusted OR to be SO high?  Is this normal?  The previous meta-analysis used R instead of SAS.  But, this shouldn't cause such a discrepancy, should it?

Super User
Posts: 11,343

Re: Cochran-Mantel-Haenszel Odds Ratios

Posted in reply to bmacdonald627

Without data it is hard to say specifically why any result is obtained. Comparison with another product or procedure and differences of result, assuming same data is used, often comes down to either different algorithms used or options that may be optional in one but defaut in the other. Again without seeing both code, and knowing both languages (I don't speak R for instance) it would be hard to say.

New Contributor
Posts: 3

Re: Cochran-Mantel-Haenszel Odds Ratios

[ Edited ]

Here is the data:

The first column is just the observation numbers assigned by SAS.  The second column is the study.  The fourth column identifies it as a deletion or duplication (I am just looking at deletions at the moment).  The fourth column identifies where it was a case (case=1) or control (case=0).  "X", or the fifth column, identifies how many cases or controls possessed this deletion, and the final column has the total number of cases or controls in the study.

 

Obs study location deldup case x n

113q29duplicat13420
213q29duplicat02416
343q29deletion123391
443q29deletion003181
543q29duplicat103391
643q29duplicat013181
793q29deletion11172
893q29deletion00160
9123q29deletion11245
10123q29deletion00490
11153q29deletion146882
12153q29deletion006316
13173q29deletion101438
14173q29deletion0133246
15193q29deletion164719
16193q29deletion005917
17193q29duplicat114719
18193q29duplicat005917
19203q29duplicat1108290
20203q29duplicat007431
21213q29deletion11150
22213q29deletion00268
23233q29deletion153945
24233q29deletion003611
25263q29deletion11834
26263q29deletion00672
27273q29deletion121259
28273q29deletion001173

 

Hopefully this can help!

New Contributor
Posts: 3

Re: Cochran-Mantel-Haenszel Odds Ratios

Posted in reply to bmacdonald627

This is the outcome for the raw, unadjusted odds ratio:

Screen Shot 2016-10-19 at 6.58.29 AM.png

 

This is the outcome for the mantel-haenszel odds ratio:

Screen Shot 2016-10-19 at 6.57.19 AM.png

Ask a Question
Discussion stats
  • 3 replies
  • 286 views
  • 0 likes
  • 2 in conversation