Mantel-Haenszel difference in proportions controlling for strata

[lcmichael_unc]

Hi Folks,

I want to compute the overall stratum-adjusted difference in proportions (SVR12*TRTWKS) and confidence limits for the variable SVR12 in the following tables statement in a frequency procedure. I need the overall difference (and C.I.) in SVR12, controlling for CIRRHOSIS (Y/N) and HCVGTCAT (A/B) simultaneously but the following code is producing individual differences controlling independently for each pair of strata (e.g., CIRRHOSIS='Y', HCVGTCAT='A') and not controlling for all levels of both covariates simulateously. Many thanks for all insights.

proc freq data=&library..SVR_GP_mITT nlevels order=data;
tables CIRRHOSIS*HCVGTCAT*TRTWKS*SVR12 / riskdiff (column=both CL=(WALD WILSON)) chisq fisher cmh cl agree alpha=0.05;
format SVR12 svrfmt.;
where FIRSTREC=1;
ods output CrossTabFreqs=crstabfrq RiskDiffCol1=rskdfcol1 PearsonChiSq=pearchisq PdiffCLs=PdiffCLs ;
run;

[lcmichael_unc]

All responders:

I think I found my answer: it requires the specification of the 'common' sub-option in the riskdiff option per:

tables CIRRHOSIS*HCVGTCAT*TRTWKS*SVR12 / tables HCVGTCAT*TRTWKS*SVR12 / 
riskdiff (common column=both CL=(WALD WILSON)) chisq fisher cmh cl agree alpha=0.05;

Please confirm. Thanks.

[Watts]

Yes, the RISKDIFF(COMMON) option provides confidence limits for the overall stratum-adjusted risk (proportion) difference. Also, the COMMONRISKDIFF option provides additional confidence limit types, tests, and stratum weights.