Common Risk Difference

[mb_1]

Need to calculate common risk difference estimate for >=5% reduction in weight adjusting for sex between treatment groups 1,2,3,4 and the placebo group with 95% CI. The risk difference and 95% CI are based on the minimum risk weight approach by Mehrotra and Railkar. Currently I have data in the below format.

Subject	Whether >=5% reduction in weight	Treatment	Sex
1	1	1	1
2	0	2	1
3	1	3	1
4	1	4	0
5	1	1	1
6	0	0	0
7	0	0	0
8	1	4	1
9	0	2	0
10	0	3	0

How should I calculate the common risk difference?

[StatDave]

Use one of the methods discussed and illustrated in this note. 

[mb_1]

Are any of these methods based on the minimum risk weight approach by Mehrotra and Railkar?

[Ksharp]

You gotta see this :

Solved: commonriskdiff using MR method in the case of double zero events - SAS Support Communities

[Ksharp]

/*Here is an example*/
data have;
 set sashelp.heart(obs=1000 keep=bp_status status sex rename=(bp_status=treatment));
run;

proc freq data=have;
table treatment*status*sex/commonriskdiff(column=2 cl=MR );
run;

[mb_1]

In the example code, what does the column=2 option work?

Also, it is giving me the following output.

Does "controlling for treatment" mean treatment is the stratifying factor?

I actually want the output in the following format where I get the common risk difference adjusted for the stratifying factor i.e., sex comparing between each treatment group and placebo.

	Placebo (Normal)	Treatment 1 (High)	Treatment 2 (Optimal)
Estimated Common Risk Difference (95% CI)	-	xx (xx.x, xx.x)	xx (xx.x, xx.x)
p-value	-	0.xxx	0.xxx

[Ksharp]

“In the example code, what does the column=2 option work?”
That means using the second column 's risk to build riskdiff .
Here is risk when sex=Male (a.k.a count of Male When Status='Alive' / count of Status='Alive') <---Risk of Row1

or (a.k.a count of Male When Status='Dead' / count of Status='Dead') . <-----Risk of Row2

RiskDiff= Risk of Row1 - Risk of Row2 .


"Does "controlling for treatment" mean treatment is the stratifying factor?"
Yes. That is correct.


"i.e., sex comparing between each treatment group and placebo."
You need to know which variable(the strata variable) to adjust .
Only include the strata/third variable, you could get COMMON Risk Diff.
and make sure the ROW and COLUMN variables both must have TWO levels.

[mb_1]

The strata variable for the data I have is sex with 2 levels : male and female. There are 7 different treatment groups including placebo. The outcome variable has 2 levels: 0 and 1. Can you help me get the common risk estimates, min risk CI and p-value in the format I mentioned in last comment? 

[Ksharp]

To calculate COMMON Risk Diff , you should have a nx2x2 contingency table.

So you need organize your data to form a nx2x2 table.

NOTE： updated

data have;
 set sashelp.heart(obs=1000 keep=bp_status status sex rename=(bp_status=treatment));
run;

/*
1)Firstly , Get data for Optimal and Normal ,and using "Death"  as a reference level (column=2)
2)Here the first row is Normal and second row is Optimal,so Risk Diff is Normal - Optimal 
*/
proc freq data=have(where=(treatment in ('Optimal' 'Normal')));
table sex*treatment*status/commonriskdiff(column=2 cl=MR test=MR);
ods output CommonPdiff=CommonPdiff1 CommonPdiffTests=CommonPdiffTests1;
run;

/*
1)Secondly , Get data for Optimal and High ,and using "Death" as a reference level (column=2)
2)Here the first row is High and second row is Optimal,so Risk Diff is High - Optimal 
*/
proc freq data=have(where=(treatment in ('Optimal' 'High')));
table sex*treatment*status/commonriskdiff(column=2 cl=MR test=MR);
ods output CommonPdiff=CommonPdiff2 CommonPdiffTests=CommonPdiffTests2;
run;



/*After getting datasets "CommonPdiff1" and "CommonPdiff2" , you could combine them together.*/
data want;
 merge CommonPdiff1 CommonPdiff2(rename=(value=v lowercl=l uppercl=u)) 
       CommonPdiffTests1 CommonPdiffTests2(rename=(Prob=P));
 label='Common Risk Diff(95% CI)';
 Optimal=.;
 Normal=cats(put(value,8.2),'(',put(lowercl,8.2),',',put(uppercl,8.2),')');
 High=cats(put(v,8.2),'(',put(l,8.2),',',put(u,8.2),')');
 output;
 label='Pvalue';
 Optimal=.;
 Normal=vvalue(Prob);
 High=vvalue(P);
 output;
run;

/*Print it*/
proc report data=want nowd;
column label Optimal Normal High;
run;