Calculating GMR(Geometric Mean ratio) at two different visits

[mounikag]

Hi ,

I have a parameter with lab test values(aval) at two different visits(day1, week4). So i need calculate a Geometric Mean Ratio for  (Week 4 Antibody Level / Day 1 Antibody Level). for a parameter by vaccine group at week 4

Below is the example:

subject      paramcd      avisit             aval       GMR

101            aaa             day1             1.3          ?

101            aaa             week4          2.2

102            aaa             day1             1.1

102            aaa             week4           3.3

103            aaa             day1             2.8

103            aaa             week4           4.7

How to calculate GMR varaible and also  a 95%cI

Please provide your insights with an example . so it would be helpful to me .. Thanks in Advance

[FreelanceReinh]

Hi @mounikag,

Following Example 127.5 Equivalence Testing with Lognormal Data from the PROC TTEST documentation I suggest this -- assuming that your dataset is already sorted by PARAMCD SUBJECT:

/* Create sample data */

data have;
input subject $ paramcd $ avisit $ aval;
cards;
101 aaa day1 1.3
101 aaa week4 2.2
102 aaa day1 1.1
102 aaa week4 3.3
103 aaa day1 2.8
103 aaa week4 4.7
;

/* Get Day 1 and Week 4 values into separate variables */

proc transpose data=have out=trans(drop=_name_);
by paramcd subject;
var aval;
id avisit;
run;

/* Compute the GMR Week 4 / Day 1 for AVAL (for each PARAMCD) */

ods select none;
ods output conflimits=want(keep=paramcd ratio--UpperCLGeomMean
                           rename=(GeomMean=GMR LowerCLGeomMean=LowerCL_GMR UpperCLGeomMean=UpperCL_GMR));
proc ttest data=trans dist=lognormal;
by paramcd;
paired week4*day1;
run;
ods select all;

proc print data=want;
run;

Result:

                                              LowerCL_    UpperCL_
Obs    paramcd       Ratio             GMR      GMR         GMR

 1       aaa      week4 / day1      2.0426      0.8933      4.6704

[mounikag]

Hi reinhard,

Thanks for quick reply.. *Below is the sample code i am using to generate the output to pass the GMR variable value into the below aval value.
I am using proc means procedure dataset to generate 95%CI. Could you please use below code and i need to pass the GMR value of week4/day1
instead of aval value. I will provide the sample output how i need. As you suggested, I am using the proc ttest but i am not getting the answer.
Basically i need to pass GMR value between the visits instead of aval value

* Create sample data */;

data have;
input subject $ paramcd $ avisit $ aval;
cards;
101 aaa day1 1.3
101 aaa week4 2.2
102 aaa day1 1.1
102 aaa week4 3.3
103 aaa day1 2.8
103 aaa week4 4.7
104 aaa day1 3.8 2
104 aaa week4 4.8 2
105 aaa day1 2.8 3
105 aaa week4 5.7 3
106 aaa day1 3.7 3
106 aaa week4 4.8 3
;
run;


proc sort data=have ;
by avisit trt01pn ;
run ;
proc means data=have;
class avisit trt01pn;
var aval;
output out=mean n=n nmiss=nmiss mean=mean lclm=lclm uclm=uclm;
run;

data gma;
set mean;
keep avisit n1 gmc1 lcl1 ucl1;
where _type_=7 and &tgroup.=1.;
n1=n;
gmc1=10**mean;
lcl1=10**lclm;
ucl1=10**uclm;
run;

data gmb;
set mean;
keep avisit n2 gmc2 lcl2 ucl2;
where _type_=7 and &tgroup.=2.;
n2=n;
gmc2 =10**mean;
lcl2=10**lclm;
ucl2=10**uclm;
run;

proc glm data=have PLOTS(MAXPOINTS= 10000);
by avisit ;
class trt01pn;
model aval = trt01pn ;
means trt01pn / t cldiff clm;
ods output cldiffs=dmean ModelAnova=anova;
run;

data rgm ;
set dmean;
where comparison="1 - 2";
keep avisit rgmc rlcl rucl;
rgmc=10**difference;
rlcl=10**lowercl;
rucl=10**uppercl;
run;

proc sort data=gma;
by avisit ;
proc sort data=gmb;
by avisit ;
proc sort data=rgm;
by avisit ;
run;

data final;
merge gma gmb rgm ;
by avisit ;
run;

[FreelanceReinh]

Thanks for providing code and a sample output. I'm not sure I understand what you're trying to do, partly because of inconsistencies in your code:

• Dataset HAVE does not contain a variable TRT01PN (but a variable PARAMCD), so neither the first PROC SORT step nor the PROC MEANS step will work.
• The WHERE conditions in the DATA steps following the PROC MEANS step seem to use a treatment group variable, but the definition of macro variable TGROUP is not shown.
• The WHERE condition _type_=7 will never be met as long as the PROC MEANS step uses only two CLASS variables. Three (or more) CLASS variables would be needed.
• You write "I am using the proc ttest," but it looks rather like you are using PROC GLM.
• The footnote of your sample output states that the confidence intervals are calculated using the Clopper-Pearson method. I'm only familiar with the Clopper-Pearson confidence interval for a binomial proportion, but I don't see where a binomial proportion is involved in your data.
• What do you mean by "pass the GMR variable value into the below aval value"?

[mounikag]

HI ,

1. Sorry i forgot to add treatment variable in the dataset. I added now its in the below dataset and i updated my code as per your comments .
2. I forgot to update tgroup macro to required variable now its updated in the below code.
3. Yes you are right i should use _type_=3 in proc means updated in the below code.
4. what i meant is previously i used proc ttest in separate code but it didnt worked so i changed the code to apply PROC GLM
5. We need to calculate 95% CI using clopper pearson method for particular geometric ratio
6. what i meant is i need to keep the value of geometric mean ratio between day1 / week 4. so i need to calculate GMR across two visists for each parameter. so how i have to calculate to get the value of GMR? Please take the example of 101 subject at day1 the avla value is 1.3 and week 4 value is 2.2 so what will be the GMR value between these two aval values? could you please let me know your inputs with a code

data have;
input subject $ paramcd $ avisit $ aval; trt01pn
cards;
101 aaa day1 1.3 1
101 aaa week4 2.2 1
102 aaa day1 1.1 2
102 aaa week4 3.3 2
103 aaa day1 2.8 1
103 aaa week4 4.7 1
104 aaa day1 3.8 2
104 aaa week4 4.8 2
105 aaa day1 2.8 3
105 aaa week4 5.7 3
106 aaa day1 3.7 3
106 aaa week4 4.8 3
;
run;


proc sort data=have ;
by avisit trt01pn ;
run ;
proc means data=have;
class avisit trt01pn;
var aval; (GMR value ?)
output out=mean n=n nmiss=nmiss mean=mean lclm=lclm uclm=uclm;
run;

data gma;
set mean;
keep avisit n1 gmc1 lcl1 ucl1;
where _type_=3 and trt01pn =1;
n1=n;
gmc1=10**mean;
lcl1=10**lclm;
ucl1=10**uclm;
run;

data gmb;
set mean;
keep avisit n2 gmc2 lcl2 ucl2;
where _type_=3 and trt01pn=2;
n2=n;
gmc2 =10**mean;
lcl2=10**lclm;
ucl2=10**uclm;
run;

proc glm data=have PLOTS(MAXPOINTS= 10000);
by avisit ;
class trt01pn;
model aval = trt01pn ;
means trt01pn / t cldiff clm;
ods output cldiffs=dmean ModelAnova=anova;
run;

data rgm ;
set dmean;
where comparison="1 - 2";
keep avisit rgmc rlcl rucl;
rgmc=10**difference;
rlcl=10**lowercl;
rucl=10**uppercl;
run;

proc sort data=gma;
by avisit ;
proc sort data=gmb;
by avisit ;
proc sort data=rgm;
by avisit ;
run;

data final;
merge gma gmb rgm ;
by avisit ;
run;

[FreelanceReinh · Accepted solution]

Thanks for the improved code and the explanations.

---

@mounikag wrote:
5. We need to calculate 95% CI using clopper pearson method for particular geometric ratio

---

The geometric means are calculated from antibody levels (measured on a continuous scale). As I said, the "Clopper-Pearson method" I know is applicable to binomial proportions (see PROC FREQ documentation), which I don't see in your data or code.

6. what i meant is i need to keep the value of geometric mean ratio between day1 / week 4. so i need to calculate GMR across two visists for each parameter. so how i have to calculate to get the value of GMR? Please take the example of 101 subject at day1 the avla value is 1.3 and week 4 value is 2.2 so what will be the GMR value between these two aval values? could you please let me know your inputs with a code

---

Geometric means (and means in general) are calculated as a summary of two or more values, not for a single value of one subject. So I would expect these to be calculated from the AVAL values of several subjects (as shown in my first reply). The GMR would serve as a summary measure of all the individual ratios (like 2.2/1.3 for subject 101), e.g., within a treatment group.

---

[mounikag]

Thanks .. for clarification. I got your point .. thanks for taking time to answer my questions. It was so helpful