crossover with lognormal endpoint

favola — Tue, 28 Feb 2012 21:14:01 GMT

I have to calculate the sample size of a 2x2 crossover study.

The primary efficacy endpoint is a continuous variable (FeNO) with a lognormal distribution, measured before (pre) and after (post) treatment.

<1> Can I use proc power with paired t tests for Mean Ratio with Lognormal Data or is there a better way?

proc power;

pairedmeans test=ratio DIST=logNormal

meanratio =

cv =

corr =

npairs = .

alpha = 0.05 power = 0.9;

run;

I have data from a pilot study very similar. The following procedure was used to analyze the efficacy endpoint.

proc glm data=pilot;

class seq drug period pat;

model post_log = seq pat(seq) drug period pre_log/ ss3; *pre_log and post_log = before and after treatment suitably log trasformed.

test h=seq e=pat(seq);

lsmeans drug /pdiff stderr;

run;

<2> It is correct?

<3> There is a proc glmpower corresponding to this ANCOVA analysis?

To obtain the data for my proc power instead I should perform the following proc t test:

proc ttest data=pilot dist=lognormal;

var post1 post2 / crossover= (drug1 drug2);

run;

<4> meanratio = is geometric mean of treatment ratio with both sequence ?

<5> cv = is geometric CV of treatment ratio with both sequence?

<6> corr = how can I calculate it?

If I submit the following procedure:

proc ttest data=pilot dist=lognormal;

var post_ratio_drug1_drug2;

run;

I get the same geometric mean but another CV.

<7> Which CV should I use?

But using only the post-treatment values I have to assume that values before treatment are homogeneous between treatment groups.

<8> To avoid this how can I do?

<9> I can analyze the post/pre ratio?

<10> If it is distributed normally have to use a proc power with DIST = Normal?

I hope not too 10 questions.

Thanks in advance (Please use plain English)

Favola

crossover with lognormal endpoint

favola — Mon, 12 Mar 2012 13:20:11 GMT

<7> Let's use an example.

data pilot;

input Drug1 $ Drug2 $ post1 post2 @@;

label Drug1='Drug Period 1' Drug2='Drug Period 2' post1='FeNO period 1' post2='FeNO period 2';

datalines;

R T 24 15

T R 15 27

R T 28 23

T R 30 28

T R 6 12

R T 32 22

T R 23 19

R T 32 17

T R 76 41

R T 15 10

T R 20 25

R T 14 15

T R 21 18

R T 66 25

T R 25 42

R T 38 35

R T 20 19

T R 21 19

R T 27 51

T R 24 72

R T 14 14

R T 22 24

T R 22 24

R T 28 36

T R 22 37

T R 27 51

R T 20 22

T R 28 73

R T 44 29

T R 52 23

T R 18 19

R T 29 34

T R 33 45

R T 63 42

T R 27 58

R T 29 30

R T 19 19

T R 29 32

R T 35 22

T R 26 52

;

run;

proc ttest data=pilot dist=lognormal;

var post1 post2 / crossover= (drug1 drug2);

run;

/* Geometric mean for treatment ratio = 1.2347 and CV = 0.2224 */

/* Calculate treatment ratio */

data pilot1; set pilot;

label post_trt1='FeNO T' post_trt2='FeNO R' post_ratio_drug1_drug2='FeNO R/T';

if drug1='T' then post_trt1=post1; else post_trt1=post2;

if drug1='R' then post_trt2=post1; else post_trt2=post2;

post_ratio_drug1_drug2=post_trt2/post_trt1;

run;

proc ttest data=pilot1 dist=lognormal;

var post_ratio_drug1_drug2;

run;

/* Geometric mean = 1.2347 and CV = 0.4590 */

The geometric mean is the same but CV is different.

Which CV should I use in proc power?

proc power;

pairedmeans test=ratio DIST=logNormal

meanratio =

CV =

corr =

npairs = .

alpha = 0.05 power = 0.9;

run;

Thanks in advance

Favola

topic crossover with lognormal endpoint in SAS Health and Life Sciences

crossover with lognormal endpoint

crossover with lognormal endpoint