Bioequivalence and Bioavailability Forum • SAS: Dropping a treatment from a 3-way study

jag009
★★★

NJ,
2013-08-01 23:08
(4280 d 20:16 ago)

Posting: # 11154
Views: 11,591

SAS: Dropping a treatment from a 3-way study [Software]

Hi all, especially Detlew :-)

Do I need to do anything out of the ordinary in SAS to analyze a 3 treatment study dataset having one of the treatment arms removed?

I use PROC GLM to analyze a BE dataset from a 3-period, 3-treatment, 6 sequence crossover study. For some reason when I removed the data from one treatment arm (I just put "." for the data, leaving the "subject, period, treatment, sequence" info intact), SAS failed to generate results for T/R Ratio, 90% CI for the treatment that I kept.

SAS errror message (with respect to PROC GLM):
NOTE: TYPE I EMS not available without the E1 option.
WARNING: More coefficients than levels specified for effect TREATMENT. Some coefficients will be ignored.
NOTE: Test2 - Reference is not estimable.
NOTE: The data set WORK.L_S_MEANS has 5 observations and 9 variables.

WARNING: Output 'Estimates' was not created. Make sure that the output object name, label, or path is spelled correctly. Also, verify that the appropriate procedure options are used to produce the requested output object. For example, verify that the NOPRINT option is not used.

My SAS code:

PROC GLM data=&glmdata order=internal outstat=tanova; 

  CLASS TREATMENT PERIOD SEQUENCE SUBJECT;

  MODEL &pk= SEQUENCE SUBJECT(SEQUENCE) PERIOD TREATMENT/SS1 SS3 CLPARM alpha=0.1;

  output out=pred p=yhat r=resid student=stresid;

  RANDOM SUBJECT(SEQUENCE) /TEST;

  LSMEANS TREATMENT PERIOD /stderr pdiff out=lsmeanl;

  estimat = ESTIMATE "Test2 - Reference" TREATMENT 0 1 -1;

run;

Thanks
John

Helmut
★★★

Vienna, Austria,
2013-08-02 00:17
(4280 d 19:07 ago)

@ jag009
Posting: # 11155
Views: 10,003

Delete instead of missing?

Post reply

Hi John,

I don’t speak SAS, but until Detlew shows up you can try something: Instead of setting Test1’s observations to “.”, delete the respective rows completely.
Try in the last line to delete 0 from the ESTIMATE statement.

Edit: PHX seems to be more tolerant. Same results whether I exclude one treatment (my standard), set them to “.”, or delete complete rows. In the second case it doesn’t matter whether I set one of the coefficients to 0 or leave it empty. Now for the interesting part: One of my studies, n=18, 6×3, all effects fixed.

Full model (df 30)

       PE      90% CI    ISCV   MSE

T1/R  98.91 94.73 103.27 7.42 0.005485

T2/R 101.59 97.29 106.07 7.42 0.005485

Crippled models (df 14 each)

       PE      90% CI    ISCV   MSE

T1/R  99.00 93.64 104.66 9.19 0.008413

T2/R 101.52 98.15 105.02 5.53 0.003108

CIs in the crippled models are wider or narrower due to the different pooled variances and dfs. The PEs are also slightly different (closer to 1), which was a little bit surprising to me.
:confused:

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

d_labes
★★★

Berlin, Germany,
2013-08-02 11:06
(4280 d 08:18 ago)

@ jag009
Posting: # 11165
Views: 9,909

SAS: Dropping a treatment from a 3-way study

Post reply

Hi John!

Full ACK with Helmut's suggestion. Seems he has basic SAS speak knowledge acquired :cool:

.
If you set the values of one treatment just to missing SAS nevertheless tries to estimate the treatment effect for it with is of course missing. Result: Errors.

Also full ACK with the estimate statement. After removing one treatment it has only two entries.

BTW: Be careful with this estimate. The factors +1, -1 and zero depend on the coding of our treatment factor. It's lexicographic. T1, T2 come after R if you have coded as usual T1, T2, R.

—
Regards,

Detlew

Helmut
★★★

Vienna, Austria,
2013-08-02 19:23
(4280 d 00:00 ago)

@ d_labes
Posting: # 11173
Views: 9,824

SAS: Dropping a treatment from a 3-way study

Post reply

Hi Detlew!

❝ Full ACK with Helmut's suggestion. Seems he has basic SAS speak knowledge acquired :cool: .

Just following simple analogies of the type: “Since I’m fluent in BASIC, I can dabble in FORTRAN.” ;-)

Any opinion on the different PEs I observed in my example?

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна!
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

jag009 ★★★ NJ, 2013-08-02 19:38 (4279 d 23:46 ago) @ d_labes Posting: # 11175 Views: 9,813	SAS: Dropping a treatment from a 3-way study Post reply
	Thanks guys! I should've thought of that. John

BRN ☆ Turkey, 2015-07-15 19:00 (3568 d 00:24 ago) @ d_labes Posting: # 15089 Views: 8,513	SAS: Dropping a treatment from a 3-way study Post reply
	Hi Detlew, ❝ BTW: Be careful with this `estimate`. The factors +1, -1 and zero depend on the coding of our treatment factor. It's lexicographic. T1, T2 come after R if you have coded as usual T1, T2, R. How should we write the sas code for the estimate to compare - T vs. R - T vs. R1 in a 4 sequence study (R1TR2R R2RR1T RR1TR2 TR2RR1)? Thanks. — Regards, BRN

d_labes
★★★

Berlin, Germany,
2015-07-17 10:37
(3566 d 08:47 ago)

@ BRN
Posting: # 15101
Views: 8,440

SAS: estimate statement in Proc GLM

Post reply

Dear BRN;

❝ How should we write the sas code for the estimate to compare

❝ - T vs. R

❝ - T vs. R1

❝

❝ in a 4 sequence study (R1TR2R R2RR1T RR1TR2 TR2RR1)?

That depends (as always in life :cool:

).
If you evaluate all data at once it reads (lex order R, R1, R2, T):

estimate 'T vs. R' tmt -1 0 0 1;

estimate 'T vs. R1' tmt 0 -1 0 1;

If you evaluate according to the EMA crippled analysis you have to use only the data for the comparison under consideration, i.e. you have only T|R or T|R1 at hand. Then the estimate statement reads:
estimate 'T vs. R' tmt -1 1;
and in an other Proc GLM call
estimate 'T vs. R1' tmt -1 1;

Hope that helps.
BTW: Sometimes RTFM helps also :-D

—
Regards,

Detlew