Bioequivalence and Bioavailability Forum

Hi,

I have a 3 period, 2 treatment crossover (TRR, RTT) design and used the following code (as per FDA guidance) to analyse the data:

PROC MIXED DATA= cmax ;
  CLASSES armcd usubjid period treat ;
  MODEL LOG_PARM = armcd period treat / DDFM=KR;
  RANDOM treat/TYPE=FA0(2) SUB=usubjid G;
  REPEATED/GRP=treat SUB=usubjid;
  ESTIMATE 'B vs A' treat -1 1/CL ALPHA=0.1;
  LSMEANS treat/ cl alpha=0.1;
RUN;

The above analysis resulted in the following note in the SAS log:
"Estimated G matrix not positive definite"
suggesting that "one or more variance components on the RANDOM statement is/are estimated to be zero and could/should be removed from the model."

My colleague is suggesting I switch the covariance structure to VC however I'm concerned that this negates the REPEATED statement and is questionable in a 3x2 design. What is your opinion on this point?

I used the CSH structure but, as expected, encountered the same notes in the log.

Looking through the covariance structures, Compound Symmetry (CS) looks more appropriate as it accounts for correlation & assumes it is constant regardless of the lag between pairs of repeated measurements (sounds OK given the study design).

Any opinions on the above points gratefully received.
Thanks,
RB

---

Edit: Formatted. You may use BBCodes (see here). [Helmut]