RE: eigenvalues

Pavel: In NONMEM, the best variance-covariance is obtained during evaluation of the objective function by the importance sampling step (expectation only, EONLY=1, so the final parameters do not change from what was assessed by SAEM). The SAEM step only evaluates the S matrix version of the variance-covariance. To obtain the R matrix version, set $COV MATRIX=R, and evaluate the importance sampling step, from which you get the proper marginal density objective function, as well as a full Monte-Carlo assessed variance-covariance matrix. For example: $EST METHOD=SAEM …. $EST METHOD=IMP NITER=10 EONLY=1 MAPITER=0 ISAMPLE=1000 … $COV MATRIX=R PRINT=E UNCONDITIONAL You should assess the eigenvalues from the variance-covariance evaluated at the IMP step. Robert J. Bauer, Ph.D. Vice President, Pharmacometrics R&D ICON Early Phase Office: (215) 616-6428 Mobile: (925) 286-0769 [email protected]<mailto:[email protected]> http://www.iconplc.com