RE: Number of subject and population PK/PD modeling

From: "Serge Guzy" GUZY@xoma.com Subject: RE: [NMusers] Number of subject and population PK/PD modeling Date: Fri, January 14, 2005 12:40 pm My experience with both Winnonlin and the population approach is that the mixed effect approach always gave me better estimates of at least the population means. Using Winnonlin and averaging the PK estimates never gave me superior average values of the main PK parameters. On the other hand, sometimes I saw problems with the estimates of the covariance components of the population variance covariance matrix when dealing with small number of patients and in a rich data environment. Correlation sometimes would be drifted to 1 with the log-likelihood being flat across a big range of correlation values. In that case, rich data analysis using Winnonlin would give me better estimates of the true correlation between the PK parameters. My experience was also that population variances were estimated equally in a rich data environment and better of course in a semi rich data environment (some patients did not have enough information to be analyzed with Winnonlin). My conclusion was that a mixed effect approach is always as good as and most of the time better than a Winnonlin approach expect for very rich data environment where we are interested to estimate the population covariance. Serge Guzy President POP_PHARM