RE: posthoc step

From: "Ludden, Thomas (MYD)" luddent@iconus.com Subject: RE: [NMusers] posthoc step Date: Mon, December 20, 2004 11:27 am Nick et al. NONMEM uses the same algorithm for all ETA searches. However, the details of the algorithm differ between version V and version VI beta. When MAXEVAL is not zero, the ETA search may benefit from different THETA, OMEGA and SIGMA estimates as the overall search continues. As already noted, rather small differences in the priors may alter the OFV shape. For instance, using the same data (T=1,2,3,Y) as the example but somewhat different priors, K~N(9.49,var=4.67), err~N(0,var=0.0371) instead of K~N(10,var=4.), err~N(0,var=0.04), the Bayes OFV yields a shoulder over the range of K=8 to K=9 but not a minimum. With this prior NONMEM V returns a K value of 0.943 for the individual. As noted before, if one used a design appropriate for the prior such that the data were (0.1*T,Y), NONMEM V estimates K=1.01. In this case, the predicted Y values, based on the prior, are above the err-SD. In fact, supplementing the previous data set with one additional observation at T=0.1 provides a K estimate of about 1. Best wishes to all for the holidays. Tom _______________________________________________________