We’ve been trying to fit what we believe is extravascular 2 cmpt tlag model. We can obtain reasonable estimates using ELS but the bootstrap performs poorly. Cannot get estimates from LB - the model converges but cannot get standard errors but maybe because it will not invert the Hessian matrix or because the estimates were too close to the upper bound. Tried to work with QRPEM but nonphysiologic estimates, perhaps my inputs were incorrect. Would appreciate advice on any obvious model/error misspecifications for why we cannot use FOCE LB, the limitations of settling with ELS assuming I can improve that. The sampling values we have themselves may contribute. Thanks. [file name=Velicade_7_23_13_shell.phxproj size=6178390]Certara | Drug Development Solutions (5.89 MB)
I refit your model to the data from scratch. Here is what I did. 1: Your data had true DV =0 which is not relevant and can affect the fit especially if you are using proportional error. I remade your data set and removed all DV=0 2: I started with 1c, then added tlag. Tlag was not stat significant. 3: I shift to 2c , first accepted all fixed and random, use as initial V2 and CL2 what I got as final for V and Cl and then fit. I used proportional error which is the most relevant for PK data 4: The fit was good and then I made a copy, accept all fixed and random and did 20 bootstraps. The results to me look good. estimate se % cv tvKa 0.013924391 0.0029675749 21.312062 tvV 1222.8705 426.4322 34.871412 tvV2 44876.499 17007.217 37.897824 tvCl 9.1950585 2.3373909 25.420076 tvCl2 9.330355 1.1491501 12.316253 stdev0 0.29258593 0.039859464 13.623165 I cannot get directly the se’s as apparently the Hessian cannot be inverted. The project is attached. I hope it helps. best serge [file name=velcade_serge.phxproj size=3466181]Certara | Drug Development Solutions (3.31 MB)