Posthocs

From: Baerbel Wittke 41 61 68 85067 <BAERBEL.WITTKE@roche.com> Subject: Posthocs Date: 16 Jul 1997 12:44:20 -0400 Hello, I have a data set from a phase III study containing steady state plasma concentrations of a compound after an iv infusion for about 50 % of the patients. Dosing and covariable information is available for all patients. I could obtain good posthoc estimates for drug clearance for all patients with plasma levels. My goal is to estimate the steady state plasma levels of the remaining patients (without plasma concentration measurements). When I repeated the posthoc analysis with the complete data set (DV = 0 for the patients without measurements), the posthoc estimates of CL were much too high for these patients. I got reasonable results for the posthocs when I used the predicted plasma levels (based on TVCL) instead of DV = 0 in the data set, but I don't know whether this is the correct way to obtain posthoc estimates for patients without concentration measurements. Has anybody experience with this ?? Thanks, Baerbel Wittke (e-mail: baerbel.wittke@roche.com)

From: R.Port@dkfz-heidelberg.de (Ruediger Port) Subject: POSTHOC estimates Date: 16 Jul 1997 13:31:14 -0400 This is in reply to Baerbel Wittke's question: I think you should use data records where EVID=2 and MDV=1 for those patients where no concentration measurements are available. This will produce predicted plasma levels (under PRED) where the prediction is based on TVCL, the available covariate information, and the (fixed effects) parameters which link individual CL to the covariates. DV=0 without MDV=1 is interpreted as a real zero measurement, hence the exceedingly high POSTHOC estimates of individual CL. As I understand, no real POSTHOC estimates of individual clearance can be obtained without concentration measurements because all eta's are estimated to be zero in this case (the most likely value for eta when there is no concentration measurement that could suggest that this individual differs by something other than the covariates from a typical individual). Good luck! Ruedi Dr. R.E. Port, Dept. 0420, German Cancer Research Center P.O. Box 10 19 49, D-69009 Heidelberg phone: x49-6221 42-3385 -3347 fax: -3346 e-mail: r.port@dkfz-heidelberg.de