residual error model, endogeneous drug

From: "R. E. Port" <R.Port@DKFZ-Heidelberg.de> Subject: residual error model, endogeneous drug Date: Tue, 22 May 2001 12:25:42 +0200 (CEST) Dear Paul, a few weeks ago you posed a question to the group about how to model an endogenous baseline concentration (CBAS) of a compound which is also administered exogenously; specifically, the question was how to model the interindividual variation of this baseline concentration: > My questions to the group: > - is there a method of coding such that F but not CBAS is allowed to have > an EPS assigned to it > - would this remove the bias in the estimation of CBAS, > - can one still get correct values for W and IWRES? Lewis, in his reply, mentioned a way of modeling which ... > involves conditioning on the baseline observation, but recognizing that it > has error of the same magnitude as any other. My own experience with modeling an endogenous baseline value ("BASE") has been like yours, namely that describing BASE using a theta and some kind of random variation didn't satisfying results, especially in the presence of large interindividual variation in BASE. On the other hand, I get nice results (with $PRED, I haven't tried $PK) with a procedure where no mean baseline (theta) is estimated at all and where "all the variability of the baseline (is) "absorbed" by the eta ... recognizing that it has error of the same magnitude as any other (observation)" (Lewis). The code is as follows (assuming that your baseline measurements were taken at time zero and that there are no data records at times < 0): IF (TIME.EQ.0) BASEfee = DV ; fixed-effects estimate of baseline BASEmee = BASEfee + THETA(1)*ETA(1) ; mixed-effects estimate of baseline ... ; mixed-effects prediction for what was measured at later times: mep = BASEmee + increment ; "increment": effect of treatment Y = mep + THETA(1)*EPS(1) $OMEGA 1 FIXED $SIGMA 1 FIXED This code sets PRED = DV for time zero. Setting MDV=1 for this data record prevents it from having an EPS assigned to it, such as you wanted. This lets NMTRAN issue a warning message which, I hope, doesn't matter in this case. ETA(1) and EPS(1) have the same standard deviation (THETA(1)) recognizing that the baseline measurement has the same residual error as all the others. The procedure has been mentioned in a publication (Port et alii, Br J Clin Pharmacol 46, 461 - 466, 1998) but may not have been thoroughly examined by a statistician, so if someone finds it flawed for some statistical reason I'ld be grateful to learn about. With best regards, Ruedi -------------------------------------------------------------------------- Dr. R.E. Port, German Cancer Research Center, D-0200 P.O. Box 10 19 49, D-69009 Heidelberg, Germany phone: +49-6221 42-3385 -3347 fax: -3382 e-mail: r.port@dkfz.de