From: "Piotrovskij, Vladimir [JanBe]" <VPIOTROV@janbe.jnj.com>
Subject: One more way to approach the problem of interindividually varying SIGMA
Date: Thu, 10 Aug 2000 11:15:47 +0200

Dear nmusers,

In a recent discussion (see related thread) Mats suggested the following to account for the interindividual variability in residual variance:

Y=F+EPS(1)*EXP(ETA(1))

This requires METHOD=1 INTERACTION and is therefore computationally very intensive.

A simplified approach that works fine using the FO method consists in assuming a population being a mixture of two or more subpopulations differing in SIGMA. Thus, no individual SIGMA, but a few subpopulation-related SIGMAs:

$PK
...
EST = MIXEST
SP1 = 0
SP2 = 0
IF (MIXNUM.EQ.1) THEN
SP1 = 1
ELSE
SP2 = 1
ENDIF
$MIX
NSPOP= 2
P(1) = THETA(.)
P(2) = 1 - P(1)
$ERROR ; proportional
Y = F*(1 + SP1*ERR(1) + SP2*ERR(2))

Best regards,
Vladimir
----------------------------------------------------------------------
Vladimir Piotrovsky, Ph.D.
Janssen Research Foundation
Clinical Pharmacokinetics (ext. 5463)
B-2340 Beerse
Belgium
Email: vpiotrov@janbe.jnj.com