Re: [Fwd: CLIN PHAR STAT: Mixed Vs Fixed]

Date: Wed, 09 Aug 2000 12:41:14 +1200 From: Nick Holford <n.holford@auckland.ac.nz> Subject: Re: [Fwd: CLIN PHAR STAT: Mixed Vs Fixed] Steve, Stephen Duffull wrote: > Here is a simple example of the prior setup for the WinBUGS > code for implementing a Bayesian PK analysis. This is the nmusers list. I have deleted that WinBuggy filth :-) > The difficulty I have lies in not knowing exactly what > NONMEM does when it does the NONMEM Bayesian approach > (please note that I have never used this feature of NONMEM > and indeed don't know how too). Here is a "simple" example of how to use NONMEM with priors on THETA and OMEGA. It requires a user defined prior subroutine which follows the control stream: $PROB theophylline pharmacodynamics $DATA theopd.dat IGNORE # $INPUT ID TIME THEO AGE WT GEND RACE DIAG PEFR=DV $ESTIM PRINT=1 POSTHOC MSFO=theopd.msf $COV $THETA (0,146.,) ; E0 $THETA (0,164.,) ; EMAX $THETA (.001,6.55,) ; EC50 $OMEGA 0.00178 ; CVE0 $OMEGA 0.239 ; CVEMAX $OMEGA 1.46 ; CVEC50 $SIGMA 6590. ; SD 1 $SUB PRIOR=theopdB.for ;Prior normal mode for THETA $THETA 146. FIX ; theta1 $THETA 164. FIX ; theta2 $THETA 6.55 FIX ; theta3 ;Prior uncertainty on normal mode for THETA $OMEGA BLOCK(3) FIX 122 -125 427 6.05 20.1 3.82 ;Prior OMEGA $OMEGA 0.00178 FIX ; omega1 $OMEGA 0.239 FIX ; omega2 $OMEGA 1.46 FIX ; omega3 ;Degrees of Freedom for OMEGA (Nsubjects-1) $THETA 152 FIX ; dfomega1 $THETA 152 FIX ; dfomega2 $THETA 152 FIX ; dfomega3 $PRED e0= THETA(1)*EXP(ETA(1)) EMAX=THETA(2)*EXP(ETA(2)) EC50=THETA(3)*EXP(ETA(3)) Y = E0 + EMAX*THEO/(THEO+EC50) + ERR(1) $TABLE ID TIME THEO AGE WT GEND RACE DIAG E0 EMAX EC50 Y NOPRINT ONEHEADER FILE=theopd.fit theopdB.for SUBROUTINE PRIOR (I,CNT,NTHP,NETP,NEPP) DOUBLE PRECISION CNT IF (ICALL.LE.1) THEN NTHP=3 NETP=3 NEPP=1 ENDIF CALL NWPRI(CNT) RETURN END > In contrast WinBUGS is > quite transparent about both its hierarchical structure and > about methodology it uses to get around nasty integration > problems. Its all clear as mud. > IMHO I think that if a Bayesian method is desirable then a > Bayesian program should be used - but I may be biased :-). Not biased -- just different set of priors. -- Nick Holford, Division of Pharmacology & Clinical Pharmacology University of Auckland, Private Bag 92019, 85 Park Road, Auckland, NZ email: n.holford@auckland.ac.nz tel:+64(9)373-7599x6730 fax:373-7556 http://www.phm.auckland.ac.nz/Staff/NHolford/nholford.htm