repeating cases

Date: Wed, 17 Nov 1999 11:25:40 -0600 From: Dr Ted Whittem <whittem@uiuc.edu> Subject: repeating cases I would be grateful for direction to reading on a method to handle within NONMEM those individual cases that re-enter a study multiple times. My particular problem involves kinetics of an antibiotic for mastitis in cows. An animal may be successfully treated on one occasion, but re-enter the study several weeks/months later with a new infection (different pathogen, changed covariables etc). Obviously these are not truly independent... I am a new user and have not yet discovered answers to what are probably FAQs! Ted Whittem

Date: Wed, 17 Nov 1999 13:57:18 -0800 From: Lewis Sheiner <lewis@c255.ucsf.edu> Subject: Re: repeating cases The basic idea is to treat the individual as the "same" on each occasion he/she/it is encountered. This means, practically, having all records for an individual be contiguous in the data file, all with the same ID. You can restart the time scale by using a record with EVID = 3 0r 4. Different covariates do not present a problem as the $PK block recomputes the parameters at each event record. If, however, variability between occasions is envisaged, so that the person/animal is not the same between occasions, although perhaps more similar to itself on another occasion than it is to a random other individual (with the same covariate values), then you may wish to allow for both inter-individual variability (IIV) and inter-occasion (IOV) variability. You can read about how one might do this (a bit clumsily) using NONMEM in Karlsson MO, Sheiner LB: The importance of modeling inter-occasion variability in population pharmacokinetic analyses, J Pharmacokin Biopharmaceut, 21: 735-750, 1993. Good luck, LBS. PS. Note that, as happens more often, at least with human data, than one might have thought, the IOV>IIV, then treating each occaasion as though it were a distinct individual is a reasonable approximation.

Date: Thu, 18 Nov 1999 11:32:01 +1300 From: Nick Holford <n.holford@auckland.ac.nz> Subject: Re: repeating cases Looks like an ideal case for using a Between Occasion Variability (BOV) model. These have been described by Karlsson & Sheiner as Inter Occasion Variability (IOV) but I prefer BOV. You can use the different covariates for the animal to predict *systematic* changes in CL from study to study but the BOV describes the *random* variability about the mean (covariate adjusted) CL for that individual. The code for describing Between Subject Variability (BSV) and BOV for CL could be like the following. OCC is a data item that indicates the occasion of the study e.g OCC=1 would be all concs and doses on study occasion 1 and OCC=2 would be all concs and doses on study occasion 2, etc. The $OMEGA BLOCK(1) SAME record allows the ETA to be different for each occasion for an in individual but ensures that all ETAs are sampled from a distribution with the same SD. It is not possible to identify BOV without this constraint. $INPUT TIME OCC AMT DV $THETA 1 ; CL $OMEGA BLOCK(1) .25 ;Occ 1 $OMEGA BLOCK(1) SAME ;Occ 2 ; Repeat the $OMEGA BLOCK(1) SAME record for each occasion after the first $OMEGA .5 ; BSVCL $PK IF (OCC.EQ.1) THEN BOV=ETA(1) ENDIF IF (OCC.EQ.2) THEN BOV=ETA(2) ENDIF ; Repeat above IF/ENDIF block for each occasion BSV=ETA(1) CL=THETA(1)*EXP(BSV+BOV) Karlsson, M. O. Sheiner, L. B. The importance of modeling interoccasion variability in population pharmacokinetic analyses. J Pharmacokin Biopharm 1993; 21:735-50 -- Nick Holford, Dept Pharmacology & Clinical Pharmacology University of Auckland, Private Bag 92019, Auckland, New Zealand email:n.holford@auckland.ac.nz tel:+64(9)373-7599x6730 fax:373-7556 http://www.phm.auckland.ac.nz/Staff/NHolford/nholford.htm