All, It feels like there might be some clarification in order. What I was suggesting in the below was a plot of the individual random effect values vs. pregnancy term, in order to explore possible (non-random) covariate effects of pregnancy stage. I think this is consistent with what Nick, Stephen and others are suggesting. It may eventually be appropriate to include changes in IIV variance vs. pregnancy term, but I would be looking for different fixed effects first, and would be fairly judicious in adding additional random effects. Best, Kevin
-----Original Message----- From: owner-nmus...@globomaxnm.com [mailto:owner-nmus...@globomaxnm.com] On Behalf Of Kevin Dykstra Sent: Tuesday, March 01, 2011 11:59 AM To: yu...@pitt.edu; 'nm nm' Subject: RE: [NMusers] [Fwd: occasions during pregnancy] Paul, You might try plotting your etas 6-9 vs. trimester (coded at four levels) to ensure that the IOV is truly random by occasion, as your model assumes. Obviously, it is not unheard of that the IOV should be much larger than IIV, but I wouldn't start with that assumption. Usually there is at least some correlation within an individual. Good luck. Kevin Kevin Dykstra, PhD, FCP +1 978.655.1943 (O) +1 978.289.2987 (M) kevin.dyks...@qpharmetra.com | http://qPharmetra.com -----Original Message----- From: owner-nmus...@globomaxnm.com [mailto:owner-nmus...@globomaxnm.com] On Behalf Of yu...@pitt.edu Sent: Tuesday, March 01, 2011 10:53 AM To: nm nm Subject: [NMusers] [Fwd: occasions during pregnancy] ---------------------------- Original Message ---------------------------- Subject: occasions during pregnancy From: yu...@pitt.edu Date: Tue, March 1, 2011 10:49 am To: "nm nm" <nmus...@blobomaxnm.com> -------------------------------------------------------------------------- Hi all nmusers, I thank all who responded my questions yesterday. Almost all the responses suggested that several occasions of one patient should be put under one ID #. I re-code my control stream and adjusted the data file as following: $PK K12 = THETA(1)*EXP(ETA(1)) CL= THETA(2)*EXP(ETA(2)+ETA(6)*TRI1+ETA(7)*TRI2+ETA(8)*TRI3+ETA(9)*TRI4) $OMEGA .8; .1 .8; .1 .1 .8; .1 .1 .1 .8; .1 .1 .1 .1 .8; $OMEGA BLOCK(1) 0.9; $OMEGA BLOCK(1) SAME; $OMEGA BLOCK(1) SAME; $OMEGA BLOCK(1) SAME; where TRI1,TRI2,TRI3, and TRI4 are different stages of pregnancy. This model fits poorly for the data (from the plot of PRED, IPRED VS. DV), although the estimates are stable and reasonable. If I treat the different occasions as different patients, ignoring the correlation within the same patients, then the model fits quite well and the results are reasonable. I also noticed one note from Lewis Sheiner: 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: Wed, 17 Nov 1999 13:57:18 -0800 From: Lewis Sheiner <le...@c255.ucsf.edu> Subject: Re: repeating cases--------- The parameters during pregnancy change quite large, so I am not sure if it is a reasonalble approximation to treat occasions as distinct individual, or I have to search the better models of putting those occasions under one ID? and what is the direction to improve the model? Any suggestion is greatly appreciated. Paul School of Pharmacy University of Pittsburgh 716 Salk Hall 3501 Terrace Street Pittsburgh, PA 15261 Phone: 412-648-8546 E-mail: yu...@pitt.edu Yuanyue (Paul) Gao School of Pharmacy University of Pittsburgh 716 Salk Hall 3501 Terrace Street Pittsburgh, PA 15261 Phone: 412-648-8546 E-mail: yu...@pitt.edu
