PK analysis of data from different occasions

Dear nmusers, I'm working on an PK analysis of a data set in NONMEM and have run into some questions. I would really appreciate if any of you have the chance to give me some suggestions. The data set I'm analyzing consist of concentration-time profiles obtained at three different occasions for every subject (n=20). Non-compartmental analysis and individual modeling indicate changes in CL/F between the occasions, possibly explained by renal function. If I analyze the occasions separately, a two-compartment model describes the data well at occasions 2 and 3, but not at occasion 1. At occasion 1 large inter-individual variation is observed and a one-compartment model can to some degree describe the data. If I analyze all occasions simultaneously with between-occasion variation, a one-compartment model is able to fit the data reasonably well (two-compartment model is not considered robust). The parameter estimates are in the right ball park, but GOF plots from the simultaneous analysis indicate model misspecification typical for a one-compartment model fit of "two-compartment data". My question is therefore how I should proceed? Will separate or simultaneous modeling be most ideal and what should I consider to get most relevant information? Best regards, Havard Havard Thogersen, Master student University of Oslo/Cincinnati Children's Hospital Medical Center Pediatric Pharmacology Research Unit and Laboratory of Applied Pharmacokinetics & TDM Cincinnati Children's Hospital Medical Center 3333 Burnet Avenue, MLC 6018 Cincinnati, OH 45229-3039 Phone: (513) 636-9011

Hi, Havard: I assume the compound is orally administered. A simultaneous modeling is of course a must for data containing three occasions. Here are few my comments without actually looking at data, GOF and knowing your objectives: 1. identify the most robust absorption model; 2. incorporate IIV and IOV on relative bioavailability; or just IOV on F to explain the interoccasion variability in PK 3. obtain the best IIV model for two compartment model since it describe your data at most occasions except 1. 4. explore different residual error models best, david > The data set I'm analyzing consist of concentration-time profiles > obtained at three different occasions for every subject (n=20). > Non-compartmental analysis and individual modeling indicate changes in > CL/F between the occasions, possibly explained by renal function. If I analyze > the occasions separately, a two-compartment model describes thedata well at > occasions 2 and 3, but not at occasion 1. At occasion 1 > large inter-individual variation is observed and a one-compartment model > can to some degree describe the data. If I analyze all occasions > simultaneously with between-occasion variation, a one-compartment model > is able to fit the data reasonably well (two-compartment model is not > considered robust). The parameter estimates are in the right ball park, > but GOF plots from the simultaneous analysis indicate model > misspecification typical for a one-compartment model fit of > > "two-compartment data". > > My question is therefore how I should proceed? Will separate or > simultaneous modeling be most ideal and what should I consider to get > most relevant information? > > Best regards, > Havard > > Havard Thogersen, > Master student > University of Oslo/Cincinnati Children's Hospital Medical Center > > Pediatric Pharmacology Research Unit and Laboratory of Applied Pharmacokinetics & TDM > > Cincinnati Children's Hospital Medical Center > 3333 Burnet Avenue, MLC 6018 > Cincinnati, OH 45229-3039 > Phone: (513) 636-9011 begin:vcard fn:David Dai n:Dai;David org:Bristol-Myers Squibb Company;Strategic Modeling and Simulation, Clinical Discovery adr:Mail stop E14-07;;Route 206 & Province Line Road;Princeton;NJ;08543;USA email;internet:[EMAIL PROTECTED] title:Senior Research Investigator tel;work:609-252-6342 tel;fax:609-252-7821 version:2.1 end:vcard

Hej Havard, I agree with David that you are much better off with a model that includes all your data. I think the idea of modeling is to be able to explain the whole system and not just a part of it. It seems that you have some changes from one occasion to another. The logical procedure would be to come up with a hypothesis to explain the reasons for the observed change. Adding variability terms might improve the fit but the real question is why there are such big variations between the occasions that your structural model changes from one day to another. I am not dismissing the idea of estimating IIV or IOVs (inter occasional variability). This might very well be real and due to e.g. intake of food on one day and no food on another day. However, if you think your compound changes its own duration of absorption (e.g. through a change in gastric motility), changes its own extent of absorption (e.g. inducing or inhibiting gut wall or liver enzymes), or affects its own elimination, you should add it in your structural model. There are several publications readily available that deal with these types of changes. The problem you might face could insufficient data due to e.g. non-optimal study design. Ha det! Toufigh

Havard, I'd point out that a 1 compartment is just a 2 compartment model where some of the parameters are not identifiable (due to non-optimal sample times, issues with absorption profiles, perhaps flip-flop pk etc). No drugs are one or two (or three or four) compartment. This is a gross oversimplification we impose on a very complex system because it explains data we observe (i.e., the line goes through the dots). So, I suspect your drug isn't "switching" between one and two compartments. Rather it is always 10 or 20 compartments, only 1 or 2 of which you can identify. It seems to me that the answer is that you have a two compartment model where with some subsets of the data some parameters cannot be identified. It isn't a requirement to be able to identify any given model with only a subset of the data. Note that this is perhaps a fixed effect not a random effect, so IIV may or may not be the solution, but per! haps a function of changing renal function, differences in absorption profiles or bioavailability. You'll have to go into some detail on what about the two compartment model isn't robust (no covariance step? sensitive to initial estimates?, high estimation of correlation between parameters?) If your GOF plots are better for the two compartment (and/or better NPDE, better objective function, better PPC), then I'd suggest the two compartment model is better, and you should focus on making it robust. So, it shouldn't bother us to admit that sometimes we can't identify a compartment sometimes, when in reality we can only identify a tiny fraction of the true biological compartments. Mark Sale MD Next Level Solutions, LLC www.NextLevelSolns.com > -------- Original Message -------- Subject: [NMusers] PK analysis of data from different occasions From: "Havard Thogersen" <[EMAIL PROTECTED]> Date: Tue, July 31, 2007 10:47 am To: < > > [email protected] > > > Dear nmusers, I'm working on an PK analysis of a data set in NONMEM and have run into some questions. I would really appreciate if any of you have the chance to give me some suggestions. The data set I'm analyzing consist of concentration-time profiles obtained at three different occasions for every subject (n=20). Non-compartmental analysis and individual modeling indicate changes in CL/F between the occasions, possibly explained by renal function. If I analyze the occasions separately, a two-compartment model describes the data well at occasions 2 and 3, but not at occasion 1. At occasion 1 large inter-individual variation is observed and a one-compartment model can to some degree describe the data. If I analyze all occasions simultaneously with between-occasion variation, a one-compartment model is able to fit the data reasonably well (two-compartment model is not considered robust). The parameter estimates are in the right ball park, but GOF plots from the simultaneous analysis indicate model misspecification typical for a one-compartment model fit of "two-compartment data". My question is therefore how I should proceed? Will separate or simultaneous modeling be most ideal and what should I consider to get most relevant information? Best regards, Havard Havard Thogersen, Master student University of Oslo/Cincinnati Children's Hospital Medical Center Pediatric Pharmacology Research Unit and Laboratory of Applied Pharmacokinetics & TDM Cincinnati Children's Hospital Medical Center 3333 Burnet Avenue, MLC 6018 Cincinnati, OH 45229-3039 Phone: (513) 636-9011