- On 5 Dec 2014 at 12:07:19, Roger Jelliffe (jelliffe.at.usc.edu) sent the message

Back to the Top

Dear Colleagues:

1. What method of population modeling gives you the most likely results given the data? NONMEM,

other parametric population modeling approaches, or what? Would you be surprised if it was Pmetrics?

And why is this?

2. What type of population model lets you develop dosage regimens which are specifically

designed to hit a clinically selected target goal (serum concentration, effect, etc.) with maximum

precision? And why is this?

About question #1 - Parametric population modeling approaches ASSUME that the model parameter

distributions have a certain shape. They are ASSUMED to be normal, lognormal, Poisson, etc. The

parameters estimated are those in the equation which determines the specific shape of the

distribution (mean, SD, covariances, etc.). They can also be bimodal, but the ASSUMPTION must be

made in advance of the analysis, which is then CONSTRAINED by that assumption, and because of this

constraint, will never obtain the most likely parameter distributions given the data.

On the other hand, nonparametric modeling approaches such as the NPML of Mallet and Pmetrics of

Neely et al, make no such constraints on the model parameter distributions. They use UNCONSTRAINED

PARAMETER DISTRIUTIONS (UPD’s). Because of this, they always get results that are more likely. You

might look at

Bustad A, Terziivanov D, Leary R, Port R, Schumitzky A, and Jelliffe R: Parametric and Nonparametric

Population Methods: Their Comparative Performance in Analysing a Clinical Data Set and Two Monte

Carlo Simulation Studies. Clin. Pharmacokinet., 45: 365-383, 2006.

About question #2 – With parametric pop models, one can only take clinical action based on the

estimated CENTRAL TENDENCIES of those ASSUMED model parameter distributions. There is only one value

for each parameter. ONLY ONE dosage regimen gets evaluated. It is ASSUMED to hit the target exactly.

On the other hand, what would be the very best population model one could ever have? Somehow, one

might pick up the red phone and call GOD. HE would then tell us each subject’s EXACT model parameter

values. One can never do better than to know the exact values for each subject. The distribution

would be discrete, NOT continuous. As the number of subjects studied approaches infinity, the

distribution, with all its genetic subpopulations, would converge to the true distribution.

In contrast, if a parametric model with an ASSUMED distribution were to be used, the distribution,

with infinite subjects, would converge to the most likely ASSUMED distribution which fitted the data

the best. But that is all.

Models with UPD’s use the ENTIRE model parameter distributions, which consist of many support

points, up to one for each subject studied. With them, one can use multiple model (MM) dosage

design. It works as follows. Give a candidate regimen to each model support point. Each point

generates predictions into the future. At the time the specific target goal is desired, it is easy

to calculate the weighted squared error (each support point is weighted by its probability in the

population) of the failure of the regimen to hit the target. Then one simply finds the regimen which

minimizes the weighted squared error with which the target goal is hit. Now, we have the most

precise dosage regimen.

This is a unique strength of models with UPD’s. If one uses parametric models, one NEVER becomes

aware of this issue, and can do NOTHING to maximize the precision of any dosage regimen, as there is

no provision to do this. UPD models, yes. Parametric models, no. A reasonable reference is

Jelliffe R, Bayard D, Milman M, Van Guilder M, and Schumitzky A: Achieving Target Goals most

Precisely using Nonparametric Compartmental Models and "Multiple Model" Design of Dosage Regimens.

Therap. Drug Monit. 22: 346-353, 2000.

Also, you might look at

Neely M, van Guilder M, Yamada W, Schumitzky A, and Jelliffe R: Accurate Detection of Outliers and

Subpopulations with Pmetrics, a Nonparametric and Parametric Pharmacometric Modeling and Simulation

Package for R. Therap. Drug Monit. 34: 467-476, 2012.

All best wishes for the Holidays,

Roger Jelliffe

Roger W. Jelliffe, M.D., F.C.P., F.A.A.C.P.

Professor of Medicine Emeritus,

Founder and Director Emeritus

Laboratory of Applied Pharmacokinetics

USC School of Medicine

Consultant in Infectious Diseases,

Children’s Hospital of Los Angeles

4650 Sunset Blvd, MS 51

Los Angeles CA 90027

Want to post a follow-up message on this topic?
If this link does not work with your browser send a follow-up message to PharmPK@lists.ucdenver.edu with "Population PK/PD modeling in general" as the subject |

Copyright 1995-2014 David W. A. Bourne (david@boomer.org)