Pharmacokinetic/
Pharmacodynamic Modeling:
Concepts and Methods
Alan Hartford
Agensys, Inc.
An Affiliate of Astellas Pharma Inc
Outline
Introduction to Pharmacokinetics
Compartmental Modeling
Maximum Likelihood Methodology
Pharmacodynamic Models
Relevance of NONMEM
(A few examples fitting nonlinear mixed
models with R included through-out as
time allows)
2
Introduction
Pharmacokinetics is the study of what a
body does with a dose of a drug
kinetics = motion
Absorbs, Distributes, Metabolizes, Excretes
Pharmacokinetics
Endpoints
AUC, Cmax, Tmax, half-life (terminal),
C_trough, Clearance, Volume
PK/PD Modeling
Procedure:
Estimate exposure and examine correlation between
exposure and PD or other endpoints (including AE
rates)
Use mechanistic models
Purpose:
Concentration
Cmax
AUC
Tmax
Time
6
Figure 2
Compartmental Modeling
A persons body is modeled with a system of differential
equations, one for each compartment
If each equation represents a specific organ or set of
organs with similar perfusion rates, then called
Physiologically Based PK (PBPK) modeling.
The mean function f is a solution of this system of
differential equations.
Each equation in the system describes the flow of drug
into and out of a specific compartment.
9
First-Order 1-Compartment
Model (Intravenous injection)
Input
Central
Vc
Elimination
k10
Solution:
10
Choice of Parameterization
For making distribution assumptions for
parameters, it is more physiologically
relevant to assume that systemic
clearance a random effect instead of
elimination rate.
Because clearance and volume are
assumed to be independent, this reduces
the number of parameters in the
covariance matrix.
11
First-Order 1-Compartment
Model (Intravenous injection)
Parameterized with Clearance
Input
Central
Vc
Elimination
k10
Solution:
12
First-Order 1-Compartment
Model (Extravascular Administration)
Input
ka
Central
Absorption depot:
Central compartment:
Vc
Elimination
k10
Solution:
F = Bioavailability
13
(i.e., amount absorbed)
First-Order 1-Compartment
Model (Extravascular Administration)
Parameterized with Clearance
Input
ka
Central
Vc
Solution:
Elimination
k10
F = Bioavailability
14
(i.e., amount absorbed)
Parameterization
ka, k10, V
Micro constant
ka, Cl, V
Macro constant
Technical Considerations
Outline
General form of NLME
Parameterization
Error Models
Model fitting
(Approximate) Maximum Likelihood
Fitting Algorithms
16
17
and
18
Error Models
Error models used for PK modeling:
Additive error
Proportional error
Exponential error
19
Distribution of Error
In each case, the errors are assumed to be
normally distributed with mean 0
In PK literature, the variance is assumed to be
constant (2)
Heteroscedastic variance is modeled, by
pharmacokineticists, using the proportional error
term
Statisticians, in general, use the approach with
additive error model assuming a variance
function R() where is an m x 1 vector which
can incorporate , D and other parameters, e.g.,
R()=2[f()]2, =[, ]
20
Elimination
k10
21
24
Approach: Approximate ML
Maximum Likelihood
Given data yij, we use maximum likelihood to
obtain parameters estimates for , D, and
2.
Because the mean function, f, is assumed to
be nonlinear in i in pharmacokinetics,
least squares does not result in equivalent
parameter estimates.
26
Approximate Methods
Options:
Approximate the integrand by something we
can integrate
First Order method (Taylor series)
Algorithms Used
Available in NONMEM
First Order
Laplaces Approximation
Importance Sampling
Gaussian Quadrature
Spherical-Radial
Gibbs Sampler
Monolix
Approximate integrand
28
33
Numerical Integration:
Importance Sampling
Consider a function g(b).
Problem!
If each evaluation of the likelihood surface requires
a resampling, then you introduce a randomness
to your likelihood surface.
The likelihood surface would have small
perturbations which would affect your
determination of a maximum.
Solution: sample once and re-use this sample for
each evaluation of the likelihood.
36
Additional Note
When your model does not converge,
often its because you have the wrong
model.
Dont switch algorithms just because of
nonconvergence. First plot data and
scrutinize choice of model.
39
Software
NONMEM (industry standard, 1979,
FORTRAN)
SAS
R and S-Plus
Monolix
WinBugs (PKBugs)
Phoenix (new 2008)
40
Using R
Nonlinear mixed effects fitting function:
nlme (provided by Pinheiro and Bates)
(You also need the lattice package.)
Pre-written PK models available in PKFIT
package
http://www.pharmastatsci.com/pharmacokineti
cs.htm (provided by In-Sun Nam)
41
Hessian Matrix
The Hessian Matrix is the symmetric matrix of second
partial derivatives of the objective function
The 2nd derivative test can be use to confirm
minimization
If the Hessian is positive definite (equivalently, have all positive
eigenvalues) then the objective function has been minimized at
the solution
43
44
First-Order 2-Compartment
Model (Intravenous Dose)
Parameterized in terms of
Micro constants
Note that including Vp overparameterizes the model since
Input
k12
Peripheral
Central
(Vp)
Vc
k21
Elimination
k10
45
Web Demonstration
http://vam.anest.ufl.edu/simulations/secon
dorderstochasticsim2.html#sim
(Requires installation of Adobe
Shockwave player.)
46
First-Order 2-Compartment
Model (Intravenous Dose)
Input
k12
Peripheral
Central
(Vp)
Vc
k21
Elimination
k10
General form of
solution:
47
Another, preferred
parameterization (macro constants)
Q is the inter-compartmental
distribution parameter
Input
Peripheral
Central
Vc
Cl
Elimination
Vp
48
49
Modeling Covariates
Assumed: PK parameters vary with respect to a
patients weight or age.
Covariates can be added to the model in a secondary
structure (hierarchical model).
50
51
53
54
Super-position Principle
Assume dosing every 24 hours
Assume concentration for single dose is
Then concentration, C(t) is
55
56
57
Pharmacodynamic Model
PK: nonlinear mixed effect model
PD:
now assume predicted PK parameters are
true
less PD data per subject (or more, e.g. EKG
data)
nonlinear fixed effect model (mechanistic)
58
Emax Model
Emax * Conc
E= EC50+Conc
59
Mechanistic Models
(from Bill Jusko course 2007)
Reversible
Direct (example: Emax model)
Rapid (CNS, CV)
Slow (Ab, Ca-Ch-BI)
Indirect
Synthesis, secretion
Cell trafficking
Enzyme induction
Irreversible
Chemotherapy
Enzyme Inactivation
60
Effect =
RT * Conc
________________
kD + Conc + K2*Conc2
61
K2=0.01
K2=0.05
K2=0.5
62
Which PD model?
If mechanism is known, then choice of
model is more clear.
If mechanism not known, then trying
different models leads to suggestions
about mechanism.
63
64
Kinetics Diagram
Dose
k12
Plasma
Compartment
V1
k10
Excretion and
Metabolism
Effect
Compartment
V2
k20
Elimination from
V2
65
Kinetics Equations
Param
Description
C1
C2
k10
k12
k20
66
Description
Measured effect
E0
Baseline effect
Emax
EC50,prot
Cprotein
EC50,drug
67
Acknowledgements
Thanks to Huafeng Zhou, Bill Denney and Banmeet Anand
for help with concepts and examples!
Thanks also to Yao Huang for reviewing slides.
Huafeng Zhou, Gilead, Biostatistician
Bill Denney, Pfizer, Pharmacokineticist
Banmeet Anand, Agensys, Pharmacokineticist
Yao Huang, Agensys, Biostatistician
Also referenced was a PD Modeling short course by Bill
Jusko, SUNY Buffalo.
69
References
Davidian, M. and D. Giltinan, Nonlinear Models for
Repeated Measurement Data, Chapman and Hall, New
York, 1995.
Gabrielsson, J. and D. Weiner, Pharmacokinetic and
Pharmacodynamic Data Analysis: Concepts and
Applications, Swedish Pharmaceutic, 2007.
Pinheiro, J.C. and D.M. Bates, Approximations to the
log-likelihood function in the nonlinear effects model, J.
Comput. Graph. Statist., 4 (1995) 12-35.
Pinheiro, J.C. and D.M. Bates, Mixed-Effects Models in
S and S-Plus, Springer, New York, 2004.
The Comprehensive R Network, http://cran.r-project.org/
Pharma Stat Sci, http://www.pharmastatsci.com/
70