“FUTURE CONSIDERATIONS FOR PK/PD RESEARCH” Terrence F. Blaschke, M.D. Professor of Medicine and Molecular Pharmacology Stanford University.

“FUTURE CONSIDERATIONS FOR PK/PD RESEARCH”

Terrence F. Blaschke, M.D.Professor of Medicine and Molecular Pharmacology

Stanford University

Issue for discussion:

Can PK/PD modeling help to devise dosage regimens that will have better efficacy and/or safety without adding

time/cost to drug development?

Premise:• There is a need for alternative dose-finding

methods since all reasonable regimens cannot be studied using the current standard of a 48 week controlled study of efficacy and safety– Patient resources are limited– Time requirements would be excessive, and delay

patient access to alternative regimens– HIV therapeutics is a fast-moving field, and approved

regimens may not be acceptable as controls to patients or investigators

M =n!

p!(n-p)!

IF:

FOR: n=31 M=4495

n=23 M=1771

n=14 M=364

n=6 M=20

(p=3)

Combinatorials: the numbers problem

“PK/PD Modeling”What is meant by this expression?

Pharmacokinetics (PK) describes the time course of drug concentrations in plasma (and sometimes in other fluids and tissues) resulting from a particular dosing regimen

Pharmacodynamics (PD) expresses the relationship between drug concentrations in plasma (and sometimes in other fluids and tissues) and a resulting pharmacological effect

A PK/PD Model combines • A model describing drug concentrations vs. time (PK) with• A model describing the relationship of effect vs. concentration (PD), and• A statistical model describing variation in intra- and inter-individual PK/PD models

to predict the time-course and variability of effect vs. of time.

Note: Only mechanistic PK/PD models can be relied upon for extrapolation (I.e., for prediction vs. description)

Process:

• Build PK Model• Build PD Model• Link PK and PD models• Simulate treatment regimens or trials for

useful predictions

An Example:(Next few slides courtesy of Abbott Laboratories and Pharsight Corporation)

This simple model links adherence, pharmacokinetics, and viral pharmacodynamics to

treatment outcome in a patient population.

Adherence Pharmaco-kinetics

Pharmaco-dynamics

PrescribedPI Doses Actual

DosePlasmaConc

Antiretroviral Experience,Disease Severity

Model

DataSource

Random,(beta distribution)

fractional adherence rate

Two one-comp.PK models with

enzyme inhibition and induction

Standard two-strainviral model

MEMS data,Public literature

Two multiple-dosePhase I studies,

One Phase II study

In-vitro data,

ViralLoad

Pharmacokinetic Modeling: The PK model accounts for dose-dependent bioavailability, competitive inhibition, and

exposure-dependent enzyme induction.

PI

RTVEnzyme Induction

Enzyme induction

whenapplicable

CompetitiveEnzymeInhibition

AbsorptionSite

Plasma

AbsorptionSite Plasma E

l i m

i n

a t

i o nTime

Time

Fraction Absorbed

WhenApplicable

Fraction Absorbed

Pharmacodynamic Modeling

The model was previously published.# This simple PD model includes two viral strains (wild type and a pre-existing mutant), long-lived infected and

actively infected cells, and different sites of action by PIs and NRTIs.

V2

virus:1 = wild-type

virus:2 = mutant

V1 Tmutate:

m1

1–m1release

infect

proteaseinhibitors

r. t.inhibitors

long-lived cells

actively infected

mutate:m2

1–m2

Tinfect

proteaseinhibitors

r. t.inhibitors

TL2

TA2

TL1

TA1

release

# Hsu A, Wada DR, Liu M et al., PK/PD Modeling of ABT378/Ritonavir Clinical Trials, Including an Adherence Factor. Seventh European Conference on Clinical Aspects and Treatment of HIV Infection, 1999, Oct 23-27.

– For assessing the effect of PK and adherence variability, 400 subjects were simulated for 48 weeks for each of the six regimens, for a dose-time perturbation of 1.6 hr. Adherences with a beta distribution and with a mean of 81% and SD of 0.20 were used for BID regimens, and a mean of 84% and SD of 0.19 were used for QD regimens.

Simulation

Abbott used this approach to compare various combinations PI dosing regimens which included low and moderate dose ritonavir and were able to predict:

• The range of peak and trough concentrations for each of the PI’s in the regimen, and the ratio of trough concentrations to IC50 values

• The effect of varying degrees of nonadherence on the fraction of patients who were likely to experience virological failure

The PK/PD model and the simulations done with it were observed to be consistent with data from several actual trials carried out by Abbott

Building and Evaluating PK/PD Models

• PK models– As part of conventional PK studies,

information on inter- and intra-subject variability is needed

– For drug combinations, interactions should be evaluated at steady-state with dose regimens that include/bracket those likely to be used

– Consider measuring binding proteins such as 1 acid glycoprotein and unbound drug concentrations




Pharmaco-dynamics


DosePlasmaConc


Model

DataSource

Single-coin model,beta distribution

of fractional adherence






One Phase II studyIn-vitro data,

ViralLoad

DATA NEEDED TO CREATE PK/PD MODELS(Much of it is pre-existing scientific knowledge!)

• PD models– Require a combination of in vitro and in vivo

data incorporated into a mechanistic model of viral dynamics (which incorporates baseline CD4, HIV RNA copy number, possibly prior treatment as well)

• Relate in vitro and in vivo sensitivities using early monotherapy data from naïve subjects with wild-type virus

• Expand model to pretreated patients using additional in vitro data using various resistant mutants found in vivo





Pharmaco-dynamics


DosePlasmaConc


Model

DataSource









ViralLoad


In Vitro Pharmacokinetic-Pharmacodynamic System

• Evaluate PK/PD model by comparing outcome of trial simulations to actual data from trials in experienced patients– Response variables: treatment failure and/or presence

of genotypic or phenotypic resistance

– Must incorporate realistic estimates of drug-taking behavior into the simulation

• For the clinical trial used for comparison, actual measures of adherence would be preferable since the effect of different adherence patterns on resistance development is not known





Pharmaco-dynamics


DosePlasmaConc


Model

DataSource









ViralLoad


A simple PK/PD relationship to help understand the

potential consequences of changes in dose regimens or

formulations

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000

Concentration (ng/ml)

EC50=3

Concentration-Response Relationships for Different Quasispecies

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3


0

8

16

24

Do

sing

T

imes

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3


0

8

16

24

Do

sing

T

imes

99%Inhibition @ trough

(Note that the overall antiviral response is the integrated response over time)

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3


0

12

24

Do

sing

T

imes

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3


0

12

24

Do

sing

T

imes


10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3


0

24

Do

sing

T

imes

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3


0

24

Do

sing

T

imes


0

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3 EC50=12


0

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3 EC50=12


0

8

16

24

Do

sing

T

imes

0

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3 EC50=12


0

8

16

24

Do

sing

T

imes


0

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3 EC50=12


0

12

24

Do

sing

T

imes

0

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3 EC50=12


0

12

24

Do

sing

T

imes


0

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3 EC50=12


0

24

Do

sing

T

imes

0

10

20

30

40

50

60

70

80

90

100

1 10 100 1000 3000


EC50=3 EC50=12


0

24

Do

sing

T

imes


PK/PD modeling for AIDS: Where do we stand today?

• PK models for antivirals are generally well-defined

• Several good models of viral dynamics have been developed

• For PIs and NNRTIs, plausible mechanistic relationships between drug concentrations in plasma and inhibition of viral replication have been proposed

• Although simulations using full, mechanistic PK/PD models are consistent with observed data, the robustness of such models in a variety of settings and dosing regimens has not yet been demonstrated

• It is too soon to conclude that PK/PD modeling can substitute for confirmatory trials

General PK/PD modeling: Where do we stand today?

• Continue to improve and refine mechanistic PK/PD models, using in vitro and in vivo data– for individual drugs, in vitro data needs to be

related to in vivo data, including the effect of protein binding, early in development when monotherapy data are being generated

• Generate concentration-response data in early development

PK/PD modeling: Where do we go from here?

• Use PK/PD models to plan trials, limiting dosing regimens and drug combinations to those likely to demonstrate acceptable efficacy/toxicity, and be robust to non-adherence

• Measure adherence as part of the trial


• Consider whether PK/PD modeling based on short term (e.g., 24 weeks) studies can be used as surrogate evidence of long term efficacy– Differences in outcome between 24 and 48

weeks are more likely due to non-adherence rather than regimen failure (use-effectiveness vs. method effectiveness)


“FUTURE CONSIDERATIONS FOR PK/PD RESEARCH” Terrence F. Blaschke, M.D. Professor of Medicine and Molecular Pharmacology Stanford University.

Documents

statistical model

pk model accounts

simple pd model

time pk

viral load slide

viral model mems data

pk models

simple model links adherence