Transcript How can Statistics Facilitate the Clinical/Nonclinical

Using PK/PD Modeling to Simulate Impact
of Manufacturing Process Variability
Alan Hartford
Agensys
Tim Schofield
Biologics Consulting Group, Inc.
The 32nd Annual Midwest Biopharmaceutical Statistics Workshop
May 18 – 20, 2009, Muncie, Indiana
Introduction


The manufacturer has the responsibility of keeping
manufacturing process variability of the dose in
control.
One method for assuring that a product, after a reformulation, is viable is to perform a clinical trial
showing bioequivalence (BE) of exposure endpoints.
Investigate with Modeling


PK/PD models can be used to simulate the impact of
variations of dose on the response cascade of
dose → exposure → pharmacodynamics → clinical
outcome
These models can address the appropriateness of




the BE study
choice of bounds
Specific information from clinical development is
needed as input for this PK/PD modeling.
Important information from nonclinical development
can also be incorporated to save clinical resources.
Outline







Introduction
Bioequivalence (BE) bounds
PK/PD Models
Predict effect of process variation on clinical
outcome
Required clinical information
Collaborative modeling nonclinical/clinical
Summary
Current Practice


When a new formulation is developed, the
current practice is to perform a clinical study
to show the new formulation is
“bioequivalent” to a previously studied
formulation.
This allows for inference of conclusions from
earlier studies for the new formulation.

i.e., efficacy results from a Ph III study can be
inferred to a new formulation
BE Requirements
Strict bioequivalence (BE) bounds are used for
exposure endpoints (AUC and Cmax)



The geometric mean for AUC and Cmax is
calculated for both formulations.
If the formulations are similar, the ratio of exposure
for new formulation / old formulation ≈ 1.
For BE, AUC and Cmax of new formulation
compared to approved formulation must have 90%
CI of GMRs to be within (0.80, 1.25)
6
BE Requirements (cont.)




This strict BE requirement is standard for many
clinical comparisons (e.g., interaction studies,
elderly/young studies, insufficiency studies)
But (0.80, 1.25) may not be appropriate for clinical
reasons
(0.80, 1.25) is standard for when no clinical
justification can be given for other bounds
If victim drug has wide therapeutic window, then
wider bounds are appropriate
7
BE Requirements (cont.)


For drug interaction studies, FDA
suggests that boundaries can be justified
by a sponsor based on population
average dose, concentration-response
relationships, PK/PD models, or other
So the onus is on the sponsor to justify
other comparability bounds
FDA Guidance: Drug Interaction Studies--Study Design, Data Analysis, and
Implications for Dosing and Labeling (draft 2006)
Example of Using Alternate
Comparability Bounds

In the case of testing if a new antibody has
an effect on the exposure of a standard of
care chemotherapy



Not ethical to sample many patients with
cancer for this interaction trial
Variability of the chemo (AUC, Cmax) was high
FDA accepted plan with small N which required
GMR to be within (0.80, 1.25) but that the
90% CI to be within (0.70, 1.43)
FDA Guidance to Clinical


FDA Guidance for some studies (e.g.,
interaction studies) allows for some leeway
for sponsor to clinically justify alternative
bounds
PK/PD modeling can be used to justify
different bounds for comparing formulations
10
Modeling & Simulation

PK/PD M&S and Clinical Trial Simulation
can provide insight to



Effect of variation in dose on exposure
Effect of variation in exposure on PD
Effect of PD on clinical endpoint
11
Example: Selection of Dose to
Achieve 40% Effect
Example: Assume Emax model is appropriate for a drug
Target of 40% Response
This target of exposure not appropriate.
Does not allow for variability in exposure or effect.
12
Suppose Clinical Selection of
Dose was to Achieve 40%†
†For
this example, 40% was chosen completely arbitrarily and not generally a target
chosen for most drugs.
Range of exposure for patient population
due to variability in PK parameters
{
Increase dose to ensure
mean exposure high enough
(~38) to conclude statistical
significantly  40% effect
13
Assume the following dose-linear
relationship is observed/verified
Need exposure of ~38
for desired clinical effect
Needed dose is then ~15 mg
And should have been confirmed
in clinical development
14
Manufacturing Variability




Every manufacturing process has
specification limits
Product from this process is allowed to
vary within these limits
In this manner, the dosage of drug is
not constant across a batch
The effect of the manufacturing
variability is what we need to
understand
Incorporating Manufacturing
Variability
To determine effect of manufacturing variability on the
sequence of
Dose:Exposure:Response
Perform simulations
1. Assume manufacturing variability limits
2. Using dose linear relationship and incorporating PK
model with inter-subject variability, determine effect of
additional variability due to manufacturing process on
exposure
3. Using PK/PD model (e.g., Emax), determine effect of
compounded variability in exposure in step 2 on clinical
or PD effect
16
Effect on Exposure due to
Manufacturing Variability and
Subject-to-Subject Variability in PK
May need to increase dose
beyond 15 mg to ensure
exposure for all above 38, but
only if safe
17
Increase Effect Target to Account
for Variability in Exposure
Mean target for response is now
>40% to account for variability in
exposure
18
Limits for Effect



Note that the 40% effect size was determined
from Ph III development and not an effect
size targeted in earlier studies
Likewise, an upper limit for effect to be
determined by the safety profile observed
throughout clinical development
Simulations using PK/PD models will help to
determine acceptable limits of manufacturing
variability
19
Including Safety Information
from Clinical Development
In clinical
development,
there is a
maximum dose
studied or
maximum dose
found to be safe
This clinical information
provides an upper limit for
manufacturing variability on
dose.
20
Required Clinical Information

The information needed from clinical
development includes


Upper limit on exposure due to safety
Target response for efficacy
Required Clinical Information
(cont.)

Additionally, clinical information is
needed to build the PK/PD model


Need to sample responses across wide
range of exposure values to understand
what model is appropriate
Note that this can be at odds with goals of
adaptive designs
Dose-Exposure Relationship



Earlier, we assumed a linear doseexposure relationship
However, this relationship might not be
known for patients for a new
formulation
Nonclinical and preclinical modeling
could be used to provide this
information
Additional Modeling
Opportunities


delete
Modeling approaches are used widely
across drug development
These different modeling efforts can be
linked across nonclinical and clinical
delete
Expanded Problem statement

How can nonclinical development
collaborate with clinical development to
demonstrate that a manufacturing
process is delivering product to the
patient that is safe and effective?
Potential paths
Process Parameters (x’s)


Pro: Can directly study impact
of process parameters on
patient outcome
Con: Too many combinations
to study
Patient Outcomes (z’s)
Potential paths (cont.)
Process Parameters (x’s)


Pro: Can study many
combinations of process
parameters in a homogeneous
population
Con: Uncertain relationship of
response in animals to
response in humans
Preclinical
Models (ẑ’s)
Patient Outcomes (z’s)
Potential paths – nonclinical
and preclinical
Allometric
Scaling
Process Parameters (x’s)


Pro: Can study many
combinations of process
parameters in a homogeneous
population
Con: Uncertain relationship of
response in animals to
response in humans
Preclinical
Models (ẑ’s)
Patient Outcomes (z’s)
Exposure (ž)
Potential paths – nonclinical
Process Parameters (x’s)


Pro: Can study many
combinations of process
parameters in vitro
Con: Less certain relationship
of response in vitro to response
in humans
Quality
Attributes (y’s)
Patient Outcome (z’s)
The pieces
Specifications
Maximum Release  f(Maximum, Var)
 Maximum - Assay Variability
 Maximum - t df  s Assay

g( y)  z
Minimum Re lease  f (Minimum ,Loss, Var)
 Minimum  Loss  Combined Variability
Maximum Specification
 Minimum  bˆ  t df  sb2ˆ  s2Assay
Minimum Specification
-6
0
Capability
Release
Limits
Limits
Shelf-Life
6
12
18
24
30
36
Specifications
Starts with clinically supportable
maximum and minimum limits
(specifications)
Maximum release calculated from
release assay variability
Minimum release calculated from
shelf life, stability, and release assay
variability's
30
The pieces
Design Space
f ( x)  y
y
1
x  f ( y)
USL
LSL
Dave Christopher
PhRMA CMC SET
31
The pieces
x  f 1( y)  
Design Space (cont.)
Posterior Predicted Reliability with
Temp=20 to 70, Catalyst=2 to 12, Pressure=60, Rxntime=3.0
Rxntime
Pressure
70
 x such that Prob(Y is in A | x, data)  1  
0.7
0.6
60
= Design Space
50
x2=
0.4
Temp
Contour plot
of p(x) equal to
Prob (y is in A
given x & data).
0.3
40
0.2
30
The region inside the
red ellipse is the
design space.
0.5
0.1
0.0
20
2
John Peterson, GSK
June 11, 2008, Graybill Conference, Ft. Collins, CO
4
6
8
10
12
x1= Catalyst
32
The pieces
IVIVC

An in-vitro in-vivo correlation (IVIVC) has been
defined by the FDA as “a predictive mathematical
model describing the relationship between an in-vitro
property of dosage form and an in-vivo response”



Main objective is to serve as a surrogate for in vivo
bioavailability and to support biowaivers
Might also be used to bridge in vitro and in vivo activity
along the pathway from manufacturing process to patient
outcome
IVIV relationship (IVIVR) more appropriate to the goal –
g(y)=ž
Potential paths – IVIVC
Design
Space
IVIVC
PK/PD
Modeling
Process Parameters (x’s)
Quality
Attributes (y’s)
PK Profile (ž’s)
Patient Outcome (z’s)
IVIVC

FDA guidance offers 5levels of correlation

Level A correlation comes
closest to defining IVIVR –
the purpose of level A
correlation is to define a
direct relationship between
in vivo data such that
measurement of in vitro
dissolution rate alone is
sufficient to determine the
biopharmaceutical rate of
the dosage form
Fdiss=fraction dissolved
Fabs=fraction absorbed
The pieces
IVIVC (cont.)

IVIVR established from “link model”
among in vitro dissolution, in vivo plasma
levels, and in vivo absorption

Fraction absorbed is
obtained in one of 3-ways:

Wagner-Nelson method
C T  K E 0 Cdt
T
FT 

K E 0 Cdt
CT = plasma [C] at time T
KE = elimination rate constant

Loo-Riegelman method
CT  K 10 0 Cdt  ( Xp )T VC
T
FT 

K 10 0 Cdt
(XP)T = [C] in peripheral comp. after oral
VC = volume in central compartment
K10 = elimination rate constant after IV

Numerical deconvolution
The Full Cascade of
Information




Processing Parameters (x’s)
Quality Attributes (y’s)
PK (exposure) Parameters (ž’s)
PD or Clinical Outcome (z’s)
Potential paths (cont.)
f ( x)  y

g( y )  z

h( z)  z
xLimit  f 1( yLimit )
yLimit

 g (zLimit )
1


zLimit  h1(zLimit )
Summary


A process has been outlined for using
information from different stages of
drug development to determine process
limits
Process will inform decision about
needing additional clinical trials for new
formulations
Summary (cont.)

Clinical information is needed for
successful modeling




Target for efficacy
Safety
In total, the therapeutic window
IVIVC or IVIVR models needed to
inform about exposure
Summary (cont.)


Both PK/PD Modeling and IVIVC
modeling are time-consuming and
tedious and must be integrated early
into development
Designs of clinical trials must be
designed so that information needed for
building models is available