Transcript Slide 1

Linking Modeling & Simulation, Decision Analysis, and the Technology of Drug Formation
Kevin Dykstra, Lee Hodge, Bob Korsan, T.J. Carrothers. Pharsight Corporation, Mountain View, CA
CASE STATEMENT
THE INTEGRATED MODEL
Several attributes of
the drug were
relevant to
development
decisions.
Critical Business Decision
Is it worth trying to find a new formulation of this drug?
Identify Critical
Treatment
Attributes and
Relative Weights
1. Can the main efficacy and tolerability characteristics be
predicted from publicly available summary data?
2. Can sufficient additional patient benefit be provided by an
alternative formulation to warrant further development?
Identify Metrics
and Relevant
Response
Levels for each
Attribute
TreatmentResponse
Models
A Clinical Utility
Index (CUI) provided
a single metric for
decision-making.
Transitioning to Tactical Questions
We were now in a position to explore the implications
of our models together with the team…
CUI Elicitation
Assign
Preference
Values for each
Response Level
CUI
Framework
CUI Distributions for
Competing Treatments
1
Drug and Disease
Modeling
Probability of
Individual
Attribute Levels
E(CUIA )
E(CUIB )
Drug
ModelBuilding
B
0
Here, treatment B is
expected to be superior to A
Expert Opinion,
as needed
DMX User
Interface
“Can the main efficacy and tolerability characteristics be predicted from
publicly available summary data?”
M&S
Customer
M&S
Expert
CUI
Estimated
Product
Profile
Simulated Drug &
Disease Model Outputs
DMX® is a software visualization and communication tool to explore M&S results
Calculate Utility
–
–
–
Used by modeling experts to make M&S results available to teams and decision-makers
Used by the project team to compare performance vs. competing treatments, evaluate product
profiles, and understand trade-offs
THE PK/PD MODELS
6
8
10
SES –Two studies at
three doses with data
were suitable for
modeling
Define response levels for each attribute
●Example: A particular marker could show an effect
Worse, Equivalent, or Superior relative to the standard of care or a key competitor
●Responses to continuous variables are discretized
●Categorical variables fit naturally in this framework
100
80
80
6
60
Cp
8
40
20
0
1
Time after dose (h)
5
10
2
3
4
0
2
Time after dose (h)
-0.2
SES (cfb)
8
IR concentration (ng/mL)
0
6
4
2
4
6
8
8
10
12
0
2
4
Time (h)
6
8
10
Supported by these insights, this project continued development
12
Time (h)
SES (cfb) vs Time (h)
Cp (ng/mL) vs Time (h)
Ka factor: 1.00
Dopahexidine: 8
Ka factor: 1.00
Dopahexidine: 8
Time (h)
5.0%
mean
95.0%
Time (h)
5.0%
mean
95.0%
0.0
0.00
0.00
0.00
0.0
-0.12
-0.07
-0.01
1.5
3.95
5.75
8.06
1.5
-0.92
-0.83
-0.75
6.0
0.73
1.07
1.64
6.0
-0.19
-0.11
-0.05
10
Key learnings from this example of integrating modeling
with drug design to inform decision making
15
Data suggested sigmoidal Emax
curve vs. concentration
SES (cfb) vs Time (h)
We explored the
effects of altering PK
characteristics
SES (cfb) vs Time (h)
Ka factor: 1.00
Dopahexidine: 8
Ka factor: 0.25
Dopahexidine: 24
0
0
-0.2
-0.2
-0.4
-0.4
Study 1 Plb
Cp
cfbSES    Emax 

Cp  EC50
0.0
-0.5
-1.0
5
6
10
15
Cp (ng/mL)

- Intercept
Cp
– population-predicted plasma concentration
Graphs show an
increased duration of
effect at a higher dose,
with absorption rate
decreased by 75%.
-0.6
-0.6
-0.8
-0.8
-1
-1
0
2
4
6
8
10
12
0
2
Time (h)
4
6
8
10
12
Time (h)
Model Expectation
Efficacy (maximal change in SES scale)
1
.228
1
.207
Compliance (%)
2
.169
5
.120
Hypotension (change in mm Hg, systolic BP)
3
.139
4
.140
Drowsiness/Somnolence (% incidence)
4
.135
3
.154
Duration of effect (h)
5
.112
7
.098
Elevated LFTs (incidence over 3x uln transaminase)
6
.109
2
.182
Dyskinesia (% incidence)
6
.109
6
.101
Hypotension
(maximal change in SBP, mm Hg)
Drowsiness/Somnolence
(% incidence)
Worse than competitor X
Effectively the same
Clearly better
-30
-15
-5
-15
-5
0
0.20
0.70
1.00
Severe impact
Moderate impact
Mild impact
0.52
0.26
0.00
0.78
0.52
0.26
0.20
0.60
1.00
Worse than competitor
Effectively the same
Clearly better
0.00
2.00
5.00
2.00
5.00
12.00
0.00
0.50
1.00
Worse than competitor X
Effectively the same
Clearly better
0.35
0.15
0.00
0.40
0.35
0.15
0.15
0.70
1.00
2
1
15
20
25
Study 2 D1
2
4
6
8
10
Cp (ng/mL)
n
Example calculation for a simple CUI:
xn | treatment )   wU
i i  xi | treatment 
wi
0.40
0.30
0.30
Total CUI
Contribution
12
18.4
7.5
37.9
This was a strategic, portfolio-level effort to decide whether the program could
be sufficiently worthwhile to warrant further investment.
•
Weights here are the inter-attribute relative weights.
•
Note that U (utility) is either expressed on a {0,1} or {0,100} scale.
Senior management participation and buy-in were key to the success of the
project.
•
In this example, the NCE has a somewhat higher utility for Effect 1 efficacy but a substantially lower
utility for AE 1 … thus its CUI is lower than the comparator.
With this complete, we were in a position to examine the dose-response for
Clinical Utility and look at sensitivity to particular attributes.
SBP (mm HG)
4
6
8
10
0
2
Cp (ng/mL)
4
6
8
10
Cp (ng/mL)
BP – systolic blood pressure
Cp – predicted plasma concentration
Data (IR and MR) from two studies at five doses.
Unless the uncertainty could be better resolved, this
attribute would not influence the decision.
5.0%
mean
95.0%
16
-0.42
-0.09
0.02
24
0.06
0.07
0.11
36
-0.13
-0.03
0.04
Dose of new “Ka x 0.2” formulation (mg)
The plot shows how much better 24mg with slow absorption
(Ka factor of 0.2) is versus the best dose with “current”
absorption, 8 mg.
Note that when we remove drowsiness / somnolence as a
differentiating attribute, there is little effect on the plot
(green and brown lines are almost on top of each other).
– residual error
1 Plot
Model of dyskinesia was
problematic due to lack of
consistent data
Ka x 0.2
100
Sensitivity analysis for CUI –
duration of efficacy drives
CUI more than does
drowsiness / somnolence
95
2
ei
-1
Ui
30
61.2
25
-2
Attribute
Effect 1
AE 1
Effect 2
-3
Contribution
16
9.6
7.5
33.1
-4
wi
0.40
0.30
0.30
Total CUI
-5
Ui
40
32.1
25
BPi   0  1C p   i ,
-6
Attribute
Effect 1
AE 1
Effect 2
i 1
Comparator
NCE
0
A simple linear model vs. concentration described changes in SBP well
-7
Clinical Development (VP)
Project Management
New Product [Commercial] Development
Biostatistics and DMPK (periodically)
CUI ( x1 , x2 ,
120
120
115
110
SBP (mm HG)
100
95
0
Graph and table compare various
doses of a new formulation (with
absorption 20% as fast as current)
versus the best dose of the current
formulation
Study 2 D2
105
120
115
110
105
100
95
For systolic blood
pressure, two trials
at three doses were
available
– logit function
Predicted Cmax (ng/mL)
SBP (mm HG)
Duration of effect (h)
logit()
CUI scen.
Difference in Clinical Utility Index vs Ka x 0.2
Ka x 0.2
Ref: Current: 8
0.2
Meanwhile, when we remove duration as a differentiating
factor, there is a bigger change (green versus purple).
logit  MuscWkn   Interc  Slope Cmax
0
ineff=0
drows. fixed
durat. fixed
Difference in
Clinical Utility Index
Weight
10
115
Rank
5
CUI scen.: ineff=0
Reference: Current: 8
110
Weight
Techniques like CUI put unwieldy dimensions into a common
currency.
105
Rank
0
Thus the major driver making 24mg with Ka factor of 0.2
better than 8mg at “current” Ka is duration.
-0.2
-0.4
4
6
8
10
-0.6
0
Predicted Cmax (ng/mL)
10
20
30
Dose of new “Ka
0.2” formulation (mg)
Ka xx0.2
logit  MuscWkn   Interc  Slope Cmax
PD Modeling - Musclew k-asthenia Tue Mar 07 18:55:43 EST 2006
40
Enabling software that lets non-modelers "play" and test their
intuition and assumptions can be a powerful way to unite models
with $$ impact.
Difference in CUI vs Dose for
Ka x 0.2
We also examined the effects
of alternative formulations on
overall patient benefit, as
quantified by the CUI
-1
Team
-2
Treatment Attribute
Low
-0.3
-0.6
-1.2
– maximum concentration (ng/mL)
predicted by concentration model
-3
Physician Survey
Not clinically meaningful
Moderately effective
Excellent (super drug)
CMax
-4
Efficacy
(maximal change in SES)
Range values
High
Value
0
0.00
-0.3
0.60
-0.6
1.00
-5
Range name
DrowSom – Fraction of pop experiencing
drowsiness/somnolence
-6
Endpoint
logit  DrowSom  Interc  Slope * CMax
-7
Meaningful response ranges and their
preference values were assessed
Muscle Weakenss (LOGIT of fract incid)
The clinical team identified and
ranked the most important attributes
of the drug
Incidence of
Drowsiness /
Somnolence was
modeled as a
function of Cmax
(IR and MR data from
3 studies at 6 doses)
Time of onset (not in CUI) and maximal change in SES are similar.
Duration of efficacy is substantially increased.
0
Assign weights to the importance of the attributes relative to each other
90% Prediction Interval
-1
4.
Specific, actionable recommendations for a reformulated drug that is likely
to have superior benefit to the current formulation.
Time after dose (h)
-1.0
Study 2
4

-0.6
-1
2
Outcomes were simulated for a variety of drug absorption scenarios.
Results of the simulations were explored with the project team, leading to…
-0.4
-0.8
0
Study 1 D2
-0.5
uscle Weakenss (LOGIT of fract incid)
●
●
●
●
10
Graphs show simulated
time course for Cp and
SES for an 8 mg dose.
Tables give quantitative
predictions for arbitrarily
selected times.
0.0
0
Quantify relative preference for the levels of responses within each attribute
Dyskinesia (% incidence)
The client’s project team:
Ka factor: 1.00
Dopahexidine: 8
Study 4
60
4
Study 1 D1
The analysis was to inform a vital decision and consequently
involved stakeholders from many functions and levels
– elimination rate constant
cfbSES – change from baseline in SES scale
3.
Dopahexidine is eliminated quickly.
● To achieve acceptable duration of effect, relatively large doses must be given
multiple times daily
● This implies brief exposure to high plasma levels and increased incidence of
concentration-related adverse events
Published data did not necessarily include ideal PK/PD information.
● A number of small studies over several years
● Studied either PK or efficacy, but not necessarily at the same time
● Literature was not particularly systematic
Ke
Cp
2
Study 1 D1
SES Change from Baseline
2.
– absorption rate constant
40
0
-2
Incidence of adverse events is also related to drug level.
Examples: Major efficacy endpoint, adverse effects, compliance-related issues, including those affecting key competitors
Ka
20
12
-3
Efficacy is closely related to drug levels (concentration in plasma).
● Measured according to a standard efficacy scale (SES)
Determine the critical attributes affecting the utility of the treatment: the product profile
– 0 if PK assay is Method 1;
We used DMX to
explore and
communicate models
findings for the major
endpoints
SES (cfb)
4
Logit(Drowsiness)
Dopahexidine is indicated for a chronic neuro-muscular disorder.
1.
Assay
100
100
Cp
2
0
Dopahexidine is a mature drug with substantial historical data and
significant tolerability issues
– Assay effect on apparent
bioavailability
Study 3
60
80
60
Cp
0
Time after dose (h)
CUI Elicitation: The Clinical Utility Index assessment process has several distinct steps
Eassay
SES (cfb) vs Time (h)
Ka factor: 1.00
Dopahexidine: 8
SES (cfb)
It is not…
– dose, mg
IR concentration (ng/mL) vs Time (h)
–
–
0
●
D
0
This presentation will provide an example of linking M&S and DA to a formulation decision. The analysis
involved a PK/PD M&S exercise that, when combined with a Clinical Utility Index (CUI, a multi-attribute
evaluation), gave the client an ability to see if altering the formulation of a drug could improve the net
patient benefit and thus further inform the investment decision.
– concentration at time t, ng/mL
0
40
Meanwhile, the use of analytical techniques such as Modeling and Simulation (M&S) and Decision
Analysis (DA) has increased. These methods are enhancing pharmaceutical companies’ ability to make
clear formulation decisions before making huge investments.
a systematic approach to understanding subjective preferences
a transparent way of weighing tradeoffs
knowledge-driven; available data are used; if not, we rely on expert opinion
closely related to the Target Product Profile
a Multi-Attribute Utility Function
an “objective” measure in the sense of a physiological measurement, such as blood
pressure.
Study 2
Study 1
20
It is…
A single compartment
model was adequate
0
●

Cp(t)
1 if PK Assay is Method 2
40
The Clinical Utility Index (CUI) quantifies these tradeoffs by providing a single metric
for the multiple dimensions of benefit and risk.

D Ka
C p (t ) 
( E Assay * Assay ) e  Ket  e  K at
V ( Ka  Ke )
20
Another challenge is the increasing need for drug companies to revisit existing drugs for continued
profitability. Discovery of new drugs has diminished in recent years. Thus companies are looking to
reformulations as a means of finding new ways of differentiating their products and offering patients
benefit.
The PK Model was
based on a pooled
analysis of 6 studies
0
Every drug has benefits and risks. As a result, tradeoffs have to be made among the
drug attributes in the product profile. The relative importance of these characteristics
depends on the indication, patient population, and decision at hand.
100
Historically many companies have approached pharmaceutical development decisions with considerable
ambiguity as to how available options impact the net patient benefit. In particular, since many drug
features “trade off” (e.g., more efficacy is typically accompanied by increased problems with safety or
tolerability), it is difficult for company decision-makers to know the optimal combination of “knob
settings” among their formulation options.
80
THE CLINICAL UTILITY MODEL
Information was gathered information from a variety of public sources.
PK/PD models were constructed describing the main efficacy and tolerability endpoints.
Input was gathered from a variety of internal stakeholders and integrated into a single
metric of clinical benefit.
“Can sufficient additional patient benefit be provided by an alternative
formulation to warrant further development?”
Used by teams to capture knowledge and update as new information becomes available
BACKGROUND
Did we address the tactical questions
(and thus the strategic decision)?
Drug Model Explorer® (DMX®)
A
P(CUI < X)
Dopahexidine is an anonymized mature drug
indicated for a chronic neuromuscular condition
CONCLUSIONS
COMMUNICATING MODEL FINDINGS
To make M&S more regularized and accessible, we must continue to
focus on communication.
Development of staff who can “speak the language” of both modeling
and business decisions.