Transcript Enhanced Quantitative Drug Development (EQDD)

Enhanced Quantitative
Decision Making
- Reducing the likelihood of incorrect decisions
Mike K. Smith, Jonathan French, (Pfizer)
Ken Kowalski, (A2PG)
Wayne Ewy (formerly Pfizer, retired).
Global
Pharmacometrics
Six Components of
Model-Based Drug Development*
PK/PD &
Disease Models
Trial Performance
Metrics
Decision
Criteria
Model-Based Drug
Development
Competitor Info.
& Meta-Analysis
Design & Trial
Execution Models
Data Analysis
Model
2
* Lalonde et al, Clin Pharm & Ther, 2007; 82: pp21-32
Quantitative Decision Criteria
• “I’ll know it when I see it…”
• “Evidence of an effect…”
• “Reasonable efficacy and safety tradeoffs”
• WRONG!!!
Quantitative Decision Criteria
• 2 points improvement over placebo.
– Better.
– At least it’s quantitative
• How sure do you want to be?
– Mean 2 points?
– Lower CI 2 points?
– Mean 2 points and lower CI > 0?
P(Criteria|Data)
• Not just P(… | Data)
– Data
– Prior data, model assumptions, parameter
uncertainties
– Trial design
– Dropouts, imputation methods etc.
– Data analytic method
Truth vs Trial
• For a given set of model parameters /
assumptions there will be a “true” outcome
against the decision criteria.
– What is the chance of achieving 2 points
improvement given current information?
– For a given set of parameters we will know whether
we achieve 2 points improvement or not.
• Then for this same set of parameters, apply
design, dropout / imputation models, analytic
technique and assess decision criteria.
Truth vs. Trial - Formally
•  is the true (unknown) treatment effect
– =f(, , ) is specified for a given set of
model assumptions
•
•


•

vector of fixed effects parameters
covariance matrix for between-unit
(subject or study) random effects
covariance matrix for within-unit
(subject or study) random effects
Truth vs. Trial - Formally
• Define quantitative decision rule under
truth () and data-analytic results (T),
e.g.,
– Truth:
– Data:
Go if TV,
No Go if <TV
Go if TTV, No Go if T<TV
• Note TV denotes the Target Value
• Note T could be a point estimate or confidence
limit on estimate/prediction of 
Operating Characteristics
Total
Correct No Go
Incorrect Go
P(True No
Go)
Incorrect No Go
Correct Go
P(True Go)
P(Trial No Go)
P(Trial Go)
1.0
P(Go)
PTS
Total
“True”
No Go
Trial Go
“True”
Go
Trial No Go
11
P(correct)
Example
• Comparing SC-75416 with ibuprofen in
dental pain.
– Published in Kowalski, K.G, et al. “Modeling
and Simulation to Support Dose Selection and
Clinical Development of SC-75416, a
Selective COX-2 Inhibitor for the Treatment of
Acute and Chronic Pain”.
• Decision criteria based on 3 point
difference from ibuprofen in TOTPAR6
endpoint.
Example
PTS = P(  3) = 67.2
360 mg SC-75416 vs 400 mg Ibuprofen
2500
Mean = 3.27,
SD=0.60
F req u en c y
2000
Obs Mean = 3.3
1500
1000
500
0
0
1
2
3
Delta-TOTPAR6
From Kowalski et al: A model-based framework for quantitative decision-making in drug development
Presentation at ACOP, Tuscon, AZ 2008.
4
5
Example
Trial
Trial No Go
LCL95  0 or Mean<3
Trial Go
LCL95> 0 and Mean3
Total
k<3
2081
20.81%
1199
11.99%
3280
32.80%
k3
1729
17.29%
4991
49.91%
6720
67.20%
Total
3810
38.10%
6190
61.90%
10,000
100%
Truth
P(correct) = 70.72%
From Kowalski et al: A model-based framework for quantitative decision-making in drug development
Presentation at ACOP, Tuscon, AZ 2008.
P(Go) = 61.90%
PTS = 67.20%
17
“Nominal” values for OCs
• P(Correct) can be fixed at >=80%
• PTS for initiating a new trial depends on
quadrant, portfolio, stage of development.
– Perhaps minimal “dignity level” for starting a trial.
• Fixing these two implies P(False GO) and
P(False NO GO) must float, depend on
P(Correct) and PTS.
– Driven by decision criteria.
• E.g. For P(Correct) = 80%, P(Incorrect) = 20%,
spent across P(False GO), P(False NO GO).
Iterate / Optimise
• If the operating characteristics “don’t look
good”…
– Change the data analytic model
– Change the design constraints (↑ n /group)
– Change the data-analytic decision criteria for the trial.
• If we fix one or more of the above (e.g. n /group)
then there is limited other things that can
improve OCs.
– Change the data analytic model, change data-analytic
decision criteria for the trial.
The components may
change over time
• “Truth” model / prior will be refined over time.
– P(“True” Go given current knowledge / model)
changes.
• Decision criteria may change.
– Commercial viability changes. [This may change both
our compound target criteria – truth decision rule, as
well as the data-analytic decision rule]
– Acceptable level of confidence for Trial Go decision
changes. [This applies only to data-analytic decision
rule]
Final Remarks (1)
• Greater collaboration required among
kineticists/modelers, statisticians and clinicians
• Kineticists/modelers:
– Explicit and transparent about the assumptions and
limitations of their PK/PD and disease models
– Think strategically about how model will be used to
influence internal decision-making
– Avoid excessive use of NONMEM-jargon and write
reports to broader audience
– Calibrate models against data-derived (non-modelbased) statistics of interest
Final Remarks (2)
• Statisticians:
– Embrace assumption-rich nonlinear models for
decision-making especially in early clinical
development
– Avoid “Phase 3” mentality when designing Phase 2
studies…relying on empirical (assumption-poor)
models to make decisions in early clinical
development can be costly
• Clinicians:
– Quantitatively define clinically relevant effects and
commercial targets
– Explicitly and quantitatively defined decision rules
Bibliography
1.
2.
3.
4.
Kowalski, K.G., Ewy, W., Hutmacher, M.M., Miller, R., and
Krishnaswami, S. “Model-Based Drug Development – A New
Paradigm for Efficient Drug Development”. Biopharmaceutical
Report 2007;15:2-22.
Lalonde, R.L., et al. “Model-Based Drug Development”. Clin
Pharm Ther 2007;82:21-32.
Kowalski, K.G., Olson, S., Remmers, A.E., and Hutmacher, M.M.
“Modeling and Simulation to Support Dose Selection and Clinical
Development of SC-75416, a Selective COX-2 Inhibitor for the
Treatment of Acute and Chronic Pain”. Clin Pharm Ther, 2008;
83: 857-866.
Kowalski, K.G., French, J.L., Smith, M.K., Hutmacher, M.M. “A
model-based framework for quantitative decision making in drug
development”. Presentation at ACOP, Tuscon, AZ. 2008.
http://tucson2008.go-acop.org/pdfs/8-Kowalski_FINAL.pdf