1. How Modeling and Simulation has
impacted Decision-Making in New
Drug Development
Raymond Miller, D.Sc.
Pfizer, Inc. Ann Arbor, MI
2. Outline
⢠Pre-clinical PK/PD in experimental models
⢠PK/PD in experimental models in patients
or healthy subjects
⢠PK/PD in dose-ranging study in patients
⢠PK/PD analysis in confirm studies for
efficacy and safety
⢠NDA â PK/PD in regulatory decisions
3. Early Drug Development
Objectives
⢠Select promising compounds.
⢠Identify safe and effective doses and dosing
regimens.
⢠PK/PD helps with compound selection and
then guide an efficient clinical development
strategy.
4. Preclinical Model of Behavioral Activity
to Predict Potency and Time-Course of
Response in Humans
⢠Single time point dose response and time-
course of response studies conducted in rat
⢠Plasma samples collected from a subset of
the experimental group
⢠Naïve Pooled PK model parameters
obtained
⢠PK parameter estimates utilized in a link
PK/PD model to estimate EC50 and Ke0
5. PK/PD Model:
Oral Dose (1)
Ka
Central
Compartment
(2) V2
Cp
K K23 (=.001*K)
Effect Site (3, Ceff)
Emax*Ceff
Elimination Effect =
Keo EC50+Ceff
Effect site equilibration
Parameters of greatest Interest, relative potency and equilibration of effect
6. Reference Drug One Compartment PK Model
DOSE: 30
Cp Tmax ~ 1 hour
12000
1000
DOSE: 10
Observed, ng/mL
Predicted, ng/mL
12000
1000
0 2 4 6 8 10 12
Time, hr.
7. Test Drug One Compartment Model
0 2 4 6
DOSE: 30 DOSE: 100
50000
Cp Tmax ~ 0.5 hr
50000
20000
20000
DOSE: 3 DOSE: 10
5000
1200
2000
100
0 2 4 6
Time, hr.
9. Effect Site Concentration vs. Observed and Predicted Response
DRUG: Test Compound
100
Observed Response
0
DRUG: Reference Compound
100
Observed Response
0
Predicted Response
0 10000 20000 30000 40000 50000 60000 70000
Ceff, ng/mL
10. PK/PD Results
⢠Effect Site Equilibration for Test
Compound 12 times faster than for
Reference
⢠Test Compound is Ÿ as potent as Reference
11. Simulated Human Response to the Test Compound
7
6
5 1X mg Reference
Response
4
4X mg Test (t1/2=4.0 hrs)
3
2
1
0 2 4 6 8 10 12 14 16 18 20 22 24
Time Post-Dose (hr)
12. Test Ke0 14 times faster, 1/12 as potent
4.0
Reference 1x
Test 8x
3.5 Test 12x
3.0
Response
2.5
2.0
1.5
1.0
0 2 4 6 8 10 12
Time Post-Dose (min)
13. Phase 2 Clinical Studies
Objectives
⢠Does the drug work
⢠Does any tolerable dose provide a minimal
clinically acceptable level of response
⢠What is the lowest dose that provides the
minimal clinically acceptable level of effect
⢠What is the lowest dose that provides the
âbestâ benefit to the patient.
⢠What are the best doses and regimens for
use in Phase 3
14. Planning Phase 2a dose-ranging trial in
Alzheimerâs Disease (AD)
⢠Primary question: Does drug have benefit?
â Useful efficacy would be > tacrine:
⢠3 points on ADAS-Cognitive (ADASC) after 12 weeks
⢠Important secondary question: Is more drug
better than less?
â Preclinical data suggested dose response (DR) could be
U-shaped or monotonic
⢠Limited to 12 weeks of therapy (Toxicology)
15. CATD
⢠PK/PD information from preclinical, Phase
1, Phase 2, and the literature to build drug
and disease models from which realistic
virtual patients can be simulated.
⢠Simulations to assess the probability
distribution of clinical trial outcomes, the
sensitivity of these outcomes to
uncontrollable factors, merits of alternate
study designs.
16. Simulation methods
⢠Pharsight Trial Simulator for the final
simulations
⢠SAS for the analysis of the simulated
datasets.
⢠For each treatment sequence, a population
of patients (n=1500) was created and these
were sampled with replacement to generate
individual clinical trials (from 100 to 2000
depending upon the precision needed for the
particular question).
17. Trial Performance Criteria
(âTrial Metricsâ)
1. Statistically significant evidence of some
beneficial drug effect:
p< 5% for dose group vs. placebo (mult. adjusted)
2. Correct characterization of dose response pattern
As monotonic or U (reversal) or flat
3. Sufficiently accurate estimation of effect size
â Despite the likely carry-over
â Acknowledging that short-term treatment will not
show us full steady-state effect
19. Trial Designs Compared
⢠All had similar size (~cost):
â N~60, 4-8 observations/patient, 10-12 weeks on drug
⢠Latin squares (no washout between periods):
â 6x6 (6 sequences, 6 2-week periods)
â 4x4 (4 sequences, 4 4-week periods)
⢠Parallel group (4 or 6 groups, 12 weeks treatment)
⢠Incomplete block designs (6 seq., 3 4-week
periods)
⢠Combinations: 4x4 LS and 2 group parallel group
⢠Also compared variety of analysis models
â How to assess/deal with carry-over
â How to discriminate monotonic from U-shape
20. Other Assumptions
⢠Drop-out rate =1%/week
⢠Specific population mix (gender, smoking,
ADASC) based on prior experience
⢠Disease progression: 6 points/year
⢠Placebo response ( peak day 7, fade by 42)
⢠MSE=16
21. Metrics for Each Trial
⢠âPositive Trial?â P<.05
â Alternatives include LQ trend test, Hochberg
⢠âCorrect Shape Characterization?â
â Based on observed pattern in LSMeans
â Correct if characterization matches truth
⢠âEffect Estimate Close Enoughâ
â Close enough if within 1 point of truth
22. Power for âPositive Trialâ
% of 100 Trials with Significant Drug Effect
Effect Onset Slow Tacrine
#8: #6: #1
4x4 4x4 6x6
Design 4 Wk 3 Wk 2 Wk
Response Model
Linear 84 51 41
Emax 88 67 43
SEmax 96 85 68
U-Fat 57 49 39
AVERAGE 81 63 48
These are the 3 best designs â all others had less power
23. Power for âCorrect Shapeâ
% of 100 trials correctly characterized
(Relaxed rule for study significance: 20% 2-tail)
Effect Onset Slow Tacrine
#8: #6: 6x6
4x4 4 4x4 3 2
Design Wk Wk Wk
Response Model
Linear 96 72 53
EMax 84 74 44
SigEmax 96 89 64
UFat 45 39 45
AVERAGE 80 69 52
24. Simulation Conclusions
Design
⢠4x4 LS with 4-week periods using bi-
weekly measurements
â Was best among alternatives considered for
detecting activity and identifying DR shape
â Met minimum design criteria (80% average
power)
25. Results
⢠4x4 LS design was accepted, conducted,
and analyzed more-or-less as recommended
⢠Unfortunately, drug didnât work
â But we were able to find this out more quickly
and with less resources than with conventional
design
26. Trial Simulation
⢠We used trial simulation to
â Compare performance of alternative designs
⢠Across a range of possible data models and other assumptions
â Explore alternative analysis methods and decision
criteria
⢠In process, we developed âevidenceâ that the
proposed design would work and be cost-
effective.
â Very useful for during management review
27. DMX
(Drug Model Explorer)
Drug Model Explorer (DMX) Simulate
Response
Space
Drug and Disease Drug
Modeling-
Model Outputs
Building
DMX User Interface
DMX User
M&S Group
DMX is an easy to use, interactive tool to help users quickly
query quantitative dose-response information (safety &
efficacy) for the development candidate & key competitors
Helps team address critical clinical development questions:
dose-response, response in a target population, probability
of a particular response at a given dose, dose-range to
achieve a target response, probability of superior response
vs. comparators
28. Actual User Interface of DMX Tool
Endpoints
Plots Display Trends
Uncontrollable
Variables
&
Assumptions
Tables Display Detail
Controllable
Variables
Output Controls
29. Interpreting â Competitive Environment
Integrated analysis of all relevant data on new compound and key competitors on all key
endpoints
Graphical and tabular display comparing the results of our new compound to a key
competitor
Estimate the probability that the mean response for A is at least X% better than the key
competitor
1 Plot LDL % change from baseline 1 Plot LDL % change from baseline
against against
Treatments Treatments Atorvastatin (mg)
10
10
LDL % change from baseline
LDL % change from baseline
0
-10
-20
-30
(mg) 0.00
(mg) 3.00 (mg) 0.00
(mg) 6.00 (mg) 1.00
-40
(mg) 9.00
-50
-70
-60
0 2 4 6 8
0 2 4 6 8
(mg)
Atorvastatin (mg)
30. Phase 3 Clinical Studies
Objectives
⢠Confirm exposure-response â Phase 2
⢠Assess significant covariates
⢠Assess inter- and intra-individual variability
⢠Biomarker prediction of efficacy
⢠Predict drug interactions
⢠Understand relationship between drug
exposure and clinical outcome
31. Gabapentin â Neuropathic Pain
NDA
⢠Two adequate and well controlled clinical
trials submitted.
⢠Indication â post-herpetic neuralgia
⢠Trials used different dose levels
â 1800 mg/day and 2400 mg/day
â 3600 mg/day
⢠The clinical trial data was not replicated for
each of the dose levels sought in the drug
application
32. Gabapentin Study Designs for
PHN
Overview of PHN Controlled Studies: Double-Blind Randomized/Target Dose and ITT Population
Duration of Double-Blind Phase Number of Patients
Final Gabapentin Dose, mg/day
Fixed Overall Any All
Titration Dose Duration Placebo 600 1200 1800 2400 3600 Gabapentin Patients
4 Weeks 4 Weeks 8 Weeks 116 -- -- -- -- 113 113 229
3 Weeks 4 Weeks 7 Weeks 111 -- -- 115 108 -- 223 334
4 Weeks 4 Weeks 8 Weeks 152 -- -- -- 153 -- 153 305
379 0 0 115 261 113 489 868
t included in study design
All randomized patients who received at least one dose of study medication.
⢠Used all daily pain scores (27,678
observations)
⢠Exposure-response analysis included titration
data for within-subject dose response
34. Model Predicted Gabapentin Effect (Less Placebo) Plot of Model Predicted Gabapentin Effect by
1.6 Total Daily Dose and Estimated Dose Absorbed
1.5
1.4
1.3
1.2
1.1
1.0
0.9
0.8
0.7
0.6
0.5
0.4 Total Daily Dose
0.3 Estimated Dose Absorbed
0.2
0.1
0.0
0 500 1000 1500 2000 2500 3000 3500 4000
Gabapentin Dose (Total Daily or Total Daily Absorbed)
35. FDAMA 1997
FDA review staff decided to explore whether PK/PD analyses
could provide the confirmatory evidence of efficacy.
ââbased on relevant science, that data from one adequate and
well controlled clinical investigation and confirmatory evidence
(obtained prior to or after such investigation) are sufficient to
establish effectiveness.â
36. Important Features of the Data
⢠Adequate and well controlled trials
⢠Analysis prospectively planned
⢠Response variable is clinical endpoint
⢠Ample data â several dose levels + placebo
⢠Longitudinal data
⢠Well characterized PK
37. Results
⢠Summary statistics showed pain relief for
both studies at different doses concur.
⢠M & S showed pain scores for both studies
can be predicted with confidence from the
comparative pivotal study (cross
confirming).
38. Conclusion
⢠The use of PK/PD modeling and simulation
confirmed efficacy across the three studied
doses, obviating the need for additional
clinical trials and thus supporting the
approval of the product.
⢠The package insert states
âpharmacokinetic/pharmacodynamic
modeling provided confirmatory evidence
of efficacy across all dosesâ
39. Commentary
From discovery to market, drugs have a high rate of attrition.
Because of the complexity, risk, and cost of drug discovery
and development, drug companies must apply the best
scientific methods and technology, as well as decision making
processes, to facilitate early termination of nonviable
candidates while rapidly advancing viable ones.
PK/PD modeling and clinical trial simulation provide useful
insight at every stage to help identify optimal candidates as
early and as with few resources as possible.
At each decision point the objective is to use the best
exposure-response and other scientific evidence to make
decisions.
40. Final Note
FDA guidance's and publications have emphasized the
importance of integrating pharmacokinetic and
pharmacodynamic (PK/PD) information and drug
development and its impact on decision making.
Pfizer has created a list of activities at specific decision
points from discovery to registration that should be
required in order to make the most informed decisions
based on all relevant PK/PD information.
41. Acknowledgements
⢠Jeffrey Koup ⢠David Hermann
⢠Daniele Ouellet ⢠Brian Corrigan
⢠Wayne Ewy ⢠Bill Frame
⢠Peter Lockwood ⢠Richard Lalonde
⢠Pharsight