Experiences with Program-level Modeling and Simulation

Nitin Patel, Ph.D. Chairman, Co-Founder and CTO Cytel Inc.

Disclaimer

•  The views and opinions expressed in the following PowerPoint slides are those of the individual presenter and should not be attributed to Drug Information Association, Inc. (“DIA”), its directors, officers, employees, volunteers, members, chapters, councils, Communities (formerly known as SIACs) or affiliates, or any organization with which the presenter is employed or affiliated.

•  These PowerPoint slides are the intellectual property of the individual presenter and are protected under the copyright laws of the United States of America and other countries. Used by permission. All rights reserved. Drug Information Association, Drug Information Association Inc., DIA and DIA logo are registered trademarks. All other trademarks are the property of their respective owners.

2

Acknowledgements •  Jaydeep Bhattacharya, Cytel inc. •  Armaan Pai, U. Penn. for their excellent support in developing and

running simulations •  My colleagues in the DIA Scientific Working

Group on Adaptive Programs chaired by Carl-Fredrik Burman, especially Jim Bolognese and Christy Chuang-Stein

3

Outline

•  Case study of neuropathic pain Phase 2 study design using simulation of Ph2-Ph3-RevenueModel

•  Advantages and challenges of program level design modeling and simulation (M&S)

4

Program level M&S experiences •  Vehicle for constructive dialog between various

stakeholders in drug development process (NEWDIGS project at MIT)

•  Optimizing drug development process for pharma cos. –  Early stage studies:

•  Ph1 PoC based on biomarkers to dose selection for Ph 3 (with Merck & co)

•  Asset development for multi-indication oncology drug –  Late stage studies: Ph2 to Ph3 to launch (DIA ADSWG, Adaptive

Programs team – Carl-Fredrik Burman (chair). Subteams on applications:

•  Neuropathic pain (led by Nitin Patel) •  Diabetes (led by Zoran Antonijevic) •  Oncology (led by Olga Marchenko) •  Papers published in 2012- 2013

5

Neuropathic Pain •  Investigate impact of various factors on a

Ph2+Ph3 development program for Neuropathic Pain –  Different methods for selecting dose for Ph3 –  Ph2 sample size

•  Hybrid Bayesian/Frequentist approach

–  Statistical analysis of trial data from regulatory perspective is frequentist,

–  Go/NoGo decision making and dose selection for Ph3 trials by sponsor analysis is Bayesian with non-informative prior

–  Posterior distribution of true mean response at each dose and isotonic estimate of pr(AE) from Ph2 data used to choose dose(s) for Ph3

6

Efficacy, Adverse Events, Clinical Utility •  0-10 pain scale used to measure efficacy for treatment of

neuropathic pain in both Ph2 (12 wks.) and Ph3 (12 months) –  Target level of efficacy (mean difference from placebo) = 1 unit –  SD of efficacy response in Ph2 and Ph3 = 2 units –  DR profile used for simulations was Sigmoid Emax

•  Adverse Events (AE’s): non-transient, not manageable by other means (e.g. weight gain, sexual function AE’s) but tolerable to different degrees by patients. AE rates are non-decreasing with dose.

•  Clinical Utility is an index that combines the efficacy measure and AE rate and represents the trade-off between AE rate and Efficacy that would be made in clinical practice

7

Clinical Utility Assessment Clinical utility function was developed using judgments of D Hewitt, MD (Merck) and A Gammaitoni, MD (Nuvo Research), both are clinical experts in Neuropathic Pain.

Clinical utility value of 1 corresponds to Efficacy =1 and AE Rate = 0.2 of leading competitive drug on market

8

0 0.1 0.25 0.40 0 0 0 00.4 0 0 0 00.9 1 1 0.75 0.251.25 1.5 1.5 1 0.51.75 2 2 1.5 1

AE  RateEfficacy                      (diff.  from  Pbo)

8

Clinical Utility Function

Effi

cacy

AE rate 9 9

•  Outcome assessed at program-level by: –  PoS: Probability of Success –  TP: Time under patent after market launch –  EU: Expected clinical utility of launched drug

•  Ph2 results lead to “Go to Ph3” decision if:

•  trend test for drug effectiveness shows significance with1-sided α = 0.05, and

•  dose selection criteria for Ph3 are met

•  PoS for the Ph 2+Ph3 program is prob. of going to Ph3 and having each of 2 concurrent Ph3 trials showing significant drug efficacy over Pbo –  Sample size in each Ph3 trial has power = 0.95 at drug efficacy difference

from Pbo = 1

•  MCMC and Direct MC Simulation necessary to assess outcomes

Outcomes

10

Time Estimates

Total patent life 17yrs. Duration of Dev. Time before the Ph2 trial 2yrs.

Patient Accrual per month per site in Ph2 trial 0.5 # Sites in Ph2 trial 50

Lag between end Ph2b trial and start Ph3 trial 6 mo. Patient Accrual per mo. per site in Ph3 trial 1

# Sites in each Ph3 trial 80

Duration between end of Ph3 trials and launch 12 mo.

11 11

Dose selection for Ph3 We compare two methods for selecting dose at the end of Ph2:

1. Traditional method is to select smallest dose with estimated mean efficacy diff from Pbo ≥ cut-off value near target. 2. Selecting dose with highest utility estimate

We also compare 3 methods to estimate the dose response profile

from the Ph2 data (listed in order of increasingly stronger assumptions about the profile)

Non-parametric ANOVA: No model used relating mean efficacy response at different

doses ANOVAC: ANOVA with constraint requiring means to be monotone

increasing with dose Parametric 4PL: Mean efficacy response follows 4 parameter logistic curve

(used as working model for dose selection)

12

Base Case

D0 D1 D2 D3 D4 D5 D6 D7 D8

Efficacy (Sigmoid Emax)

0.000

0.001

0.03

0.22

0.58

0.854

1.00

1.07

1.10

AE Rate (Binomial

Prob.)

0.10

0.10

0.10

0.10

0.10

0.10

0.125

0.15

0.175

Ph2 Sample Size varies from 135 (15/arm) to 810 (90/arm) # simulations of Ph2 trial = 10,000 (Phase 3 results can be calculated analytically)

Efficacy and AE rate profiles

13 13

PoS: Traditional vs. Utility Max. dose selection

Utility maximization is much better – PoS is almost double for any sample size and across all 3 assumptions for the efficacy profile

Note: The cut-off value for traditional method = 0.8 was chosen to maximize PoS

14

Expected Utility of marketed dose (Utility maximization dose selection)

ANOVA is clearly inferior 4PL and ANOVAC are close but 4PL begins to lose to ANOVAC with increasing sample size due to diff. between Sigmoidal Emax simulation data and 4 PL shape used as working parametric model

15

Trade-off: PoS vs. Remaining Patent Time

16

•  Commonly used criterion that strikes balance increasing Ph2 SS and reducing time on market under patent:

–  expected Net Present Value (NPV)

–  NPV is cash flow (Revenue – Cost) that is discounted over time by the discount rate used by a firm to evaluate investments.

•  Also need to consider –  Probability of Success (PoS) of program –  Expected Utility of marketed dose

Outcome Assessment

17

18

REVENUE

0.00.1

0.20.3

0.40.50.60.70.8

0.91.0

1.11.2

1.3

YEAR

0 10 20

Revenue over time for Effective Patent Life TP=3,7,10,13(S5=$1B, b=0.03, c=1)

Time Profile of Net Revenue

Patent Expiration

Slope after 5th year = b, Decay parameter for period after patent expiration = c

5th Year Revenue is proportional to Clinical Utility

18

5

18

Revenue and Cost Estimates

Cost per site $15K Cost per patient $3.5K

Start-up Cost of manufacturing and marketing $1M Slope of ENPV from 5th year to Patent expiration (b) 0.1

ENPV decay parameter after Patent expiration (c) 0.5 Discount rate per year 10%

19 19

Comparison using Expected NPV

Clearly ANOVA is inferior to ANOVAC and 4PL for estimation of efficacy response about 12% smaller ENPV Optimum Ph 2 Sample size is 270 for all three.

20

                                                                                                                                  Uncertainty in Efficacy and AE profiles

AE Profile D0 D1 D2 D3 D4 D5 D6 D7 D8 Likely 0.10 0.10 0.10 0.10 0.10 0.10 0.125 0.15 0.175

High 0.10 0.10 0.10 0.10 0.15 0.20 0.25 0.30 0.35

0 1 2 3 4 5 6 7 8

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

Doses

Mean Re

sponse

Dose Response curves

1.5 x EfficacyEfficacyHalf Ef ficacyFlat

Doses  

  Mean  Effi

cacy  Respo

nse  Op0mis0c  Efficacy  

 Likely  Efficacy    Pessimis0c  Efficacy      (Sigmoidal  Emax  curves)  

Four  dose  scenarios  have  doses  D0,  D2,  D4,  D6,  and  D8  21

Probability estimates for scenarios

22

1 0.55 Likely 0.5 0.12 0.55 High 0.5 0.13 1.1 Likely 0.5 0.34 1.1 High 0.5 0.35 1.65 Likely 0.5 0.16 1.65 High 0.5 0.1

Prior  Prob  AE  Profile

Jt.  Prior  Probability

Scenario

0.2

0.6

0.2

AE  Profile

Prior  Prob  Eff  Profile

Max  Efficacy  response

22

•  Expected PoS across the Efficacy and AE

profiles

•  Expected clinical utility of marketed dose

•  Time in market from launch to patent expiry

(Dose selection method is utility maximization)

Criteria for Ph2 SS selection

23

Comparison

24

ENPV using Different Models for efficacy profile

4PL is best, however ANOVAC is close. Optimum Ph 2 Sample size is 225 for all 4PL and ANOVAC.

25

Conclusions

•  We found that Program-level simulation for Ph 2 designs can substantially improve the program. Major advantages – Aligns design quantitatively to commercial

objectives like ENPV – Enables evaluation and optimization of several

important design choices: e.g. sample size, dose selection methods, number of doses, dose response model selection

26

Challenges

•  The most significant challenges were: – Time needed to develop and run simulation

models – Need for cross- functional inputs (e.g. clinical

utility functions for safety-efficacy trade-offs, commercial model)

27

Thank you! nitin@cytel.com

www.cytel.com

28