Understanding Parkinson

1
Understanding Parkinsons Disease Model Based
Approach

• Venkatesh Atul Bhattaram,
• Ohid Siddiqui, Joga Gobburu
• - Pharmacometrics, OCP, CDER/FDA
• - Biometrics, OB, CDER/FDA

2
Acknowledgements

• External
• Clinical
• Stanley Fahn MD, Parkinsons Study Group
• Karl Kieburtz MD, NET-PD Steering Committee
• Statistics
• David Oakes PhD, University of Rochester
• Jordan Elm MS, Medical University of South
  Carolina
• Programmer
• Arthur Watts BS, University of Rochester

3
Acknowledgements

• Internal
• Robert Temple MD, Associate Director for Medical
  Policy
• Division of Neuropharmacological Drug Products
• Russell Katz MD, John Feeney MD, Leonard Kapcala
  MD
• Office of Biostatistics
• Jim Hung PhD
• OCP/DCP-1
• Mehul Mehta PhD, Ramana Uppoor PhD
• Pharmacometrics Group, OCP

4

• The objective of this part of the presentation is
  to exemplify the application of disease models.
  Trial design and endpoints will be discussed at a
  future meeting.

5
Impetus

• Drugs to slow the progression of diseases such as
  Parkinsons, Alzheimers are under development.
• Innovative trial designs/endpoints/analyses with
  model based statistical methodologies being
  proposed to discern protective drug effect.
• FDA is asked to comment on the acceptability of
  these trial designs and pre-specified analyses.
• Critical to understand disease/baseline
  characteristics, disease progression,
  placebo/drug effects, and statistical issues
  (Missing data, etc).

6
Preliminary M/S March 05
Initial Thoughts Dec 04
Concept Development Jan 05
1st Internal Meeting Feb 05
Data Collection Sep 05
OCP/OB Group Dec 05
2nd Internal Meeting Oct 05
ACCP Symposium Sep 05
3rd Internal Meeting April 20, 06
4th Internal Meeting Aug 2nd, 06
Clinical/Stat Spring 07
CPSC Oct 06
DIA Jan 07
7
Single point analysis will not differentiate
between protective and symptomatic effects
Unified Parkinson Disease Rating Scale (UPDRS)
The UPDRS is a rating tool to follow the
longitudinal course of Parkinson's Disease. It
is made up of the 1) Mentation, Behavior, and
Mood, 2) ADL and 3) Motor sections. These are
evaluated by interview. 199 represents the worst
(total) disability), 0--no disability.
8
Modeling Cycle
2
Extract Clinical Trial Information

• BASELINE EFFECT/ MODEL
• PLACEBO MODEL
• DROP-OUT MODEL
• DESIGN
• PATIENT DEMOGRAPHICS

UPDATE
4
Plug Sponsor Data, Play
Decide (Go/No Go, trial
design)
Variety of model validation approaches were
employed
9
Key Scientific Questions

1. What are the influential demographic factors
   influencing the baseline clinical scores and
   progression?
2. How do we describe the progression of Parkinsons
   disease (Linear/Nonlinear)?
3. Why patients drop-out of these trials?

10
Parkinsons Disease Database
Data Source Patients Trial Duration
Trial1 NDA 400 1yr 3yr follow-up
Trial2 External 400 1yr follow-up
Trial3 NDA 900 9mo follow-up
Trial4 NDA 200 9mo follow-up
Trial5 External 300 1.5yr
11
Patient Population Model
12
Demographics

• Influence of various demographics such as age,
  gender, disease duration, smoking, caffeine
  intake on baseline UPDRS scores were evaluated
  using regression techniques.

13
Disease Progression Characteristics
14
Selegiline
Mean (SD) of Total UPDRS scores for patients with
Parkinsons disease treated with levodopa alone
or in combination with selegiline for 5 years and
during the one-month washout period
Eur.J.Neurology, 1999, 6 539-547
15
Levodopa, Pramipexole
Mean (SD) of Total UPDRS scores for patients with
Parkinsons disease treated with levodopa alone
or in combination with pramipexole for 4 years
Time, months
Arch.Neurology, 2004, 61 1044-1053
16
Creatine-Minocycline
Mean (SD) of Total UPDRS scores for patients
treated with placebo, creatine, minocycline for
52 weeks.
Neurology, 2006, 66 664-671
17
Disease progression model describes typical
observed well
18
Disease progression model describes observed
distribution well
19
Disease Progression Characteristics

• A linear model can reasonably describe UPDRS
  change post 8 weeks.
• The models presented here and data from the early
  dose-finding of the new compound need to be used
  to support the design/analysis choices for the
  registration trials

20
Missing Data Mechanism
21
Understanding why patients drop-out of
Parkinsons trials

• Clearly patients who discontinued early had worse
  symptoms compared to those who stayed.
• Graphical displays were generated to understand
  the drop-out pattern.
• UPDRS scores in patients who discontinued for
  example in 0-16 versus 16-32 weeks were compared
• Specific risk factor for drop-outs (Parametric
  Hazard Models)
• ? UPDRS at last observed visit?
• Relative to baseline or previous visit?
• Rate of ? between first and last observed visit?

22
Higher scores lead to early treatment
discontinuation
Rescue medication
Time,
23
Is probability of drop-out related to change in
scores from baseline visit?
Duration adjusted UPDRS change
? 8 units
Time,
Duration20 weeks
24
Is probability of drop-out related rate of change
in scores from previous visit?
Time,
25
Is probability of drop-out related to slope?
Slope
Time,
26
Drop-model ValidationModel systematically
deviates from observed
27
Drop-model ValidationModel reproduces observed
well
28
Summary of drop-out modeling

• Predominant reason for drop-out worsening of
  symptoms
• Duration adjusted change and rate of change in
  UPDRS scores from previous visit are principal
  determinants of discontinuation
• Validation to ensure the model predicts
  discontinuation rates well across varied trial
  designs (fixed vs. titration dosing) is in
  progress

29
Statistical Issues in Model Based Analysis and
Simulations
30
Key Statistical Questions

• Does a linear disease progression model
  reasonably describe change in UPDRS post 8 weeks
  randomization?
• What are the reasonable trial design and endpoint
  choices?
• What are the false-positive and false-negative
  rates of concluding protective effect?
• How do we integrate the clinical pharmacology
  findings and statistical findings to address
  regulatory issues?

31
Longitudinal Analysis

• Across various drugs, the mean maximum
  symptomatic effect appears to be achieved within
  4-8 weeks. Beyond that point, change in UPDRS
  scores over time was described well using a
  linear model.
• Model validation was evaluated using standard
  diagnostics
• Predicted versus Observed
• Individual Fits

32
Delayed start design (Alternate Model)
33
Explored endpoints to discern protective and
symptomatic effects

• Placebo Phase
• Compare the slope difference between the placebo
  and drug groups at an alpha of 5
• Active Phase
• Compare the least square mean difference of the
  placebo (now on drug) and drug groups, using
  repeated measures at an alpha of 5

34
Disease Drug Trial Models
No protective effect Null model
35
Clinical trial simulations of a purely
symptomatic drug

• Sample Size 500
• Number of Arms 2
• Allocation 11
• Trial Duration 72 weeks
• Placebo Phase 0-26 weeks
• Active Phase 26-72 weeks
• Measurements 0, 4, 8, 16, 20, 26, 32, 42,
  52, 58, 72 weeks
• Drop-outs 30 per arm

• We considered three dropout scenarios.
• Equal dropouts in both drug and placebo groups
• (b) Unequal dropouts (Higher in placebo group vs.
  drug
• group)
• (c) Dropouts due to need for symptomatic
  treatment and
• toxicity leading to treatment
  discontinuation.

36
Delayed start design (No protective effect -
Null Model)
37
Type-I Error rate Under Null (no protective
effect) Model
Dropout Scenario Placebo Phase (Slope based Comparison-ITT sample) 1 Active Phase (Endpoint LS Means comparison)2 Active Phase (Endpoint LS Means comparison)2
Dropout Scenario Placebo Phase (Slope based Comparison-ITT sample) 1 Available cases LOCF - ITT sample
Dropout not related to drug or disease 5.20 5.00 5.80
Dropout due to lack of effectiveness (equal drop-outs) 5.15 16.35 22.60
Dropout due to lack of effectiveness (unequal drop-outs) 4.95 7.55 11.50
Dropout due to lack of effectiveness and/or toxicity 4.70 12.25 29.15
Dropout due to unobserved outcomes of the trial 6.05 30.15 40.60
1 Linear Random-effect regression model 2
Repeated measures (MMRM) analyses

• Placebo phase preserves Type I error rate

38
Manage and Leverage Knowledge
Information

• Demographics
• Time course
• Drop-out
• Drug Effects

Placebo Disease Models
Translation to recommending primary statistical
analysis methodology for disease modifying
agents in Parkinsons disease.
Knowledge
39
Questions to the Subcommittee

• Is the overall approach to quantifying various
  part of the disease models reasonable?
• Is the approach to qualifying the models
  reasonable?
• What appropriate forum does the committee suggest
  for sharing these advances with the public?