Transcript Slide

Quantitative “Learning” Approaches
Influence Drug Approval
Hao Zhu, Ph.D.
Pharmacometrics
FDA
DISCLAIMER: The opinions presented today are mine
and do not necessarily represent those of the U.S. FDA.
1
Hao Zhu, ACSMSBB 2008
Outline

Introduction
–
–

Unfractionated Heparin (UFH) Case Study
–
–
–


2
Pharmacometrics
Impact at a glance
Clinical trial
Quantitative ‘Learning’ Analysis
Results
Summary
Acknowledgement: Drs. Norman Stockbridge,
Yaning Wang, and Joga Gobburu
Hao Zhu, ACSMSBB 2008
Pharmacometrics
(or Quantitative Experimental Medicine)

Science that deals with quantifying disease and
pharmacology

Focus is on ‘learning’ rather than ‘confirmation’

Diverse expertise needed

3
Quantitative (Clinical) Pharmacologists, Clinicians, Statisticians,
Bioengineers
Hao Zhu, ACSMSBB 2008
4
Evidence of Effectiveness
Labeling
Quantify benefit/risk
Dose optimization
Dose adjustments
Trial design
1. Bhattaram et al. AAPS Journal. 2005
2. Bhattaram et al. CPT. Feb 2007
3. Garnett et al. JCP. Jan 2008
4. Wang et al. JCP. 2008 (in press)
Hao Zhu, ACSMSBB 2008
Decisions
NDA Reviews1,2
Protocols
Dose-Finding trials
Registration trials
QT Reviews3
Central QT team
EOP2A Meetings4
Disease Models
Knowledge
Management
Tasks
Pharmacometrics Scope
Impact at A Glance
5
Drug
Varenicline
Impact
Dose-response to support lower doses
in labeling
Ranolazine
Supportive evidence; Contraindication
of hepatic impaired
PAH drug
Cause of trials failure; Alleviation of
false +ve QT signal
CNS drug
Confirmatory evidence
Hao Zhu, ACSMSBB 2008
Impact at A Glance
6
Drug
Trileptal
Impact
Monotherapy in pediatrics without
controlled clinical trials
CCB
“Approvable” due to sub-optimal
dosing regimen
Zometa
Dose adjustment in renal impaired
Busulfan
Pediatric dosing
Hao Zhu, ACSMSBB 2008
UFH Case Study:
TIMI – 25 Study Meets its Primary Goal
STEMI Patient
Aspirin + fibronolytic drug
7
Enoxaparin
UFH
~ 8 Days
(control)
~ 48 hrs
Cumulative Events
Study Design
Antom EM et al. NEJM, 2006
Hao Zhu, ACSMSBB 2008
UFH Case Study:
TIMI – 25 Study Meets its Primary Goal
STEMI Patient
Aspirin + fibronolytic drug
8
Enoxaparin
UFH
~ 8 Days
(control)
~ 48 hrs
Cumulative Events
Study Design
Days after randomization
Antom EM et al. NEJM, 2006
Hao Zhu, ACSMSBB 2008
UFH Case Study:
TIMI – 25 Study Meets its Primary Goal
STEMI Patient
Aspirin + fibronolytic drug
9
Enoxaparin
UFH
~ 8 Days
(control)
~ 48 hrs
Cumulative Events
Study Design
UFH
Enoxaparin
Days after randomization
Antom EM et al. NEJM, 2006
Hao Zhu, ACSMSBB 2008
Unfractionated Heparin (UFH)
In
use for ~ 40 years.
Efficacious?
–
No incentive for new trials.
ACC/AHA
10
recommend 48 hr infusion.
Hao Zhu, ACSMSBB 2008
Objectives of ‘Learning’ Analyses

Primary objective:
–

Secondary objective:
–
11
To investigate the effectiveness of UFH in treating
STEMI.
To evaluate the current UFH dosing regimen
recommended by ACC/AHA.
Hao Zhu, ACSMSBB 2008
Framing the Hypothesis
12

Typically to test a hypothesis, two groups
are compared ( e.g., UFH versus placebo).

Challenge: To investigate the effectiveness
using clinical observations from a single
treatment arm, because:
– No placebo group in the study
Hao Zhu, ACSMSBB 2008
Value Gained by Learning
0.95
0.85
0.90
•Confirmed the
significant
difference
between the two
arms.
*Event = all-cause mortality
and recurrent MI
0.80
Endpoints Free Survival
1.00
• Analysis variable
was changed from
cumulative events
to time to event.
0
13
200
400
600
Time to Events [hr]
• K-M curve
reproduced the
shift at 48 hr postdose
Hao Zhu, ACSMSBB 2008
Constructing Treatment Variable
Risk1
Treatment = 1
Risk2
Treatment = 0
Standard Patient
P1
P2
P3
P4
14
UFH Duration
Time
Hao Zhu, ACSMSBB 2008
Cox-Ph Model

Typically, Cox-ph model can be written as:
, Where i   p  j  xij
hi (t )  ei  h0 (t )
j 1

15
hi(t) is the hazard function for ith subject and h0(t) is the
baseline hazard function. j are the coefficients of the
explanatory variables. xijs are the explanatory variables,
which are constant over time.
Hao Zhu, ACSMSBB 2008
Cox-Ph Model with Time-Dependant
Effect

Cox-ph model with time dependent variables:
p
i ( t )
,
where

(
t
)

 j 1  j  xij (t )
i
h (t )  e
 h (t )
i


0
hi(t) is the hazard function for ith subject, h0(t) is the baseline
function, j are the coefficients of the explanatory variables. Xij
(t)s are the explanatory variables, which change the values
over time.
In our analysis, X1j(t) = trt, which represents the UFH treatment.
Where
–
–
–

16
trt = 1, when t ≤ Ti, and
trt = 0, when t > Ti,
(Ti represents the UFH treatment duration for the ith subject)
Data obtained from a single-treatment arm can be used for
analysis.
Hao Zhu, ACSMSBB 2008
UFH Is Efficacious
1.00
Treatment Duration Distribution
0.90
0.3
0.80
0.2
0.1
0
0
17
200
400
Time to Events [hr]
Fraction of Treatment Duration
Median Treatment Duration
0.70
Fraction of Subjects with no Events
Kaplan-Meier curve
• Kaplan-Meier curve shifted
at 48hr (median UFH duration
time).
• Cox-ph analysis using UFH
treatment as time dependent
variable demonstrated
significance (P <0.0001)
• Hazard ratio = 31,
suggesting an average
patient once stops UFH
treatment, the risk for events
increases 31 fold.
600
Hao Zhu, ACSMSBB 2008
18
0.95
P
Hazard
Ratio
0.90
≤ 83 hr
***
32.7
≤ 48 hr
***
25.3
≤ 17 hr
***
3.4
0.85
UFH
Duration
UFH K-M Curve
10th-90th Percentile of Heaprin Treatment time
Medain Heaprin Treatment Time
0.80
Endpoints Free Survival[%]
1.00
Sensitivity Analysis Supports UFH
Effectiveness
0
200
400
Time to Events [hr]
600
***: P < 0.0001
Hao Zhu, ACSMSBB 2008
Permutation Test Supports UFH
Effectiveness
• Observed HR
is unlikely to
occur by
chance.
Percentage [%]
30
Observed HR
Permutated HR
25
20
• Evidence of
UFH treatment
is robust
15
10
5
0
0.6
19
1.0
1.4
30.0
32.0
Hazard Ratio from Permutated Data and Observed Data
Hao Zhu, ACSMSBB 2008
Value of ‘Learning’ Analyses

UFH is efficacious.
–

Maximized knowledge extraction from a costly,
challenging, big clinical trial leading to labeling
> 48 hr UFH infusion is more beneficial.
–
–
Potential for improving UFH’s therapeutic value
Need for longer infusions in future studies

20
Potential implications on non-inferiority inferences
Hao Zhu, ACSMSBB 2008
Current Label
There is a trend in favor of enoxaparin during the first
48 hours, but most of the treatment difference is
attributed to a step increase in the event rate in the
UFH group at 48 hours (seen in Figure 2), an effect
that is more striking when comparing the event rates
just prior to and just subsequent to actual times of
discontinuation. These results provide evidence that
UFH was effective and that it would be better if used
longer than 48 hours. There is a similar increase in
endpoint event rate when enoxaparin was
discontinued, suggesting that it too was discontinued
too soon in this study.
21
Hao Zhu, ACSMSBB 2008
Summary
Pharmacometric analyses will
 reduce late phase clinical trial attrition
 select more rational dosing
22
Hao Zhu, ACSMSBB 2008
Are you interested in a fellowship with
FDA Pharmacometrics?

Fellowship1
–

Fellowship 2
–
23
QT-TdP: (1 ~ 1.5 years)
Contact Person: Dr. Christoffer W Tornoe
Email: [email protected]
Tel: (301) 796-2236
Alzheimer's Disease Model: ( ~ 2 years)
AAPS
Attention: CRADA – Alzheimer’s Disease Model
2107 Wilson Boulevard, Suite 700
Arlington, VA 22201
[email protected]
www.aapspharmaceutica.com/crada
Deadline for submissions: June 30, 2008
Hao Zhu, ACSMSBB 2008
Questions and Comments
24
Hao Zhu, ACSMSBB 2008