Interchangeability of multisource drug products containing highly

Download Report

Transcript Interchangeability of multisource drug products containing highly

Pre-qualification Program: Priority
Medicines
Interchangeability of Multi Source Drug
Products
Interchangeability of Multisource Drug Products
Containing Highly Variable Drugs
SALOMON STAVCHANSKY, PH.D.
ALCON CENTENNIAL PROFESSOR OF PHARMACEUTICS
THE UNIVERSITY OF TEXAS AT AUSTIN
COLLEGE OF PHARMACY
AUSTIN, TEXAS 78712
[email protected]
Kiev, Ukraine, June 25-27, 2007
Outline
• Background
– What is a highly variable drug?
– Present bioequivalence BE study approach
– Disadvantages of present approach
• Bioequivalence Example of Highly Variable Drugs
• Reference – scaled average BE approach
– Widening the bioequivalence limits
– scaling
• Simulation Studies
• Summary and Conclusions
Questions
• Why?
– Have you ever had or heard of a therapeutic failure
• Where do we want to be?
– No therapeutic failures and no adverse events
• What assumptions are we willing to make?
– Multisource products are interchangeable with brand
products
• How sure do you want to be?
– How to protect the consumer and the industry?
DRUG DEVELOPMENT PROCESS AND
REGULATIONS
CONTROLLED CLINICAL TRIALS
EFFICACY
SAFETY
THERAPEUTIC
INDEX
SAMENESS
CLINICAL
PHARMACOLOGY
P'KINETICS
BIOPHARMACEUTICS
VARIABILITY IN
DRUG RESPONSE
PRODUCT
QUALITY
P'DYNAMICS
DISSOLUTION
BA, BE
DOSE
ADJUSTMENT
FOR RISK
GROUPS
DOSE RANGE FOR
TARGET POPULATION
DOSE RANGE IN LABEL
DOSE
B aber, N .S., B r.J.C lin.Pharmacol.,42,545,1996
INDIVIDUALIZATION
Highly Variable Drug Characteristics
• Drugs with high within subject variability (CVwr)
in bioavailability parameters AUC and/or Cmax ≥
30%
• Non narrow therapeutic index drugs
• Represent about 10% of the drugs studied in
vivo and reviewed by the OGD-FDA
HVD Drug Products
• Highly Variable Drug Products in which the
drug is not highly variable, but the product
is of poor pharmaceutical quality
– High within-formulation variability
Variability Due to Drug Substance and/or Drug
Product
• Drug Substance
– Variable absorption rate, extent
– Low extent of absorption
– Extensive pre-systemic metabolism
• Drug product
– Formulation
• Inactive ingredient effects
• Manufacturing effects
– Effects of Bioequivalence Study Conduct
• Bioanalytical Assay Sensitivity
• Suboptimal PK Sampling
Summary of the issues
• High Probability that the BE parameters
will vary when the same subject receives a
highly variable drug on different occasions
• Because of high variability the risk is to
reject a product that in reality is
bioequivalent -- Industry Risk !
FDA Study to Characterize Highly Variable Drugs in BE
Studies: methods
• Collected data from 1127 acceptable BE studies,
submitted
– In 524 ANDAs
– From 2003-2005 (3 years)
• Most sponsors used 2-way crossover studies
– Used ANOVA Root Mean Square Error to estimate
within-subject variance
• Drug was classified as highly variable if RMSE ≥
0.3 or 30%
Source: Barbara M. Davit AAPS/FDA Workshop 5/22/2077, Rockville, MD
FDA Study to Characterize Highly Variable Drugs
in BE Studies: results
• BE studies of HVD enrolled more study
subjects than studies of drugs with low
variability
– Average N in studies of HVD = 47
– Average N in studies of drugs with lower
variability = 33
• Range 18 – 73 subjects
Source: Barbara M. Davit AAPS/FDA Workshop 5/22/2077, Rockville, MD
FDA Study to Characterize Highly Variable Drugs
in BE Studies: results
• 10% of studies evaluated were HVD; of
these:
– 52% of studies were consistently HVD
– 16% were borderline
• RMSE was slightly above or below 0.3
• Average across all studies
– For the remaining 32%, high variability
occurred sporadically
• Not HVD in most BE studies
Reasons for Inconsistent Variability in BE
Studies
•
•
•
•
•
•
Differences in formulations
Bioanalytical assay sensitivity
Demographic characteristics of subjects
Subjects with irregular plasma concentrations
Number of study subjects
Whether subjects were fasted or fed
Source: Barbara M. Davit AAPS/FDA Workshop 5/22/2077, Rockville, MD
Present FDA Approach for BE of HVD
• ANDAs for HVD use the same study
design for drugs with lower variability
• Two way crossover design
• Replicate study design
• Firms are encouraged to use sequential
designs
Present FDA Approach for BE of HVD
• HVD must meet same acceptance criteria
as drugs with lower variability
• 90% CI of AUC and Cmax test/reference
(T/R ratios) must fall within:
0.8-1.25 (80-125%)
• Statistical adjustment necessary if a
sequential study design is used
Is present FDA’s approach suitable for HVD?
Approach
Disadvantage
Enrolled adequate # of
Study may require larger N
subjects (N) to show BE in 2 If study underpowered
way crossover study
must do new study
Replicate design ( 4period) study
High dropout rate; may
need to enroll larger N
EXPENSIVE
Group sequential design Must specify in protocol
a priori
Statistical Adjustment
Source: Barbara M. Davit AAPS/FDA Workshop 5/22/2077, Rockville, MD
Background for NEW approach
ACPS Meeting, April 14, 2004: Discussion on
Highly Variable Drugs
• Different approaches were considered, e.g.,
expansion of bioequivalence limits, and
scaled average bioequivalence
• Committee favored scaled average
bioequivalence over other approaches
• FDA working group was created; a research
project to evaluate scaling was initiated
ACPS = Advisory Committee for Pharmaceutical Science
The Width of the 90% Confidence Interval
• The width depends on:
– Within subject variability WSV
– The number of subjects in the study
• The wider the 90% CI, the more likely it is
to fall outside the limits of 80-125%
• Highly variable drugs are a problem
90%CIs & BE Limits
• Green
– Low WSV (~15%)
– Narrow 90%CI
– Passes
• Red
–
–
–
–
High WSV (~35%)
Wide 90%CI
Lower bound <80%
Fails
125%
100%
80%
• GMR & the # subjects
same in both cases
Chlorpromazine:
ANOVA-CV%
ln Cmax
ln AUClast
aBioequivalence
Study 1a
Study 2b
Study 3c
42.3
34.8
39.9
36.6
37.2
33.0
study, n=37 (3-period study)
study n=11 (solution, 3-period study)
cPharmacokinetic study, n=9, CPZ with & without quinidine (2-period study)
bPharmacokinetic
Cmax
Ref-1
AUClast
Ref-1
Ref-2
Ref-2
6
16
20
6
7
13
6
13
7
16
13
13
7
16
20
20
27
27
27
27
7
6
20
16
Chlorpromazine (ABE3)
3 x 37 Subjects
Measure
ln Cmax
ln AUClast
ANOVA-2 (GLM)
GMR%
CV%
90%CI
115
110
42.3
34.8
99-133
97-124
Chlorpromazine: 90%CIs
Measure
T v R1
T v R2
ln Cmax 103 - 146 89 - 126
ln AUClast 97 - 128 94 - 125
ANOVA-1 (GLM)
R1v R2
72 - 102
85 - 112
Background
ACPS Meeting, October 6, 2006
 Preliminary results of simulation study were
presented
 Committee was in favor of using a point
estimate constraint with scaled average BE
 Most members favored a minimum sample size
of 24
ACPS = Advisory Committee for Pharmaceutical Science
Research Project
Highly Variable Drugs (HVD) working group
evaluated different scaling approaches and
study designs. Outcome:
– Scaled average bioequivalence, based on
within subject variability of reference*
*
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
Objective
Determine the impact of scaled average
bioequivalence on the power
(percent of studies passing) at different
levels of within subject variability (CV%),
and under different conditions.
Methods
Study design:
• 3-way crossover, e.g., R T R
• Sample sizes tested: 24 and 36
• Within subject variability: 15% - 60% CV
• Geometric mean ratio: 1 – 1.7
Methods
Variables tested:
• Impact of increasing within subject
variability
• Use of point estimate constraint (80125%)
• σw0: 0.2 vs. 0.25 vs. 0.294
• Sample size: 24 vs. 36
Methods
Statistical Analysis:
• Modified Hyslop model*
• Number of simulations: 1 million (106)/test
• Percent of studies passing was determined
using average bioequivalence (80-125% limits),
and scaled average bioequivalence (limits
determined as a function of reference within
subject variability)
• Test performed under different conditions
*Hyslop et al. Statist. Med. 2000; 19:2885-2897. Hyslop’s
model was modified by Donald Schuirmann
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
Impact of Within Subject
Variability
• 15% CV
• 30% CV
• 60% CV
Average vs. Scaled Average Bioequivalence
CV% = 15, Simulations = 106, N = 36, w0=0.25
Percent of Studies Passing
100
Scaled ABE + Point Estimate
Average BE
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
1.5
1.6
1.7
Average vs. Scaled Average Bioequivalence
CV% = 30, Simulations = 106, N = 36, w0=0.25
Percent of Studies Passing
100
Scaled ABE + Point Estimate
Average BE
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
1.5
1.6
1.7
Average vs. Scaled Average Bioequivalence
CV% = 60, Simulations = 106, N = 36, w0=0.25
Percent of Studies Passing
100
Scaled ABE + Point Estimate
Average BE
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
1.5
1.6
1.7
Impact of Point Estimate Constraint
• Lower variability (30% CV)
• Higher variability (60% CV)
Average vs. Scaled Average Bioequivalence
CV% = 30, Simulations = 106, N = 36, w0=0.25
Percent of Studies Passing
100
Scaled ABE
Point Estimate
Scaled ABE + Point Estimate
Average BE
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
1.5
1.6
1.7
Average vs. Scaled Average Bioequivalence
CV% = 60, Simulations = 106, N = 36, w0=0.25
Percent of Studies Passing
100
Scaled ABE
Point Estimate
Scaled ABE + Point Estimate
Average BE
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
1.5
1.6
1.7
Impact of σW0
σW0 = 0.2
σW0 = 0.25
σW0 = 0.294
Impact of w0 on the Power
CV% = 30, Simulations = 106, N = 36
Percent of Studies Passing
100
Average BE
Scaled + Point Estimate (w0 = 0.2)
Scaled + Point Estimate (w0 = 0.25)
Scaled + Point Estimate (w0 = 0.294)
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
1.5
1.6
1.7
Impact of w0 on the Power
CV% = 60, Simulations = 106, N = 36
Percent of Studies Passing
100
Average BE
Scaled + Point Estimate (w0 = 0.2)
Scaled + Point Estimate (w0 = 0.25)
Scaled + Point Estimate (w0 = 0.294)
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
1.5
1.6
1.7
Average vs. Scaled Average Bioequivalence
CV% = 60, Simulations = 106, N = 36 vs. 24, w0=0.25
Percent of Studies Passing
100
Scaled ABE + Point Estimate (N = 24)
Average BE (N = 24)
Scaled ABE + Point Estimate (N = 36)
Average BE (N = 36)
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
1.5
1.6
1.7
Impact of Different Point
Estimate Constraints
• Point estimate constraint = ±15%
• Point estimate constraint = ± 20%
Scaled + Point Estimate constraint of 15%
Simulations = 1000000, N = 24
100
CV 30%
CV 40%
CV 50%
CV 60%
Percent of Studies Passing
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
1.5
1.6
1.7
Scaled + Point Estimate constraint of 20%
Simulations = 1000000, N = 24
100
CV 30%
CV 40%
CV 50%
CV 60%
Percent of Studies Passing
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
1.5
1.6
1.7
Summary
 Partial replicate, 3-way crossover design
appears to work well
 A point estimate constraint has little impact at
lower variability (~30%); more significant
effect at greater variability (~60%)
 A σW0 = 0.25 demonstrates a good balance
between a conservative approach, and a
practical one
Conclusion
 Scaled ABE presents a reasonable option for
evaluating BE of highly variable drugs
 Practical value, reduction in sample size:
Potentially decreasing cost and unnecessary
human testing (without increase in patient
risk)
 Use of point estimate constraint addresses
concerns that products with large GMR
differences may be judged bioequivalent
FDA Proposal*:
Scaled Average BE for HVA Drugs
• Three-period, partial replicate design
– Reference product (R) is administered twice
– Test product (T) is administered once
– Sequences = RTR, TRR, RRT
• Sample size: Determined by sponsor
(adequate power)
– minimum is 24 subjects
* Currently under evaluation
Source: Sam H. Haidar: AAPS/FDA Workshop 5/22/2077, Rockville, MD
FDA Proposal- continued
• BE criteria scaled to reference variability (Cmax
& AUC)
 0.223

BE limits, upper, lower  EXP  
 σ wr 
 σ w0

– Where σw0 = 0.25
• The point estimate (test/reference geometric
mean ratio) must fall within [0.80-1.25]
• Both conditions must be passed by the test
product to conclude BE to the reference product
Use of reference average BE for HVD
• BE criteria scaled to reference variability
• 90% upper confidence bound for:
Ho: (µT-µR)2 – θ σ2wr must be ≤ 0
Where θ = scaled average BE limit
and
θ = (ln Δ)2/ σ2wo
Where σwo = 0.25
Use a point estimate constraint
Both Cmax and AUC must meet criteria
Advantages of scaled BE
reference scaled
• Test product will benefit if:
– T variability < R variability
• The test product will not benefit if:
– T variability > R variability
Concerns with Proposed Approach
• Firms will conduct a replicate design study and
submit results to FDA
– If within subject variability ≥ 30%, FDA will use the referencescaled average BE approach
– If within subject variability ≤ 30%, FDA will use the unscaled
average BE approach
• What if the drug is characterized as a borderline HV
drug?
– FDA simulations showed that study outcome will be the same
whether the scaled or unscaled approach is used
• Scaling can allow AUC and Cmax GMR to be
unacceptably high or low
– Acceptance criteria will include a point estimate constraint
Concerns with Proposed Approach
• What if high variability results from
formulations problems or poor study
conduct?
– If T variability > R variability, no benefit in
using scaled approach
– The burden is on the applicant to convince
FDA that product is a HVD
Impact Of Test CV% On Study Power Using Scaled Average BE
CVRef% = 30, Simulations = 106, N = 24, w0=0.25
100
CVTest 30%
CVTest 40%
CVTest 50%
CVTest 60%
Percent of Studies Passing
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
1.5
1.6
1.7
Impact Of Small Sample Size On Study Power Using Scaled Average BE
CVRef% = 30, CVTest% = 30, Simulations = 10 6, w0=0.25
100
n = 24
n = 21
n = 18
n = 15
n = 12
Percent of Studies Passing
80
60
40
20
0
1.0
1.1
1.2
1.3
1.4
Geometric Mean Ratio
1.5
1.6
1.7
Thank you
спасибо