Transcript Multi-Region Trials

Multiregional Trials
Main features and issues raised
Byron Jones
Novartis
PSI Conference, May 13, 2014
Multiregional Clinical Trials
Definition and Motivation
 A single clinical trial that is conducted simultaneously in
multiple geographical regions under a common protocol
 Increasingly, clinical trials are run using patients from
various regions worldwide.
• More patients needed to demonstrate treatment advantages, as new
treatments may have only incremental benefits vs existing therapies.
• Local health authorities would like to see representation / evidence
within their domains.
• Varied settings may enhance confidence in observed effects.
• Expanded markets interest trial sponsors.
3 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Multiregional Clinical Trials
Advantages
 Advantages (Ando, Y. , ICSA/ISBS Conference, 2013)
• Prevents unnecessary duplication of clinical trials
• Makes drug development more efficient and cost-effective
• Enables simultaneous global drug submission and approval
• Gets effective and safe drugs to patients faster.
4 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Example: multiregional trial
679 study centres
5 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Example of an MRCT
 7,216 patients were enrolled from 5 geographical regions,
39 countries and 679 study centers.
 3,581patients were randomized to the drug group and
3,635 to the placebo group.
Number of
countries
N
Drug
(n)
Placebo
(n)
Treatment
difference
Standard
error
P-value
Asia
5
441
214
227
-6.97
1.675
<.0001
Europe
20
3819
1889
1930
-5.43
0.531
<.0001
Latin America
9
1229
630
599
-3.96
0.991
<.0001
North America
2
1525
750
775
-4.93
0.829
<.0001
Other
3
202
98
104
-3.18
2.258
0.16
Global
39
7216
3581
3635
-5.10
0.391
<.0001
Region
6 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Regulatory Guidance
ICH E5 and ICH E5 Q&A
A11. A multi-regional trial ...
The objectives of such a study would be:
(1) to show that the drug is effective in
the region and
(2) to compare the results of the study
between the regions with the intent of
establishing that the drug is not
sensitive to ethnic factors.
7 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
What is a region?
8 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
What is a region?
Is it based on geography?
9 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
What is a region?
Not necessarily defined by location
 “... region should not be limited to geographic boundaries
but should take into consideration relevant intrinsic genetic
and physiological or pathological factors as well as
extrinsic factors such as medical practice” ICH E5.
10 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Inconsistency in the definition of regions
Review of 60 FDA Advisory Committee Meetings 2008-2010
 90% of submissions were multiregional.
• “Region was most often defined based on geography, and specifically
continent ...”
 “No trends or consistency was observed in how regions
were defined within or across therapeutic areas nor any
rationale for the definition of region ...”
 “We propose that adequate justification of the definition
should take into consideration factors such as race or
ethnicity, disease epidemiology, medical practice, and
geographic proximity, among others.”
Tanaka et al. (2011)
[The PhRMA MRCT Key Issue Team]
11 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
PhRMA MRCT KIT Perspective on Region
 “... regions should be predefined in the design stage and
properly documented.”
 “... Geography alone may not be adequate when defining
regions. ... Intrinsic and extrinsic factors should be
considered.”
 “ Country and site selection should be considered at the
design stage as part of predefining regions...”
 Analytical approach to defining regions (e.g., factor
analysis, principal components).
 The number of regions should not be large.
12 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Analysis models for multiregional clinical trials
Fixed or Random effects for regions/centres
Recall: Multicentre Trials
 Fixed-effects Model [centre is a fixed factor]
– The centers have been specifically chosen. Conclusions reached here only
apply to the centers considered and can not be extended to other centers
that are not in the trial
 Random-effects Model [centre is a random factor]
– The centers are a random sample from a large population of centers.
Conclusions reached here can be extended to all the centers in the
population
13 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
For MRCTs: Are regional estimates fixed or
random?
 Surely “region” is a fixed-effect – cannot think of a random
sample of regions?
 Possible model might assume centres are randomly
nested within the levels of a fixed regional factor.
 However, this are differing opinions in the literature.
14 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
MTCT: fixed or random effects?
Random: Chen, Hung and Hsiao (2012)
 Chen, Hung and Hsiao (2012) define a random effects
model for the true treatment difference that applies to
region i, i=1,2, ..., M.
 They derive the global estimate of the treatment difference
by applying well-known results for the random-effects
estimator obtained from a meta-analysis, using the
DeSimonian and Laird (1986) estimator of the between
region variance.
 Give sample size formula based on global estimate.
15 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Shrinkage estimates of regional treatment effects
Qui et al. (2013). Statistics in Medicine
 Recommend :
• fixed-effects model to estimate global effect and
• estimates of individual region treatment differences using an
empirical shrinkage estimator based on a random effects model
• Individual region estimates borrow strength from other regions’
estimates.
16 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Consistency
Are individual region estimates similar to the global estimate?
 It is (or should be) a basic premise of an MRCT that that
there is no, or at most only a small amount of, regional
variation
• Regional variation can be reduced by good design and by inclusion
of region-specific covariates in models for the response.
 Should testing for such consistency be part of the analysis
plan?
• Sponsor more interested in global estimate
• Regulator more interested in local estimate for their region
• Ideally, global estimate of treatment difference is significantly
different from zero and all regional estimates are significantly
different from zero.
• Sample size implications are different for the two situations.
17 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Two well-known examples of possible
inconsistency
 PLATO trial
 MERIT trial
18 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
PLATO Trial
PLATlet inhibition and patient Outcomes trial (Wallentin, et. al., 2009)
 Compare ticagrelor (novel) vs clopidogrel (standard)
 Patients with ACS (acute coronary syndomes)
 Primary endpoint: CV death, MI, stroke
 18624 patients, followed for a year.
Endpoint ticagrelor copidogrel
HR
P-value
Primary
9.8%
11.7%
0.84
<0.001
Death
4.5%
5.9%
0.78
<0.001
 Very strong evidence that ticagrelor is superior.
 BUT...
19 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
PLATO trial
“Ticagrelor works, except if you’re an American” – Stuart Pocock (LSHTM)
20 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
PLATO trial
“Ticagrelor works, except if your an American” – Stuart Pocock (LSHTM)
21 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
PLATO: A chance result?
 Given 31 subgroup analyses were done, can this
significant interaction be due to chance alone?
 The chance of a “reversal” in sign of estimated treatment
difference is not negligible.
 But “region” is a special subgroup and will be of interest to
US regulators (FDA).
 Can this “chance” finding be explained?
 Is it caused by the Aspirin (ASA) loading dose and longterm maintenance dose that patients received on day of
randomization to treatment?
22 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Does Aspirin use explain the interaction (?)
23 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
MERIT-HF trial
Metoprolol Controlled –Release Randomised InterventionTrial in Heart
Failure
24 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
MERIT-HF trial in heart failure
Overall results
Endpoint
Death
Metoprolol
Placebo
Sample
size
1990
2001
Total
deaths
145
217
25 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
HR
P-value
0.66
0.00009
95% CL
Lower
limit
95% CL
Upper
limit
MERIT-HF trial in heart failure
Make USA a subgroup
Endpoint
Death
Metoprolol
Placebo
Sample
size
1990
2001
Total
deaths
145
217
USA
51
49
Other
countries
94
168
Interaction test: P = 0.003
26 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
HR
P-value
95% CL
Lower
limit
95% CL
Upper
limit
1.05
0.71
1.56
0.55
0.43
0.70
0.00009
Break out deaths by country and treatment
country
metoprolol
placebo
Hungary
16
29
Germany
19
31
Netherlands
14
25
Belgium
3
13
Czech Republic
9
17
Sweden
2
9
Norway
6
11
UK
4
9
Finland
0
2
Switzerland
0
1
Iceland
2
2
Poland
8
8
Denmark
11
11
USA
51
49
27 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Why concentrate
Interaction test
on USA?
Break out deaths by country and treatment
country
metoprolol
placebo
Hungary
16
29
Germany
19
31
Netherlands
14
25
Belgium
3
13
Czech Republic
9
17
Sweden
2
9
Norway
6
11
UK
4
9
Finland
0
2
Switzerland
0
1
Iceland
2
2
Poland
8
8
Denmark
11
11
USA
51
49
28 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Why focus on USA?
Unlike the PLATO
Trial, there seem no
reason to believe
Interaction is real
Are the other/better methods to test for
consistency?
 Quan et al. (2010b) proposed 5 alternative methods.
 These are of two types:
• Methods that tend to conclude consistency until there is sufficient
evidence to the contrary, e.g., interaction tests
• Methods requiring a certain strength of signal of similarity in order to
conclude consistency, e.g., Japanese MHLW proposals
 Which type is appropriate for a given situation?
 Where should the burden of evidence lie?
29 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Five methods to test for consistency
Quan et al. (2005)
1. Each region should achieve a proportion, π, of the
observed overall effect.
2. Each region should achieve a common pre-specified
constant value (b ≥ 0).
3. Demonstrate through hypothesis testing that each region
achieves a proportion, π, of the overall effect.
4. A test for treatment-by-region interaction must not yield a
significant result.
5. Tests for individual regions having effects lower than the
overall effect must all not yield significant results.
30 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Difficulties in implementation of methods: Method 1
Quan et al. (2010a)
 Trial planned to have 90% power, to detect a one-sided
difference between two treatments with significance level
0.025.
 Consider a single region (e.g., Japan) out of the set of
regions.
 Let 𝐷𝐽 be the estimated treatment effect in Japan and
𝐷𝐴𝑙𝑙 be the estimated effect over all regions
D𝐽
 Require Pr(𝐷
𝐴𝑙𝑙
> 0.5) ≥ 0.8
 If all treatment effects truly equal in all regions
• Sample size fraction for Japan = 22.4%
• Too high for a country with only 2% of the world population
31 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Conclusions
 Statistical methodology for MRCTs is still evolving
 Experience over time will determine acceptable methods
 Issues relate to conflict in the desire to estimate a global
effect versus a local (single region) effect.
 Regulatory agency involvement can focus attention on a
single region with unwanted consequences (e.g., for Type
I error rate control, effect reversal, etc. ) familiar to users
of subgroup analysis.
 Definition of “a region” needs to be clarified
32 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
References
 Ministry of Health, Labour and Welfare of Japan (2007). Basic Concepts for
Joint Clinical Trials.
 Chen, C-T, Hung, H.M.J and Hsiao, C-F, (2012). Design and evaluation of
multiregional trials with heterogeneous treatment effects across regions.
Journal of Biopharmaceutical Statistics, 22, 1037-1050
 Chen, J., et al. (2011). Consistency of treatment effects across regions in
multiregional clinical trials, Part 1: design considerations. Drug Information
Journal, 45, 595-602.
 Gallo, P., et al. (2011). Consistency of treatment effects across regions in
multiregional clinical trials, Part 2: monitoring, reporting and interpretation. Drug
Information Journal, 45, 603-608.
 Kawai,N., et al. (2008). An approach to rationalize partitioning sample size into
individual regions in a multiregional trial. Drug Information Journal, 42, 139-147.
33 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
References
 Quan, H., et al. (2010a). Sample size considerations for Japanese patients in a
multiregional trial based on MHLW guidance. Pharmaceutical Statistics, 9, 100112.
 Quan, H., et al. (2010b). Assessment of consistency of treatment effects in
multiregional clinical trials. Drug Information Journal, 44, 617-632.
 Tanaka, Y. (2011). Points to consider in defining a region for a multiregional
clinical trial: defining region workstream in PhRMA MRCT Key Issue Team.
Drug Information Journal, 45, 575-585.
 Wendel, H., et al. (2001). Challenges of subgroup analyses in multinational
clinical trials: experiences from MERIT-HF trial. American Heart Journal, 142,
502-511.
34 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
END OF PRESENTATION
35 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Models for MRCT data
Extensions of models for multicentre clinical trials (MCTs)
 Suppose we have two treatment groups, j = 1 denotes the
active treatment arm, and j = 2 denotes the placebo arm.
 Let 𝑦𝑖𝑗𝑘 denotes the response on patient k in treatment 𝑗 and
center 𝑖. Where j = 1, 2; 𝑖 = 1,2, ⋯ , 𝑐 and 𝑘 = 1,2, ⋯ , 𝑛𝑖𝑗
 The full model is
𝑦𝑖𝑗𝑘 = 𝜇 + 𝜏𝑖 + 𝛽𝑗 + 𝛾𝑖𝑗 + 𝜀𝑖𝑗𝑘
– 𝜏𝑖 is the effect of center 𝑖
– 𝛽𝑗 is the effect of treatment 𝑗
– 𝛾𝑖𝑗 is the interaction effect of center 𝑖 and treatment 𝑗
– 𝜀𝑖𝑗𝑘 ~𝑁(0, 𝜎 2 )
36 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Model II: 𝑦𝑖𝑗𝑘 = 𝜇 + 𝜏𝑖 + 𝛽𝑗 + 𝜀𝑖𝑗𝑘
 Assumption: With center effect, no treatment-by-center
interaction
 Overall treatment effect: 𝛿 = (𝛽1 − 𝛽2 )
 Estimator (weighted): ∆𝐼𝐼 =
𝑤𝑖 =
(𝑉𝑎𝑟(𝛿𝑖 ))−1
𝑐 (𝑉𝑎𝑟(𝛿 ))−1
𝑘
𝑘=1
 𝐸 ∆𝐼𝐼 = 𝛿, 𝑉𝑎𝑟 ∆𝐼𝐼 = 𝜎 2 [
𝑐
𝑖=1 𝑤𝑖 𝛿𝑖
1
1
𝑖1
𝑖2
where
= (𝑛 + 𝑛 )−1
1
𝑐
(
𝑖=1 𝑛
𝑖1
+
1
𝑐
(
𝑘=1 𝑛
𝑘1
1
+ 𝑛 )−1
𝑘2
1 −1 −1
) ]
𝑛𝑖2
 Weights for each center are chosen according to the precision
of the within-center estimate
37 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Model III: 𝑦𝑖𝑗𝑘 = 𝜇 + 𝜏𝑖 + 𝛽𝑗 + 𝛾𝑖𝑗 + 𝜀𝑖𝑗𝑘
 Assumption: With center effect and treatment-by-center
interaction
 Overall treatment effect:
𝛿𝑖
𝑐
= (𝛽1 −𝛽2 ) +
1
 Estimator (equally weighted): ∆𝐼𝐼𝐼 = 𝑐
 𝐸 ∆𝐼𝐼𝐼 =
𝛿𝑖
, 𝑉𝑎𝑟 ∆𝐼𝐼𝐼 =
𝑐
𝜎2
𝑐2
1
𝑐
(
𝑖=1 𝑛
𝑖1
+
(𝛾𝑖1 −𝛾𝑖2 )
𝑐
𝑐
𝑖=1 𝛿𝑖
1
)
𝑛𝑖2
 All centers are considered having the same importance thus
receive the same weights
 Model III is to be used when the treatment differences vary
substantially from center to center
38 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Framework for Random Effects Models for MCTs
Fedorov and Jones (2005)
 Suppose we have two treatment groups, j = 1 denotes the
active treatment arm, and j = 2 denotes the placebo arm.
 Let 𝑦𝑖𝑗𝑘 denotes the response on patient k in treatment 𝑗 and
center 𝑖. Where j = 1, 2. 𝑖 = 1,2, ⋯ , 𝑐 and 𝑘 = 1,2, ⋯ , 𝑛𝑖𝑗
 The full model is
𝑦𝑖𝑗𝑘 = 𝜇𝑖𝑗 + 𝜀𝑖𝑗𝑘 where
– 𝜇𝑖𝑗 are random and for 𝜇𝑖 = [𝜇𝑖1 , 𝜇𝑖2 ]𝑇 : 𝐸 𝜇𝑖 = [𝜇1 , 𝜇2 ]𝑇 ,
𝐶𝑜𝑣 𝜇𝑖 = 𝜎 2 Ʌ
– 𝜀𝑖𝑗𝑘 ~𝑁(0, 𝜎 2 )
 Overall treatment effect 𝛿 = 𝜇1 − 𝜇2
39 | PSI Conference 2014 | Multiregional Clinical Trials | Byron Jones
Heterogeneity in multiregional studies and a
new proposal for exact tests on interaction
Joachim Röhmel
University of Bremen
Reasons for regional differences can be
manyfold
•
•
•
•
•
•
•
•
•
•
Genetic sensitivity
Culture
Dose regimen
Application scheme
Disease epidemiology
Disease definition
Economic standing
Health care system
Medical practice
Regulatory environment
• Quality of trial conduct
• Availability of concomitant
medicines
• Evaluation of outcomes (in
particular in composite
endpoints)
• Insufficient standardisation
and validation of scores
(East Europe)
• Patient compliance
From Pocock et al.
PLATO TRIAL 2011
Estimated treatment effects by geographic region for the primary endpoint (CV death,
MI, or stroke) of the PLATO trial (hazard ratios with 95% CIs, interaction P-value <0.05).
Conclusions of the FDA statistical review (Sep 2010)
• From the additional analyses, we continue to be troubled by
the qualitative interaction between the region (US versus nonUS) and treatment.
• In our view, neither play of chance nor concurrent use of ASA
provides a satisfactory explanation for the US versus non-US
disparity observed in this trial.
• Even though multiple factors have been screened for potential
causes, the question remains unsolved.
Conclusions of the FDA statistical review (Sep 2010)
• The disparity can still be caused by the difference in standard
medical practice between US and the rest of the world, which
is hard to quantify and has not been quantified.
• We ought to seek further data to either confirm or dismiss this
disturbing finding.
• Without the data, we would recommend that this drug not be
approved.
• Another study should be required if this drug is to be approved
for use in US.
Pockock‘s conclusions
• In the PLATO trial, the between-region comparison was one
of 32 pre-planned subgroup analyses, and hence purely by
chance one could expect one or two such analyses to have
interaction P0.05.
• Furthermore, post hoc emphasis on the most striking subgroup
finding (geography, in this case) means that even if the finding
is not entirely due to chance, the observed data are prone to
exaggerate any true disparities (between regions).
• Alternatively, one can assess all 43 countries separately, and
the global interaction test for heterogeneity among the 43
hazard ratios yields P =0.95.
FDA APPLICATION NUMBER:022560Orig1s000
• The study center effect was statistically significant in the main
effect ANCOVA model. This indicates potential heterogeneity
of efficacy responses across the 6 centers.
• …
• The mean percent change from baseline BMD in lumbar spine
was ranging from
– 2.5% in the US/Canada
( 139 subjects),
– 3.1% in Hungary
( 90 subjects),
– 3.2% in Argentina
(222 subjects),
– 3.2% in France and Belgium ( 64 subjects),
– 3.8% in Poland
(147 subjects), and
– 3.9% in Estonia
( 140 subjects) .
• Results of subgroups analyses are not powered to draw any
meaningful statistical conclusion, mainly due to small number
of subjects in subgroups.
R.T.O‘Neill, May 28, 2009
R.T.O‘Neill, May 28, 2009
R.T.O‘Neill, May 28, 2009
Social Court of the Berlin-Brandenburg
Reference number: L 1 KR 140/11 KL Dec 6, 2011
• Company complains against Escitalopram being merged with all
other (generic) SSRIs, which means low reimbursement
– Company wins first stage battle in court
• Health Insurance replies (actually based on IQWiG arguments) :
– The results of the Yevtushenko study (2007) (conducted
solely in Russia) lie extraordinarily above the estimates of the
other studies. Comparability is therefore critical.
– Furthermore, the applicability of study results may not be
given in the context of German patient care. Generally, it is
necessary to take stronger regard to cultural aspects in
depression.
How do we define region?
How do we define consistency?
K. J. Caroll, AstraZeneca, 2011
What constitutes a region?
• America
– North
– Latin
– South
• Europe
– North
– East
– South
• Asia
– China
– India
– Japan
– South-East
• By Country?
• Significance of Interaction
often disappears when 3 or
more regions are included
(Caroll, 2011)
Common criteria (Quan et al. DIJ, 2010)
1. Achieving in each region a proportion of the observed
overall effect
2. Observing in each region an effect above a certain threshold
3. Tighten 1. by subsituting the lower limit of CIs instead of the
observed values
4. Absense of statistical significance in interaction tests, usually
at significance levels >>0.05
5. Lack of clinically significant differences from the overall
Strong Interaction
Strong Interaction may look less impressive
when splitting one category into two
For exploratory subgroup analyses including
regional subgroups
Baysian methods may be useful
(Penello , BASS Conference 2013)
(Penello , BASS Conference 2013)
An alternative conditional (permutation)
approach to interaction
full population decomposes into k 2x2 tables
Events
~Events
Totals
Treat R
Treat T Totals
aR1
aR0
aT1
aT0
s
f
nR
nT
n
stratum/subgroup k
stratum/subgroup1
Events
~Events
Totals
Treat R
Treat T Totals
a1R1
a1R0
n1R
a1T1
a1T0
n1T
s1
f1
n1
…
Treat R
Treat T
Totals
Events
akR1
akT1
sk
~Events
akR0
nkR
akT0
NkT
fk
nk
Totals
S11 n1R
F10
n1T
a1R1
a1R0
a1T1
b1T0
ZELEN:
S21 n
2R
F20 a2R1 n
2T
a2R0
a2T1
a2T0

F10
Sk1 nkR
akR1 n
kT
akR0
akT1
akT0
VR1
VR0
VT1
VT0
Search for all configuations
such that summation over
Columns (successes,
failures),
Rows (numbers randomised
to R or T) and strata (e.g.
regions)
gives identical results.
n1R
n1T
a1R1
a1R0
a1T1
New proposal:
a1T0
a2R0
n2R
a2R1 n
2T
a2T1
a2T0

nkR
akR1
akR0
nkT
akT1
akT0
VR1
VR0
VT1
VT0
Search for all configuations
such that summation over
Columns (successes,
failures),
Rows (numbers
randomised to R or T) per
each Stratum (e.g.
regions)
and the total numbers of
successes (and failures)
gives identical results
“Zelen/BreslowDay test (Z/BD)“ versus
“less restricted permutations (LRP)“
Z/BD | LRP
Column totals per each stratum constant


Total no of Events in Treat R (sum over all
strata) constant


Total no of Events in Treat T (sum over all
strata) constant


Row totals per stratum constant

---
Zelen‘s exact conditional approach
Ω  { τ τ  τ ijl , i  1,..., k ; j  R , T ; l  0 ,1; all margins fixed }
p  value ( τ ) 

prob ( σ )
σΩ
prob ( σ )  prob ( τ )
 n iR    n iT    N i 
 
 / 

 
i  1 b iR 1

  b iT 1   S i 1 
k
prob (  ) 
 n iR    n iT    N i 
 
 / 

   
i  1 b iR 1

  b iT 1   S i 1 
k
Breslow/Day conditional extention
(StatXact 10)
Zelen‘s set  of configurations 
p  value ( τ ) 

prob ( σ )
σΩ
2
2
χ BD ( σ )  χ BD ( τ )
!!IMPORTANT!!
„Independently ordering the sample space“ is the
principle that allows developing more exact tests
for discrepancy based on other effect measures
such as RR or  or …
Extention allows for a variety for measuring
effects in the full population
• Difference
 = (a1  m1) - (a2  m2)
• Relative risk
RR = (a1  m1)  (a2  m2)
• Odds ratio
OR = (a1 ( m1-a1))  (a2  (m2-a2))
and further measures of discrepency
• Difference
–
–
–
Difference of Differences
CI for interaction (Newcombe 1998)
k=2: discepency = subgroup 1 - subgroup 2
k>2: discepency = ij (subgroup i - subgroup j)2
• Relative risk
–
–
Ratio of relative risks
RRdiscepency =log [RRsubgroup 1/RRsubgroup 2]
RRdiscepency = ij (log [RRsubgroup i /RRsubgroup j])2
• Odds ratio
–
Zelen‘s test for homogeneity of odds ratios
Breslow-Day test
Situation investigated – typical for Biomarker + /• Binomial trials with two strata
Success
probs
Treatm1
Treatm2
Biomarker -
p1
p1
Biomarker +
p1
p2
Power to detect an interaction, when for treatm1 the prob for sucess is 0.1 in both
strata while for treatm2 the prob for success in stratum 1 is 0.1 and in stratum 2 is q.
sample sizes are n=100 per treatm and per stratum
1
0.9
0.8
Power
0.7
0.6
Rel
0.5
OR
Zelen
MaxAbs
0.4
Cochran Q
0.3
0.2
0.1
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
probability q of success in stratum 2, treatm 2
0.8
0.9
1
Power to detect an interaction, when for treatm1 the prob for sucess is 0.1 in both
strata while for treatm2 the prob for success in stratum 1 is 0.1 and in stratum 2 is q.
sample sizes are n=100 per treatm and per stratum
1
0.9
Power
0.8
0.7
Rel
0.6
OR
Zelen
MaxAbs
0.5
Cochran Q
0.4
REL
MAX
0.3
ODDS
0.2
0.1
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
probability q of success in stratum 2, treatm 2
0.8
0.9
1
More details for three or more regions
 This is an issue for the next
lesson Thank you very much
for your attention
• ( I propose to study this in the next lesson)
• Thank you very much for your attention
Multi-Regional Clinical Trials
from the Japanese viewpoint
- Current practice and future challenges Tomokazu Inomata
Novartis, Japan
PSI, May 13th, 2014
“Disclaimer : The views and opinions expressed in the following slides are
those of individual presenter and should not be attributed to Novartis”
Motivation
 MRCT has become most popular approach for efficient drug
development in Japan
 Simultaneous global drug development / submission is our
(=Novartis Japan) standard approach
 The Japanese regulatory guidance, “Basic Principles on Global
Clinical Trials (GCTs)” was issued in 2007 (Note : GCT = MRCT)
 Many discussions/conferences regarding the guidance
(Japanese sample size, consistency, ..) were held ・・・
 A good opportunity to share our experiences of MRCTs
- Statistical approach vs. Practical / Feasible approach
81 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Outline
 Trend in Japan drug development
• Relevant guidance for MRCTs
• Characteristics of MRCT
 Current practice
• An example
• Possible considerations and future challenges
 Concluding remarks
82 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Trend in Japan drug development
Relevant Guidance for MRCTs
 1998: ICH-E5
“ETHNIC FACTORS IN THE ACCEPTABILITY OF FOREIGN CLINICAL DATA”
• “Bridging study” was accepted for “resolving Drug Lag”
• Intended to extrapolate foreign data (PIII) to the Japanese population and is generally
conducted as a dose-finding study in Japanese subjects
• Many discussions on how to evaluate “Similarity” of dose-finding profile between Japan
population and Non-Japan population.
 2003: ICH-E5 Q&A’s
2006: ICH-E5 Q&A’s (R1) - Points to Consider on MRCT was added –
Japan Local guidance
 2007: Basic Principles on Global Clinical Trials
• “MRCT” was encouraged for “efficient and rapid drug development”
• Many discussions on how to evaluate “Consistency” between Japan population and
Non-Japan population.
 2012: Basic Principles on GCTs (Ref. Cases / 17 Q&A’s)
 2013: Guideline on Data Monitoring Committees
83 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Number of Approvals in Japan
- Bridging studies vs. MRCTs -
20
2006
ICH-E5
16
Q&A(R1)
18
1998
ICH-E5
2012
“Basic Principles
on Global Clinical
Trials (Reference
Cases)
2007
Basic Principles
on Global
Clinical Trials
14
12
# of
10
Approval
Bridging
8
MRCT
6
4
2
0
2006
2007
2008
2009
Year
2010
2011
2012
In 2010, more than 20% of the total
clinical trials were MRCTs
84 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business
Use Only
(Source:
Ando,Y., et al, 2012)
“Basic Principles on Global Clinical Trials”
Basic principles for design and
conduct of GCTs are provided in
the 12 Q&As
Q 1: Basic requirements
Q 2: Timing for Japan to participate
Q 3: Phase I trial or PK information
Q 4: Dose-finding study in Japanese
Q 5: Basic points to consider in designing
Q 6: Sample size determination
Q 7: Primary endpoint / evaluation index
Q 8: Position of separate domestic study
Q 9: Control arm
Q10: Concomitant medication / therapy
Q11: Recommended disease area
Q12: Decision chart for joining MRCTs
85 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Basic Principles on Global Clinical Trials
- Some of Key features -
 “Region”
• No clear definition in the Guidance,
- But consider Japan as one region
• Need a reasonable definition of “region” in each study considering
medical practice, guideline or drugs for the disease, etc.
 “Consistency”
between Japan and the entire study population
• No clear definition in the Guidance
• Need a justification as to why the entire population can be deemed as
one population.
 “Sample Size & Proportion of Japanese subjects”
• No established recommendable method....
- But two specific methods for calculating sample size are introduced
86 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Two Methods for calculating sample size
As an example a placebo-controlled study using quantitative endpoints
 Method 1
 Method 2
Probability of “DJapan / Dall > 0.5“ is
80% or higher
Probability of “Each of Di exceed
0” is 80% or higher
• Condition
• Condition
- D: Difference between the placebo
group and test drug group
- Dall: Difference in the entire population
- Di : Difference between the placebo
group and test drug group in each
region i
- DJapan: Difference in the Japanese
population
 The probability tends to increase if equal
 Minimizing the Japanese sample size
increases the total sample size
 Minimizing the total sample size increases
number of subjects is enrolled from each
region.
 The Japanese sample size can be set
without changing the total sample size
the |Multi-Regional
Japanese
sample
size.Inomata | May13 2014| Subject | Business Use Only
Clinical
Trials| Tomokazu
87
Characteristics of MRCTs in Japan
Based on 31 approved drugs between 2006 and 2013 (excluding Oncology)
 Study Phase
PIII
• PIII: 29 (94%)
PII
• PII: 2 (6%)
- Not many dose response studies were conducted as MRCTs
- But, number of MRCTs in earlier stage has been increased
 Region
• Asian trials: 13 (42%)
Asian
Global
• Global trials*: 18 (58%)
- No harmonized definition of “Asian region”
* Global trials include Asian countries
Source : PMDA review reports http://www.pmda.go.jp/
88 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Characteristics of MRCTs in Japan
Based on 31 approved drugs between 2006 and 2013 (excluding Oncology)
 Sample Size determination
• Use Method 1 or/and 2 described in the Japanese guidance
• No statistical calculation but feasible numbers
 Japanese proportion
• All MRCTs : 32% (2 - 78.5 %)
- Asian Trials: 48% (20 - 48%),
- Global Trials: 21% ( 2 - 53%)
* Larger Proportion of Japanese population in the Asian Trial
89 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Characteristics of MRCTs in Japan
Proportions of the Japanese sample size to the total sample size
based on the Clinical Trial Notifications of MRCTs
 Many cases the
proportion of the
Japanese sample size is
<0.20
 For larger sample size,
the proportion of
Japanese subjects is
<0.10
(Source :Ando and Uyama 2012)
90 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Current practice
- An example
- Possible considerations
and future challenges
91 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
An example - Seebri (COPD medication) –
Phase III MRCT
Submitted in Nov 2011
Approved in Sept 2012
 Study Design :
• A 26-week treatment, randomized, double-blind, placebo-controlled,
parallel group study to assess the efficacy, safety and tolerability of
Seebri in patients with Chronic Obstructive Pulmonary Disease
(COPD)
 Total Sample Size
• 755 (Seebri : 512, Placebo: 243)
• Japanese Sample Size : 92 (62, 30)
- 12% of the entire population
- Calculated based on Method 1
92 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
An example - Seebri (COPD medication) –
Study Design
 Region :
• 13 countries grouped by 6 regions :
1) Japan ⇒ Japan
2) Korea, Singapore, India ⇒ Asia
3) US, Canada ⇒ North America
4) Argentina ⇒ South America
5) Australia, Netherlands, Spain ⇒ European Union
6) Turkey, Romania, Russia ⇒ Eastern Europe
- No specific definition of the region documented
- Randomization was stratified by region
93 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Submitted in Nov 2011
Approved in Sept 2012
An example - Seebri (COPD medication) –
Objectives
 Primary objective
• To demonstrate that Seebri vs. placebo significantly increases
trough FEV1 (Forced expiratory volume in one second) following 12 weeks of
treatment in patients with moderate to severe COPD
 Secondary objectives
• To evaluate the effect of Seebri vs. placebo on the health status by
measuring the total score of the St George’s Respiratory
Questionnaire (SGRQ) after 26 weeks treatment.
• To evaluate the effect of Seebri vs. placebo on breathlessness
measured using the Transition Dyspnea Index (TDI) after 26 weeks
treatment.
• Many more efficacy variables were evaluated...
94 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
An example - Seebri (COPD medication) –
Results : Patient background
ALL
Japan
Seebri
Pbo
total
Seebri
Pbo
total
550
267
817
64
32
96
Age【<65 years】n (%)
278 (50.5%)
134 (50.2%)
412 (50.4%)
14 (21.9%)
9 (28.1%)
23 (24.0%)
Age【>=65years】n (%)
272 (49.5%)
133 (49.8%)
405 (49.6%)
50 (78.1%)
23 (71.9%)
73 (76.0%)
Age【Mean（SD）】yr
63.8 (9.47)
64.0 (8.96)
63.9 (9.30)
69.1 (8.00)
67.4 (9.75)
68.5 (8.61)
Weight【<60 kg】n (%)
139 (27.0%)
67 (27.6%)
206 (27.2%)
33 (51.6%)
13 (43.3%)
46 (48.9%)
Weight【>=60 kg】n (%)
375 (73.0%)
176 (72.4%)
551 (72.8%)
31 (48.4%)
17 (56.7%)
48 (51.1%)
Severity of【Moderate】n
(%)
331 (60.2%)
166 (62.2%)
497 (60.8%)
42 (65.6%)
29 (90.6%)
71 (74.0%)
217 (39.5%)
99 (37.1%)
316 (38.7%)
21 (32.8%)
3 (9.4%)
24 (25.0%)
2 (0.4%)
2 (0.7%)
4 (0.5%)
1 (1.6%)
0 (0.0%)
1 (1.0%)
Duration of COPD【Mean
（SD）】 yr
5.87 (5.798)
6.49 (6.790)
6.07 (6.143)
2.83 (3.067)
2.95 (3.328)
2.87 (3.139)
Smoking history【Exsmoker】n (%)
370 (67.3%)
176 (65.9%)
546 (66.8%)
47 (73.4%)
19 (59.4%)
66 (68.8%)
Smoking history【Current
smoker】n (%)
180 (32.7%)
n
COPD【Severe】n (%)
【very severe】n (%)
Japanese
Patients
older13 &
lighter30&(31.3%)
...
91 (34.1%)
271
(33.2%)
17are
(26.6%)
(40.6%)
95 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
An example - Seebri (COPD medication) –
Efficacy-primary endpoint- Trough FEV1 after 12 weeks treatment
ALL
Baseline
Trough FEV1
at Week 12
【LS
mean/(SE)】
Treatment
difference【LS
mean/(SE)】
[95%CI]
Japan
Seebri
(n=512)
Pbo
(n=243)
Seebri
(n=64)
Pbo
(n=30)
1.321
1.274
1.253
1.340
1.408
(0.0105)
1.301
(0.0137)
1.404
(0.0332)
1.296
(0.0432)
0.108
(0.0148)
0.108
(0.0466)
[0.0785 to 0.1368]
p<0.001
[0.0158 to 0.2011]
p=0.022
Consistent
treatment difference,
but
larger std error in
Japan due to small
sample size
ANCOVA model: Trough FEV1 = treatment + baseline FEV1 + baseline ICS use
(Yes/No) + FEV1 reversibility components + baseline smoking status + region +
center(region). region as fixed effects with center nested within region as a
random effect.
Source: CTD http://www.info.pmda.go.jp/shinyaku/P201200141/index.html
96 |Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Analysis of St George’s Respiratory Questionnaire
(SGRQ) total score at Week 26 (FAS)
ALL
Baseline
Week 26
LS Mean (SE)
Treatment
difference
LS
mean/(SE)】
[95%CI]
Japan
Seebri
(n=502)
Pbo
(246)
Seebri
(n=63)
Pbo
(n=31)
46.11
46.34
38.92
42.28
39.50
(0.813)
42.31
(0.992)
34.70
(1.790)
37.20
(2.502)
-2.81
(0.961)
-2.31
(2.613)
[-4.700 to -0.926]
P=0.004
[-7.702 to 2.683]
p=0.340
Close similarity in
treatment difference
larger std error in
Japan & quite
different p-values
ANCOVA model: SGRQ total score = treatment + baseline SGRQ score +
baseline ICS use (Yes/No) + FEV1 reversibility components + baseline
smoking status + region + center(region). Center is included as a random
effect nested within region.
Source: CTD http://www.info.pmda.go.jp/shinyaku/P201200141/index.html
97
|Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Efficacy-secondary endpoint- Analysis of Transition
Dyspnea Index (TDI) focal score at Week 26 (FAS)
ALL
Baseline
Week 26
LS Mean (SE)
Treatment
difference
LS
mean/(SE)】
[95%CI]
Japan
Seebri
(n=493)
Pbo
(240)
Seebri
(n=62)
Pbo
(n=31)
6.18
6.30
7.27
7.35
1.84
(0.257)
0.80
(0.294)
0.98
(0.558)
1.02
(0.704)
1.04
(0.235)
-0.04
(0.697)
[0.583 to 1.504]
P<0.001
[-1.426 to 1.347]
p=0.955
The treatment
differences are
quite different
Inconsistent
result ?
ANCOVA model: TDI focal score = treatment + BDI + baseline ICS use (Yes/No) +
FEV1 reversibility components + baseline smoking status + region +
center(region). Center is included as a random effect nested within region.
Source: CTD http://www.info.pmda.go.jp/shinyaku/P201200141/index.html
98
|Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
PMDA questions related to regional difference
 Explain reason for the different treatment
effects between Japanese subpopulation and
the entire population
• Why the TDI score got to worse for Japanese
population etc.?
 Explain whether any demographic difference
caused treatment effect or not
99 | Presentation Title | Presenter Name | Date | Subject | Business Use Only
Key Considerations for MRCT
- Opportunities and Challenges -
 Assessment of Consistency
• Not only the Primary endpoint but also the secondary endpoints
would play an important role for helping the assessment of
“consistency”
• Potential effect of the difference need to be evaluated in advance
• Regions should be pre-defined based on intrinsic / extrinsic
factors
<Seebri’s case >
Explain any influence of the following factors for each country, then explain if there are no
big difference in condition of COPD between Japan and Non-Japan patients
•
•
•
•
Cause of COPD, Onset of COPD
Standard Treatment, medical practice
Environment (climate etc.) ,
Disease definition, diagnostic
100|Multi-Regional
Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Key Considerations for MRCT
- Opportunities and Challenges -
 Assessment of Consistency
• Not only hypothesis testing (DJapan / DAll > 0.5, treatment by
region, etc.) but also point estimate, CIs, graphical presentation
would be helpful
• Not only comparing between Japan sub population vs. entire
study population but also displaying other regional populations
could be helpful
 Sample Size for Japanese sub-population
• A single MRCT could provide limited information
• Important for Japanese patients to enroll in MRCTs from earlier phase
and dose response study, PK/PD
- But no established method for sample size determination for DF study
- Limited number of Japanese sample size for each dose
101|Multi-Regional
Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Other Considerations for MRCT
-
- Opportunities and Challenges -
 Consistency assessment on Safety data
• No well established statistical method available
• More comprehensive approach to analyzing safety data is necessary
 Quality of Clinical Trial
• Need a good Quality Management on Operational aspects
- Operational errors in clinical trials detected in the era of complex design
clinical trials and global drug development (Ando (2012) at Bios Workshop )
 Decision making at Interim analyses
• Japanese sub-population assessment / Consistency evaluation
necessary for early termination??
• Speed of the subjects enrollment need to be well managed
102|Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Guideline on Data Monitoring Committees
Issued on April 4th, 2013
3. Setup and implementation of DMC
 3.1 Composition of DMC
When DMC is established in a large-scale global clinical trial, it is
considered appropriate essentially as possible to select a representative
for each participating region or some of regions as committees.
When Japanese subjects participate in such a global clinical trial, it may
be desirable that specialists in Japan participate as DMC committee
members considering on medical environment in Japan and existing
safety information. In case that it is difficult, it should be considered how
to evaluate safety of Japanese subjects beforehand.
Of note, when the special safety monitoring for Japanese subjects is
required by reason that experience of study treatment and safe
information is poor as compared with other regions, it is more significant
that Japanese specialists participate as DMC members.
103|Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
Concluding Remarks
for successful simultaneous global drug development
 Knowledge and experience of MRCT can be shared
among academia, industry and regulatory agencies,
 Looks similar, Not markedly different
Not surprisingly different
 How much we know/share information prior to the study is
the key to success
 Improvement of current practice, “operationally and
methodologically” is required
Japan vs.
Non-Japan
??
Statistical approach
vs. Practical/Feasible
approach
??
104|Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
References
 ICH International Conference on Harmonization Tripartite Guidance E5. Ethnic
Factor in the Acceptability of Foreign Data, (1998).
 ICH International Conference on Harmonization Tripartite Guidance E5. Ethnic
Factor in the Acceptability of Foreign Data, Questions & Answers (2006).
 Ministry of Health, Labour and Welfare of Japan (2007). Basic Concepts for
Joint Clinical Trials
 Ministry of Health, Labour and Welfare of Japan (2012). Basic Concepts for
Joint Clinical Trials. (Reference Cases)
 Ministry of Health, Labour and Welfare of Japan (2013). DMC Guidance
 Kawai, N., et al. (2008). An approach to rationalize partitioning sample size into
individual regions in a multiregional trial. Drug Information Journal, 42, 139-147
 Ando, Y. and Uyama, Y. (2012). Multiregional Clinical Trials : Japanese
perspective on Drug Development Strategy and Sample Size for Japanese
Subjects. Pharmaceutical Statistics, 22, 977-987.
105|Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
References
 Ando, Y. and Hamasaki, T. (2010). Practical issues and lessons learnted from
multi-regional clinical trials via case eamples : a Japanese perspective.
Pharmaceutical Statistics, 9, 190-200
 Tsou, H-H., et al. (2010). Proposals of Statistical consideration to evaluation of
results for a specific region in multi-regional trials – Asian perspective
 Ikeda, K., (2013) Overview of multi-regional trials conducted in Japan. Novartis
China Biostatistics Workshop
 Ando, Y., (2012) Looking beyond ICH-E9 in the Era of Global Drug
Development. Bios Summer Workshop 2012 in Osaka
 Pharmaceuticals and Medical Devices Agency. Review Report of Seebri.
106|Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only
End
107|Multi-Regional Clinical Trials| Tomokazu Inomata | May13 2014| Subject | Business Use Only