Transcript Bridging Studies Global Development

Bridging Studies 
Global Development
W. Joe Shih
Biostatistics Dept
UMDNJ-School of Public Health
Sept. 29, 2006
FDA/Industry Conference,
Washington D.C.
Clarification
• Bridging studies: in the context of “ICH E5”.
• ICH E5 includes drugs and biologics.
• Not the ‘vaccine immunogenicity bridging
trials’ (which are conducted when
manufacturing processes or storage
conditions are changed after vaccine
licensure for improving production yields
and vaccine stability or shelf life).
New Trend
• Moving from “Bridging study” toward
“Global trial”
• More fruitful to discuss issues with
“Global trial” design and analysis
My presentation: ½ “Bridging” ½ “Global”
Purpose of ICH E-5
• "To facilitate the registration of medicines among
ICH regions – US/EU/Japan…” (now expanded
to other countries)
• “To provide guidance on regulatory and
development strategies …"
 Extrapolate foreign clinical data and evaluate
influence of ethnic factors
 Minimize duplication of clinical studies
 Supply medicine expeditiously to patients
Difficulties
• “Ethnicity” is a not a simple nor precise concept
• Political and cultural sensitivities (involving
international trade of pharmaceuticals - WTO)
• Lack of consensus even regarding need for
such a guidance 
A compromise concept in E5:
 "Bridging data package" = ‘Selected information
from the complete clinical data package (CCDP)
that is relevant to the population of the new
region’ + (if needed) ‘Bridging study(ies) in the
new region.’
Difficulties
• No gold standards of “bridging studies”
• Heterogeneity among regulatory agency,
industry, and academia in the interpretation of
the ICH guideline; e.g.
 what is “extrapolation” of foreign data?
 how to evaluate the “bridging evidence”?
Practical Issues
• What type of bridging studies are required?
– Clinical studies?
– PK/PD studies?
• Sample size of a bridging study?
 Not all regions/countries act the same way
 Market size seems matter very much
(unfortunately)
Japanese Experience
Japanese Drug Organization (‘Kiko’, or OPSR):
“All the examples of bridging studies raised by the
DO are phase II/III studies which would have to
be performed exclusively in Japan.” (In addition to
PK/PD.)
Naito, C. ‘Acceptance of foreign clinical data and bridging
studies.’ Proceedings of the 19th conference on
Japanese Society of Clinical Pharmacology and
Therapeutics. Rinsho Yakuri, Jpn J. Clin Pharmacol Ther
30:505-506, 1999.
Other Asia-Pacific Experience
• Much relaxed/friendly environment; this
region’s regulatory agencies have
provided:
Criteria for bridging evaluation
Check for “need a bridging study?”
Long list of products that require no
verification of ethnic insensitivity
• Example: Taiwan, Korea, Singapore,
Thailand
Considerations for Assessing the Necessity of a
Bridging Study ( 1of 4 )
Submit relevant documents
Does the drug meet DOH requirements for waiving
YES according to DOH
a bridging study and also the criteria for exempting
requirements and request
submission of information for ethnic consideration?(1)
for waiving bridging studies
NO
Does the submitted preclinical and clinical
data package meet the regulatory requirements
(ICH E5 and DOH guidance on clinical trials)?(2)
NO
Amendment
YES
NO
Does the package include clinical data
of Asian populations?(3)
Is the medicine insensitive to both
intrinsic and extrinsic factors?
efficacy and safety insignificant?
(See ICH E5 guidelines)
YES
YES
NO
Considerations for Assessing the Necessity of a
Bridging Study
( 2 of 4 )
Does the package include clinical data
of Asian populations?(3)
NO
YES
YES
Have any early phase trials or
global clinical trials that meet the
YES
DOH requirements of bridging studies
been conducted in Taiwan?
NO
Is it reasonable to extrapolate from foreign
clinical data that the medicine is insensitive
to both intrinsic and extrinsic factors in Asians(3)
and that its clinical differences in efficacy and
safety are acceptable?(See ICH E5 guidelines)
NO
Is the medicine insensitive to both
intrinsic and extrinsic factors?
Are the clinical differences in
efficacy and safety insignificant?
(See ICH E5 guidelines)
No bridging
study required(4)
NO
Based on the result of evaluation,
an appropriately designed
protocol of a bridging study
should be submitted to
DOH for approval(5)
YES
No bridging study required(4)
Considerations for Assessing the Necessity of a
Bridging Study
( 3 of 4 )
Is it reasonable to extrapolate from foreign
clinical data that the medicine is insensitive
to both intrinsic and extrinsic factors in Asians(3)
and that its clinical differences in efficacy and
safety are acceptable?(See ICH E5 guidelines)
YES
No bridging
study required(4)
NO
Is it reasonable to extrapolate from
foreign clinical data that the
concentration (dose)-response relationship
is similar between foreign and
Asian populations(3)?
NO
Is PK and/or PD data of Asian
populations(3) available for
estimating dosage or predicting efficacy?
YES
NO
YES
Based on the result of evaluation,
an appropriately designed protocol
of a bridging study should be
submitted to DOH for approval.(5)
Considerations for Assessing the Necessity of a
Bridging Study
( 4 of 4 )
Is PK and/or PD data of Asian
populations(3) available for estimating
dosage or predicting efficacy?
YES
NO
Based on the result of evaluation,
an appropriately designed protocol
of a bridging study should be
submitted to DOH for approval.(5)
Using available data for
dose determination
(1) Apply
for waiving bridging studies with reference to DOH announcements of waiving clinical
trials. If the drug falls within the category that “requires submission of information proving no
existence of ethnic differences”, it should be evaluated following this flowchart after the one
year phase-in period.
(2) Under circumstances when evidence indicating potential intrinsic/extrinsic differences
between Chinese and other Asian populations, a bridging study in Chinese population is a must.
(3) A bridging study will be required when there exists any safety concern.
1(4) Under circumstances when evidence indicating potential intrinsic/extrinsic differences
between Chinese and other Asian populations, a bridging study in Chinese population is a must.
(5) A bridging study can be a PK and/or PD study or any clinical study that can demonstrate the
efficacy and safety of the medicine.
Products Requiring No Verification of
Ethnic Insensitivity
•
•
•
•
•
•
•
Drugs for treatment of AIDS
Drug for organ transplantation
Topical agents
Nutrition supplements
Cathartics prior to surgery
Radiolabeled diagnostic pharmaceuticals
The drug is the only choice of treatment for a given
severe disease
• Drugs for life-threatening disease have demonstrated a
breakthrough efficacy
• Lacking adequate trial subjects for any drug used for
rare disease
Products Requiring No Verification of
Ethnic Insensitivity
•
•
•
•
•
•
•
•
Anticancer drugs
Drugs with breakthrough efficacy
Drugs of single use
Drugs with different salt of the same composition and the same
administered route have been approved internal
Drugs for chronic psychologic or immunological diseases and
conducting clinical trails internal difficulty
Each compounds of new combination drug have been proved
internal, and the efficacy is the same as the single compound
Drugs with the mechanism, administered route, efficacy and
adverse effect, similar to the approved drugs
New combination composed of single compound of approved
combination or compounds of approved combination has the
same efficacy as approved combination
US FDA’s Experience
O’Neill (2003). The ICH E5 Guidance: An Update on
Experiences with its Implementation
• Majority of NDA’s contain foreign clinical trial
data, often used as primary evidence of
efficacy and safety – rarely, does the entire
data base on efficacy consist of foreign clinical
data
• Until recently, discussion have rarely been
held with sponsors during IND/NDA
development stages that specifically consider
bridging strategies when relying on foreign
clinical data
US FDA’s Expereince
• Some, but not all review divisions, during the process of
evaluation of the clinical efficacy data examine regional
differences in efficacy and safety
• Most multi-national trials have included patients from
Western Europe, U.S., Canada, New Zealand and
Australia
– Minimal but increasing experience with Latin America
and Eastern Europe
• Few examples of formal bridging studies done in the U.S.
that were performed subsequent to development of a
complete clinical data package, and that were carried out
in response to an FDA request
US FDA’s Expereince
• Generally, when FDA asks for more data/studies,
it is because the clinical trial evidence in the
NDA is not convincing and other formal phase 3
studies conducted in the U.S. are needed
• As trials come from new regions, it may become
critical to agree in advance on the sources of
data
• There has not often been a prospective
evaluation during the IND of differential PK, PD
or clinical endpoints to treatment response
Evaluating “Bridging Evidence”
Major Statistical Methods
• Shih (2001, CCT) – “Consistency”
• Liu, Hsueh and Chen (2002, BJ) –
“Equivalence”
• Chow, Shao, Hu (2002, JBS) –
“Reprodicibility/Generalizability”
• Hsiao, Xu, Liu (2003, JBS) – “Group
sequential”
• Wang/Hung (FDA) (Stat in Med, 2003) “Non-inferiority”
• Ware (Harvard) – “Hierarchical model”
• Purpose of “Bridging” is not to get an
‘overall’ estimate of treatment effect ()
by adding another study
Consistency
Shih (2001). Controlled Clinical Trials, 22, 357-366.
• Results of K reference studies (per investigator)
from the CCDP are available: W  w ,....., w 
1
• New (small) local study result from the new
region :
K
v
• First, construct the predictive probability
function pv W , which provides a measure of the
plausibility of v given the results W
Consistency
• Then compare pv W  with the plausibility of each
of the actually observed wi , i.e., Pwi W 
• The result v is considered consistent with the
previous results W if and only if
Pv W   min Pwi W , i  1,..., K 
(or, a low, say, 5%tile)
v falls within the
 Consistency
previous experience of W
 Bayesian most plausible prediction
Reproducibility/Generalizability
Chow, S.C., Shao, J., and Hu, O.Y.P. (2002). Journal of
Biopharmaceutical Statistics, 12, 385-400.
• Statistical Criteria
– Reproducibility
– Generalizability
• Sensitivity Index
– A measure, which is derived based on the difference
in two patient populations, to determine the chance of
reproducibility and generalizability based on the
observed clinical data
Similarity, Equivalence/Non-inferiority
Hung et al. (2003). Statistics in Medicine, 22, 213225.
• Let  be the therapeutic effect
 Original region 1
 New region 2
• Data (from the original region) available for 1
 1 >0 has been established
• Want to test hypotheses of 2
Similarity, Equivalence/Non-inferiority
• Non-inferiority margin
  r1
• Hypotheses
H 0 :  2  r1
vs. H a :  2  r1
Hung: “Sample size for new trial will usually be
demanding”.
Global trial: M-national M-center
(M-n/M-c) trials
• When a ‘m-national m-center’ trial is conducted,
likely it is the only study (i.e., no duplication)
• US FDA’s practice: When only one “m-center
trial” is available for establishing ‘substantial
evidence’ of efficacy and safety, the results need
to be very persuasive:
 Overall Rx effect: Highly significant (e.g., p-value
<= 0.001)
 No single site dominates the favorable effect
 Internally consistent – across centers, endpoints,
other subgroups – But ….
 Leave open how to assess ‘internal consistency’
M-n/M-c trials
FDA’s practice
• Charles Anello (FDA)
ASA FDA-Industry Workshop, Bethesda 2003
Results for individual studies should
be presented;
Extreme or opposite results need to
be discussed;
Special attention is given to age,
gender and ethnicity (or region)
That is, FDA does not seem to
differentiate ‘M-n/M-c’ trials from the
usual Multi-center trials
My View:
M-n/M-c vs. M-c trials
• In m-center trials, ni in each center is not
fixed (although anticipated) and
randomization is done within each center
• Centers compete with a total N (designed)
• When the total N is reached, recruitment
stops
• Centers with small n are usually combined
• Definition of ‘center’ somewhat arbitrary
• In a M-n/M-c trial, ‘nation’ is not arbitrary and
the sample size given for a nation has to be
minimally specified (same as in ‘bridging’)
M-n/M-c vs. M-c trials
• M-center trials (ICH E-9): Test treatment main
effect first; if significant …
• Test TxC interaction as an exploratory
analysis (no specific method or -level)
• When there are a large number of centers, it
is less important to consider interaction (since
TxC is expected? )
• M-c Focus: the ‘overall’ (main) Rx effect
• M-n/M-c focus: nation-specific effect, not
the overall Rx effect
Overall Rx effect vs.
Country specific effect
• You never (I bet) saw a drug allowed to be
marketed in every state in the U.S. except
for California (or vice versa)
• But some drugs are approved in the US, but
not in Japan; some are approved elsewhere
but not in the US
• FDA Hung’s “similarity/non-inferiority” for
evaluating new region’s 2 is ‘countryspecific’ too)
Global drug development:
M-n/M-c trials
Statistical approach: Hierarchical (multilevel) models
• Nations can ‘combine’ their data to
improve estimation for each nation/region
while allowing for heterogeneity
• (Empirical) Bayes methods/REML for
multi-level models provide approach to
estimation
Efron (1996 JASA): Hierarchical (E-) Bayes Model
=
U
Hierarchical (E-) Bayes Model
g ( )
is the direct data for 0 (of interest)
X is the other countries/centers’ data for U
Marginal sampling density: d ( x )   g (u ) f ( x )du
Conditional density of  given X: h( | x )  h( )d ( x)
Induced Prior density for 0: g x ( 0 )   h( | x) g ( 0 )d
 X0




u
 Posterior density of 0 based on all the data (X0,X):
p x ( 0 | x0 )  g x ( 0 ) f ( x0 )  g x ( 0 ) L0 ( 0 )
0
Construct hierarchical Bayes interval for 0
Hierarchical (E-) Bayes Model
 (Morris 1983) E-Bayes tactic: Replace the
induced prior with the MLE prior ( ̂ from d ( x ) )
the MLE posterior:
pˆ ( 0 | x0 )  gˆ ( 0 ) L0 ( 0 )
 The hyper-prior density h(.) is no longer needed
 Other improvements (e.g., bias correction) also
considered in Efron (1996)
 “Relevance” of X to 0 is a crucial issue
 May consider ‘mixture/weighted’ prior for g (.)
 Hierarchical generalized linear models (Lee &
Nelder, 1996 JRSS-B) are also worth studying.
Simple Model Example
 Standardized Rx effect: xi ~ N(i , 1);
 Prior i ~ N(, 2);
i = 0, 1, …, K
 Marginal: xi ~ N(, 2 +1)
 MLE: ̂  x , ˆ 2  s 2 (only use xi, i =1,...,K)
 MLE posterior for 0 ~ N( (1  B) x0  Bx , 1-B);
K
B
 E-Bayes -level interval for 0
( K  1) s 2
((1  B) x0  Bx )  z / 2 (1  B)1/ 2
 Efron (1996): Not as good as Stein estimator, which
uses x0 in MLE (improvement can be done)
 Sample size (interval width) depends only on the
K
other data, 1-B = 12
( K  1) s
 Desirable to have large K and small variability in
the data X
 H- (E)-Bayes or H-GLM models are applicable to
the bridging study scenario as well …
 But the the simultaneous multi-level modeling
makes more sense when all data (X0, X) are
from the same (M-n/M-c) trial.
An Example:
Summary of Results From a Global Drug Development for
Treatment of BPH – 14 national centers
Xi
¶Mean
(Xi-X)2
and standard deviation of the change in Total Symptom Score in the control (1) and
test (2) treatment groups respectively.
Hierarchical E-Bayes 95%-level intervals
Nation/Center-Specific Rx Effect
(Simple Model Example: BPH Data)
5
4
3
2
1
0
1
-1
-2
2
3
4
5
6
7
8
9
10
11
12
13
14
Global Trial: Final remarks
 As one global trial, all nations’ data are used to

improve estimation for each nation while allowing
for heterogeneity.
There won’t be a ‘global registration’; hence,
nation/region-specific Rx effects (assuming
properly estimated now) are more of interest than
the ‘overall’ Rx effect.
 Nation/region-specific treatment effects can be

contrasted with others, but not in the spirit of
showing ‘similarity’ or ‘non-inferiority’.
Excess heterogeneity, if happens, still needs to
be explained carefully.
Summary
 Scenario (A): When a large US/EU company
seeks a drug approval in a small AP nation, use
predictive ‘consistency’ to assess fit of the small
trial to the larger set of prior experience.
 Scenario (B): When studies are done elsewhere
and submitted for NDA in the US, the FDA will
likely ask for repetition studies in the US and
assess some kind of ‘similarity’/’non-inferiority’
(not in the ‘active control’ sense).
 Scenario (C): For global drug trials, simultaneous
multi-level models helps to better estimate nationspecific Rx.