Title: An Update on Statistical Issues Associated with the International Harmonization of Technical Standar
1An Update on Statistical Issues Associated with
the International Harmonization of Technical
Standards for Clinical Trials (ICH)
- Robert ONeill , Ph.D.
- Director, Office of Biostatistics, CDER, FDA
22nd Spring Symposium, New Jersey Chapter of ASA,
Wed. June 6,2001
2Outline of talk
- International Harmonization of technical
standards efficacy, safety, quality - statistics - where does it fit in
- Resources - who are the people and what are the
processes - A focus on a few ICH Guidances of interest
- A few issues of particular statistical concern
- The future - where do we go from here
3Harmonization of technical standards
- ICH (Europe, Japan, United States)
- Began in 1989 ICH 1 in Brussels 1991
- ICH continues today
- Outside of ICH
- APEC - Bridging study initiative , Teipei meeting
- Canada, observers, WHO
4Statistical Resources in the ICH regions
- United States
- CDER, CBER
- Europe
- U.K., Germany, Sweden
- CPMP
- Japan
- MHW advisors, university
- China, Taiwan, Canada, Korea
5Web addresses for information and guidances
- www.fda.gov/cder/guidance/index.htm
- www.ifpma.org/ich1
- www.emea.eu.int/
6ICH Guidances with statistical content
- E1 Extent of population exposure to assess
clinical safety - E3 structure and content of clinical study
reports (CONSORT statement) - E4 Dose-response information to support drug
registration - E5 Ethnic factors in the acceptability of
foreign clinical data - E9 Statistical principles for clinical trials
- E10 Choice of control group
- E11 Clinical investigation of medicinal products
in the pediatric population
7ICH Guidances with statistical content
- Safety
- carcinogenicity
- Quality
- Stability (expiration dating) Q1A, Q1E
8New initiatives from the European Regulators
(CPMP)- Points to Consider Documents
- On Validity and Interpretation of Meta-Analyses,
and One Pivotal Study (Jan, 2001) - On Missing Data (April, 2001)
- On Choice of delta
- On switching between superiority and
non-inferiority - On some multiplicity issues and related topics in
clinical trials
9Efficacy Working Party (EWP) Points to
Consider CPMP/EWP/1776/99 Points to Consider on
Missing Data (Released for Consultation January
2001) CPMP/EWP/2330/99 Points to Consider on
Validity and Interpretation of Meta-Analyses, and
one Pivotal study ( released for consultation
October 2000) CPMP/EWP/482/99 Points to
Consider on Switching between Superiority and
Non-inferiority (Adopted July 2000)
10ICH E9Statistical Principles for Clinical
Trials Contents
- Introduction ( Purpose, scope, direction )
- Considerations for Overall Clinical Development
- Study Design Considerations
- Study Conduct
- Data Analysis
- Evaluation of safety and tolerability
- Reporting
- Glossary of terms
11Study Design A Major Focus of the Guideline
- Prior planning
- Protocol considerations
12Prospective Planning
- Design of the trial
- Analysis of outcomes
13Confirmatory Study vs. Exploratory Study
- A hypothesis stated in advance and evaluated
- Data driven findings
14Design Issues
- Endpoints
- Comparisons
- Choice of study type
- Choice of control group
- Superiority
- Non-inferiority
- Equivalence
- Sample size
- Assumptions, sensitivity analysis
15Choice of Study Type
- Parallel group design
- Cross-over design
- Factorial design
- Multicenter design
16Analysis Outcome Assessment
- Multiple endpoints
- Adjustments
17Assessing Bias and Robustness of Study Results
18Analysis Sets
- ITT principle
- All randomized population
- Full Analysis population
- Per Protocol
19Data Analysis Considerations
- Prespecification of the Analysis
- Analysis sets
- Full analysis set
- Per Protocol Set
- Roles of the Different Analysis Sets
- Missing Values and Outliers
20Statistical Analysis Plan (SAP)
- A more technical and detailed elaboration of the
principal features stated in the protocol. - Detailed procedures for executing the statistical
analysis of the primary and secondary variables
and other data. - Should be reviewed and possibly updated during
blind review, and finalized before breaking the
blind. - Results from analyses envisaged in the protocol
(including amendments) regarded as confirmatory. - May be written as a separate document.
21Analysis Sets
- The ideal the set of subjects whose data are to
be included in the analysis - all subjects randomized into the trial
- satisfied entry criteria
- followed all trial procedures perfectly
- no loss to follow-up
- complete data records
22Full Analysis Set
- Used to describe the analysis set which is
complete as possible and as close as possible to
the intention to treat principle - May be reasonable to eliminate from the set of
ALL randomized subjects, those who fail to take
at least one dose, or those without data post
randomization. - Reasons for eliminating any randomized subject
should be justified and the analysis is not
complete unless the potential biases arising from
exclusions are addressed and reasonably dismissed.
23Per Protocol Set
- Sometimes described as
- Valid cases, efficacy sample, evaluable subjects
- Defines a subset of the subjects in the full
analysis set - May maximize the opportunity for a new treatment
to show additional efficacy - May or may not be conservative
- Bias arises from adherence to protocol related to
treatment and/or outcome
24Roles of the Different Analysis Sets
- Advantageous to demonstrate a lack of sensitivity
of the principal trial results to alternative
choices of the set of subjects analyzed. - The full analysis set and per protocol set play
different roles in superiority trials, and in
equivalence or non-inferiority trials. - Full analysis set is primary analysis in
superiority trials - avoids optimistic efficacy
estimate from per protocol which excludes
non-compliers. Full analysis set not always
conservative in equivalence trial
25Impact on Drug Development
- On sponsor design and analysis of clinical trials
used as evidence to support claims - On regulatory advice and evaluation of sponsor
protocols and completed clinical trials - On maximizing quality and utility of clinical
studies in later phases of drug development - On multidisciplinary understanding of key
concepts and issues - Enhanced attention to planning and protocol
considerations
26Will the Guideline Help to Avoid Problem Areas
in the Future - Maybe !
- Not a substitute for professional advice-will
require professional understanding and
implementation of the principles stated - Will not assure correct analysis and
interpretation - Most of the guideline topics reflect areas where
problems have been observed frequently in
clinical trials in drug development
27ICH Chemistry
- Q1E Bracketing and Matrixing Designs for
Stability Testing of Drug Substances and Drug
Products - Considerable new work, including extensive
simulations to evaluate size of studies and the
ability to detect important changes to expiration
date setting (incomplete blocks, alias, etc).
28ICH E10 Choice of Control Group and Related
Design Issues in Clinical Trials
- Section 1.5 is very statistically oriented
involving issues like - Assay sensitivty
- Historical evidence of sensitivity to drug
effects - Choice of a margin for a non-inferiority (dont
show a difference ) trial.
29Assay Sensitivity in Non-inferiority designs
- Assay sensitivity is a property of a clinical
trial defined as the ability to distinguish an
effective treatment from a less effective or
ineffective treatment - Note that this property is more than just the
statistical power of a study to demonstrate an
effect - it also deals with the conduct and
circumstances of a trial
30The presence of assay sensitivity in a
non-inferiority trial may be deduced from two
determinations
- 1) Historical evidence of sensitivity to drug
effects, I.e., that similarly designed trials in
the past regularly distinguished effective
treatments from less effective or ineffective
treatments, and - 2) Appropriate trial conduct, I.e. that the
conduct of the trial (current) did not undermine
its ability to distinguish effective treatments
from less effective or ineffective treatments.
can be fully evaluated only after the active
control non-inferiority trial is completed.
31Successful use of a non-inferiority trial thus
involves four critical steps
- 1) Determining that historical evidence of
sensitivity to drug effect exists. Without this
determination, demonstration of efficacy from a
showing of non-inferiority is not possible and
should not be attempted. - 2) Designing a trial. Important details of the
trial design, e.g. study population, concomitant
therapy, endpoints, run-in periods, should adhere
closely to the design of the placebo-controlled
trials for which historical sensitivity to drug
effects has been determined.
32Successful use of a non-inferiority trial thus
involves four critical steps (cont.)
- 3) Setting a margin. An acceptable
non-inferiority margin should be defined, taking
into account the historical data and relevant
clinical and statistical considerations. - 4) Conducting the trial. The trial conduct should
also adhere closely to that of the historical
trials and should be of high quality.
33Choosing the Non-inferiority margin
- Prior to the trial, a non-inferiority margin,
sometimes called a delta, is selected. - This margin is the degree of inferiority of the
test treatments to the control that the trial
will attempt to exclude statistically. - The margin chosen cannot be greater than the
smallest effect size that the active drug would
be reliably expected to have compared with
placebo in the setting of the planned trial.
based on both statistical reasoning and clinical
judgement, should reflect uncertainties in
evidence and be suitably conservative.
34Outline of the Issues
- What is the the non-inferiority design
- What are the various objectives of the design
- Complexities in choosing the margin of treatment
effect - it depends upon the strength of evidence
for the treatment effect of the active control - Literature on historical controls, and on the
heterogeneity of treatment effects among studies - The statistical approaches to each objective, and
their critical assumptions - Cautions and concluding remarks
35Non-Inferiority Design
- A study design used to show that a new treatment
produces a therapeutic response that is no less
than a pre-specified amount of a proven treatment
(active control), from which it is then inferred
that the new treatment is effective. The new
treatment could be similar or more effective than
the existing proven treatment - A non-inferiority margin ? is pre-selected as
the allowable reduction in therapeutic response.
The margin ? is chosen based on the historical
evidence of the efficacy of the active control
and other clinical and statistical considerations
relevant to the new treatment and the current
study. - ICH - E10 This delta can not be greater than
the smallest effect size that the active drug
would be reliably expected to have compared with
placebo in the setting of a planned trial. - the
concept of reliably and repeatedly being able
demonstrate a treatment effect of a specified
size !
36Non-Inferiority Design (contd)
- A test treatment is declared clinically
non-inferior to the active control if - the trial has the necessary assay sensitivity for
the trial to be valid for non-inferiority testing - the one-sided 97.5 confidence interval is
entirely to the right of - ?
37Inference for Non-Inferiority
- Delta Limits 95 Confidence Intervals
Non-inferiority shown
Non-inferiority shown
Non-inferiority not shown
Non-inferiority shown/ superiority could be
claimed
- ?
0
Control Better
Test Agent Better
Treatment Difference
38What are the various objectives of the
non-inferiority design
- To prove efficacy of test treatment by indirect
inference from the active control treatment - To establish a similarity of effect to a known
very effective therapy - e.g. anti-infectives - To infer that the test treatment would have been
superior to an imputed placebo ie.
had a placebo group been included for comparison
in the current trial. - a new and controversial
area - choice of margin is the key
39What is the Evidence supporting the treatment
effect of the active control, and how convincing
is it ?
- Large treatment effects vs. small or modest
effects - Large treatment effects - anti-infectives
- Modest treatment effects - difficulties in
reliably demonstrating the effect - Sensitivity
to drug effects - Amount of prior study data available to estimate
an effect - One single study
- Several studies, of different sizes and quality
- No estimate or study directly on the comparator -
standard of care
40How is the margin ? chosen based upon prior
study data
- For a large treatment effect, it is easier - a
clinical decision of how similar a response rate
is needed to justify efficacy of a test treatment
- e.g. anti-infectives is an example. - For modest and variable effects, it is more
difficult and some approaches suggest margin
selection based upon several objectives.
41Complexities in choosing the margin (how much of
the control treatment effect to give up)
- Margins can be chosen depending upon which of
these questions is addressed - how much of the treatment effect of the
comparator can be preserved in order to
indirectly conclude the test treatment is
effective - a clinical decision for very large
effects a statistical problem for small and
modest effects - how much of a treatment effect would one require
for the test treatment to be superior to placebo,
had a placebo been used in the current active
control study - a lesser standard than the above
42How convincing is the prior evidence of a
treatment effect ?
- Do clinical trials of the comparator treatment
consistently and reliably demonstrate a treatment
effect - when they do not, what is the reason ? - Study is too small to detect the effect - under
powered for a modest effect size - The treatment effect is variable, and the
estimate of the magnitude will vary from study to
study, sometimes with NO effect in a given study
- a BIG problem for active controlled studies
(Sensitivity to drug effect)
43How do you know which treatment effect size is
appropriate for the current active control ?How
much protection should be built into the choice
of the margin to account for unknown bias and
uncertainty in study differences ?
44Inherently, the answer relies upon historical
controls and their applicability to the current
study
- Choice of the margins should take into account
all sources of variability as well as the
potential biases associated with
non-comparability of the current study with the
historical comparisons. - A need to balance the building in of bias in
the comparison and quantifying the amount of
treatment effect preserved, as a function of the
relative amount of data from the historical
studies and the current study
45Use of historical controls in current RCTs
- Pocock,S. The combination of randomized and
historical controls in clinical trials. J.
Chronic Diseases 1976, 29 pp.175-188 - Lists 6 conditions to be met for valid use of
historical controls with controls in current
trial - Only if all these conditions are met can one
safely use the historical controls as part of a
randomized trial. Otherwise, the risk of a
substantial bias occurring in treatment
comparisons cannot be ignored.
46Importance of the assumption of constancy of the
active control treatment effect derived from
historical studies
- It is relevant to the design and sample size of
the current study, to the choice of the margin,
to the amount of bias built into the comparisons,
to the amount of effect size one can preserve
(both of these are likely confounded), and to the
statistical uncertainty of the conclusion. - Before one can decide on how much of the effect
to preserve, one should estimate an effect size
for which there is evidence of a consistent
demonstration that effect size exists.
47Explaining Heterogeneity among independent
studies Lessons from meta-analyses
- Variation in baseline risk as an explanation of
heterogeneity in meta-analysis, S.D. Walter,
Stat. In Medicine, 16, 2883-2900 (1997) - An empirical study of the effect of the control
rate as a predictor of treatment efficacy in
meta-analysis of clinical trials,
Schmid,Lau,McIntosh and Cappelleri, Stat. In
Medicine, 17, 1923-1942 (1998)
48Explaining Heterogeneity among independent
studies Lessons from meta-analyses (cont.)
- Explaining heterogeneity in meta-analysis a
comparison of methods. Thompson and Sharp, Stat.
In Medicine, 18, 2693-2708 (1999) - Assessing the potential for bias in meta-analysis
due to selective reporting of subgroup analyses
within studies. Hahn, Williamson, Hutton, Garner
and Flynn, Stat. In Medicine, 19, 3325-3336 (2000)
49Explaining Heterogeneity among independent
studies Lessons from meta-analyses (cont.)
- Large trials vs. meta-analysis of smaller trials
- How do their results compare ? Cappelleri,
Ioannidis, Schmid, de Ferranti, Aubert, Chalmers,
Lau. JAMA, 16 1332-1338, 1996 - Discordance between meta-analysis and large-scale
randomized controlled trials examples from the
management of acute myocardial infarction. Borzak
and Ridker, Ann. Internal Med.,123, 873-877
(1995) - Discrepancies between meta-analysis and
subsequent large randomized controlled trials.
LeLorier, Gregoire, Benhaddad, Lapierre,Derderian.
NEJM, 337, 536-42 (1997)
50Use of meta-analysis - necessary but not
sufficient
- Distinguish under powered studies from well
powered studies for a common effect size - if
possible - How many trials are consistent with no effect,
rather than an effect of some size - Determine between trial variability as an
additional factor to consider in choosing a
conservative margin - How do you know if the current study comes from
the same trial population, and where does it rest
in the trial distribution - critical to
assumptions for control group rate and constancy
of treatment effect - Resorting to meta-analysis of all studies, when
few individual studies reject null, tells you
something !
51Three approaches to the problem
- Indirect confidence interval comparisons (ICIC)
(CBER/FDA type method, etc.) - - thrombolytic agents in the treatment of acute
MI - Virtual method (Hasselblad Kong, Fisher, etc.)
- - Clopidogrel, aspirin, placebo
- Bayesian approach (Gould, Simon, etc.)
- - treatment of unstable angina and non-Q wave MI
52When may it not be possible to estimate a margin
or to use the non-inferiority design to infer
efficacy ?
- There is a known creep in the standard of care
over time and/or the active control treatment,
which renders any past estimates of active
control treatment effects not comparable or valid
for the current comparison, under conditions of
medical practice in the new current study - e.g. use of surfactants in neonatal treatment
53ICH E5Ethnic Factors in the Acceptability of
Foreign Clinical Data
54Key Features of E5
- Operational definition of ethnic factors
- Clinical Data Package Fulfilling Regulatory
Requirements in New Region - Extrapolation of Foreign Clinical Data to New
Region (role of ethnic factors) - Bridging Studies
- Global Development Strategies
55Ethnic Factor Definition
- intrinsic factors characteristics associated
with the drug recipient (ADME studies) - race, age, gender, organ dysfunction, genetic
polymorphism - extrinsic factors characteristics associated
with the environment and culture in which one
lives (clinical outcomes) - clinical trial conduct, diet, tobacco and alcohol
use, compliance with prescribed medications
56Assessing a medicines sensitivity to ethnic
factors(part of the screening process)
- Properties of a compound making it more likely to
be sensitive - Metabolism by enzymes known to show genetic
polymorphism - High likelihood of use in a setting of multiple
co-medications
57Assessment of the Clinical Data Package (CDP) for
acceptability
- Question 1 Meets regulatory requirements -
yes/no - Question 2 Extrapolation of foreign data
appropriate - yes/no - Question 3 Further clinical study (ies) needed
for acceptability by the new region - yes/no - Question 4 Acceptability in the new region -
yes/no
58Meets regulatory requirements
- Issues of evidence
- Confirmatory evidence two or more studies
showing treatment effects - Interpreting results of foreign clinical trials
which provide that evidence (may be one study, or
all studies, or part of a study) - Which study designs provide evidence
- Active control / non-inferiority designs
- Placebo or active control / show a difference
designs
59The sources of data for an application
(implementation)
- All clinical studies for efficacy performed in
foreign region - One study in the United States, one or more
foreign clinical studies - Multi-center/ multi-region clinical trials form
the basis for efficacy
60Considerations for evaluating clinical efficacy
between regions
- Study design differences
- Magnitude of treatment effect sizes
- Effect size variability subgroup differences
- Impact of intrinsic factors - determined when ?
- Impact of Extrinsic factors
- trial conduct and monitoring
- usage of concomitant medications
- protocol adherence
61Bridging Studies
E5 is purposely vague on how to do this or what
their design should be
62Study design and study objectives(need examples
and experience)
- What type of bridging study would be helpful for
extrapolation - - PK/PD
- Another clinical trial of the primary clinical
endpoint - equivalence/non-inferiority treatment effect
acceptably close - margin or delta - dose response study
- superiority design - estimate treatment effect
size for comparison
63E5 allows for a new study in the new region -
why is that needed ?
- When all the clinical data is derived from a
foreign region and extrapolation is an issue - When the experience with clinical trials in that
region is minimal - When there is concern with ability to confirm a
finding from a study(ies) - A confirmatory clinical trial is the bridging
study
64Developmental Strategies for Global Development
- Early vs. later strategies
- Designing population pk/pd into clinical studies
- Planning to explain effect size differences among
regions - Design of bridging studies early in development
65Study Design
- Better planning in Phase I, II, III and more
efficient study designs to address several
subgroup questions simultaneously - Design Phase III with some knowledge of PK / PD
differences in Phase I / II - Address multiple questions simultaneously for
efficiency (age, gender, ethnic)
66Study Design
- Assessing the influence of ethnic factors in each
study Phase (I, II, III) and to identify earlier
and account for, by design, the influence of
ethnic factors - Ethnic factors as another subgroup
- Age, gender, renal status, etc.
- Ethnic factors integrated with
- Dose response
- Geriatrics
- Population exposure for safety
67 Remarks
- Little experience at this time with bridging
studies - Little experience with Japanese trials in NDA
applications, or trials from Asia - More experience with foreign trials from Europe -
possible heterogeneity of treatment effects being
evaluated concern for experience in new regions
like Eastern Europe
68The future
- Appears to be increasingly dependent on
statistical input, methods, study design,
interpretation , etc. - Statistical resources (people) are needed in the
regulatory agencies in all countries/regions
serious about inference - not always present ,
maintained - cannot develop guidance documents
and consensus positions without this,nor rely on
guidances alone - Global drug development is beginning to recognize
the need for early planning for multi-regional
inference - the questions and study designs are
just unfolding