Active Control Studies
Download
Report
Transcript Active Control Studies
Equivalence, Similarity, and Non-inferiority
Clinical Trials in Neurotherapeutics
ASENT 10th Annual Meeting
March 2008
Marc K. Walton, M.D., Ph.D.
Senior Medical Policy Advisor
Office of Policy
Office of the Commissioner, FDA
The views expressed are those of the author, and do not represent an official
FDA position
Equivalence / Similarity / Noninferiority
Distinctions of study circumstances and study goal
Need to be clearly identified
Two treatments
Both known efficacious
Want to know comparative efficacy
One treatment with unknown efficacy
Second treatment known efficacious
Want to prove first is also efficacious
Comparative Efficacy
Can the two treatments be shown Equivalent?
Are two agents are essentially the same?
Implies same amount of efficacy for each treatment to
within some level of disinterest
Neither drug is better than the other
Two sided interest
Not a regulatory requirement for marketing
Not common regulatory interest (ie, sponsor interest,
possibly with desire for regulatory affirmation)
“The Same”?
With respect to characteristics planned for rigorous
evaluation
Proof of Efficacy for an Unproven Treatment
Is the new treatment efficacious?
In the same manner as the established treatment
Often a goal for regulatory decision
Based on evidence that new drug has some efficacy
No explicit requirement for same efficacy of other
drugs, or any predefined fraction of efficacy of other
drugs for same disorder.
One sided comparison of interest
Precise comparative efficacy is not a goal
Non-inferiority: A misnomer
Goal to is to show “sufficient efficacy”
Accomplished by showing “not unacceptably inferior”
to control
Does not imply truly not inferior
Similarity
Perhaps not equivalent, but close enough
‘Close enough’?
In the eye of beholder
May imply a two sided interest as well
Individual judge dependent
Difficult to define/describe
Principles of Equivalence / Noninferiority studies apply
More laxity in interpretation granted on an individual
person basis
Equivalency & Noninferiority: Active Control Studies
Why / When do Active Control studies?
Comparison of effects of two agents
May not need to exclude Placebo
Issues of assay sensitivity (later) and validity of
interpretation much easier if placebo included
Assessment of effects of one agent when placebo
control not permissible
Earlier lecture touched on circumstances
Ethics of withholding a known effective treatment
with life-saving or irreversible life-altering effect
Can be dose-ranging study with single agent
Ethical need to avoid non-effective doses
May not be a practicable approach
Remainder of talk assumes active control study of two drugs
Active Control Studies
Study aspects enabling data to be validly interpreted
Validity of Design
Integrity of Conduct
Adherence to protocol
Same overall issues with standard placebo-control
studies
Additional complexities in Active Control studies
Increased Complexities
Assay sensitivity of the design
Interpretation of the quantitative result
Determination of margin of acceptable difference
Margin
Built on combination of
Historical knowledge of comparator’s efficacy
Clinical judgment of what is an ‘acceptable’
difference
Assay Sensitivity
Can study detect a difference if one exists?
Unrecognized failure of assay sensitivity leads to:
Type II error for superiority study
Type I error for non-inferiority study
Factors which affect sensitivity can seem to impair or
advance study organizer’s goals oppositely
Promotes ability to distinguish between
Evidence of Absence
Absence of Evidence
Laboratory assays often include positive and negative
controls; rarely can be done in clinical trials
Interpretation of Study Results
Results analyzed as comparison of two groups
Need numeric criterion to form interpretation
Placebo control & other superiority studies have same
requirement, but easy to define
Show between-group difference > 0
New drug superior to placebo / other control
Superiority of new treatment over old treatment difficult
to achieve with active agents
Thus non-inferiority approach attractive
Need to have quantitative comparison criterion to
interpret study result, allowing for potentially the same
efficacy
Margin of Acceptability
How much less efficacious can the new drug be than old
drug, and still deem it acceptable to use (or equivalent)?
Two components to consider
First – what is the efficacy the old drug provides?
Statistical analysis of existing information
Can not allow new drug to be worse than old drug
by that amount, or it provides no efficacy
Second – how much of that quantitative amount of
efficacy is it permissible to give up?
Clinical judgment
Usually easier to assess after know first
component
Express as an absolute amount or relative amount
Margin of Acceptability
Two components – M1, M2
Separate basis, separate purpose
Margin of acceptable inferiority cannot be larger than:
Statistical component – M1
Clinical component – M2
Overall margin (M)
If M2 << M1: M = M2
M2 > M1: Either no need for AC study (use placebo
control study) or non-inferiority not feasible
M2 <(modest) M1 leads to complexities and anxieties
Active Control Agent Existing Knowledge
Solid knowledge of efficacy of active control agent is
essential
From prior studies – historical placebo control studies
Preferably from multiple different prior studies
Historical studies each conducted in a specific manner
Regimen of use, population of use
Concomitant care, alternative therapies
Existing knowledge applicable in circumstances of
prior experience
May not be reliable in other circumstances
Quality of historical studies needs to be considered
Adherence to protocol
Quality of data collection for outcome now of interest
(may not have been primary outcome in study design)
M1 – Statistical Margin Component
What is the treatment effect of the comparator agent?
Typical?
Reasonably likely present?
Highly likely present?
There will be no ability to actually confirm in new AC
trial
Generally derived from some form of meta-analysis
With allowance for uncertainty in historical
quantitative estimates (i.e., variance of each effect
estimate; variance of meta-estimate)
The more precisely treatment effect was estimated in
historical trials and the more trials there were, the
more precise the M1 value may be determined
M1 – Estimate from Historical Evidence
Objectively done meta-analysis is important
What studies to include
What portions to include (e.g., patient subsets)
Selection to improve strength of meta-analysis and
applicability to planned new study use of drug
How relevant is historical M1 estimate to a new study:
Relates to issues outside of purely mathematical
variance – Not in meta-analysis
Population unchanged?
Planned eligibility criteria
Unplanned shift in available population related to
changes in medical practice (e.g., development of
alternative treatments) in intervening time
Active Control’s treatment regimen unchanged?
Concomitant care impact changed?
Uncertainties Affecting Interpretation of New Study
Imprecision of control agent’s effect in each of the historical
studies
Partially estimated by statistical variance of each study
Variability of treatment effect size between historical studies
Partially estimated by meta-analysis methods
Uncertainty of quantitative extrapolation of historical treatment
effect to treatment effect of control agent in new study
The Constancy Assumption – relevance of estimate
Not statistically calculable
Can create an (arbitrary) adjustment inserted into
mathematical procedures
Uncertainty of comparison between control and new agent in
the new study
Estimated with statistical variance in new study
AS ALWAYS, Statistical approach adequate only if
non-statistical sources of bias ignorable
Circumstances for Confidence in M1 Estimate
Good quality of prior placebo control studies – A&WC
Multiple prior placebo controlled trials with comparator
Good grounds for combining data in a meta-analysis (all
randomized, combinable doses/regimen; outcome
measure defined, assessed the same way)
Consistent results across all the prior studies
Prior studies done over extended period of time, but
completed not long ago
Allows for some changes in populations, concomitant
care, etc, showing this is not critical
Prior studies done with some variations in design
Implies drug treatment effect not highly sensitive to
design
Better with large treatment-associated efficacy effect
Historical Knowledge for Non-inferiority Purposes
May be source of difficulty
Active control design being used because placebo
control cannot
Due to nature of benefit of established drug
May be few (one?) placebo controlled studies with a lifesaving drug before it becomes an established, necessary
standard of care
Limited ability to form precise estimate of treatment
effect
M2 – Clinical Component
How much efficacy is clinically acceptable to give up in
the new agent and still allow new treatment to be
medically acceptable to use in place of an already
proven active agent
Clinical judgment, given importance of endpoint, nature
of disorder, amount of efficacy of the active control agent
Absolute (% of patients, points on scale) or relative to
control’s effect (% of control’s efficacy)
M2 does not serve to make a hazy M1-statistical
analysis more reassuring by making it more stringent;
M2 is from solely clinical meaning point of view
½ often used
No automatic reason why this is the correct M2 in
different settings
Each case should be considered independently
Non-inferiority Analysis: Conceptual Methods
Sequential CI method
Dual (Double) CI method
Fixed (Pre-determined) Margin method
Synthesis method
Putative Placebo method
Strengths and Weaknesses to each
How confident can ‘we’ be in effect estimates from
historical data?
How many AC studies will we have to consider?
Given nature of benefit, how much risk of being wrong
in favor of new drug is acceptable?
How much risk of being wrong by discrediting new
drug is acceptable?
Conceptual Methods
Sequential CI method
M1 estimate of active comparator’s effect from metaanalysis confidence interval
Apply M2 limit to form overall margin (M)
New agent’s comparative effect estimated with confidence
interval from new study
Show C.I. does not exceed overall margin M
Synthesis method
Combines historical placebo-comparator data and new
study comparator-test agent data into single calculation.
Result is numeric value indicating putative placebocontrolled efficacy estimate of test agent.
Chief difference is conceptual approach of how to address
inter-study variability and strength of constancy
Quality of Study Data & Analysis
Constancy assumption needs good study quality
Effects of study flaws
Anti-conservative; opposite of superiority study
Flaws
Adherence to study protocol
Cross-overs
Completeness of data
Dropouts, missing data
Analysis population
ITT keeps errors in but is statistically pure
Per Protocol keeps errors out, but impure
Probably best to do both and show that all ways of
looking at dataset are consistent
Interpretation of Results
Examples of CIs and assessment
Apparent paradox of non-inferiority success with an inferior drug
a
b
c
d
e
f
Closing Comments
Size of non-inferiority studies
Often large to achieve sufficient precision in estimate
of New-Active effect difference so that CI of
comparison falls within the limited range defined by
margin
When M is small, study is large; when M is large,
study can be moderate in size
Biggest factor can be efficacy of active comparator
Equivalence studies
If this implies greater closeness of effect than “notunacceptably-inferior”, then M is smaller based on M2
component
Consequently sample size of study will increase
Closing Comments
Active Control-Noninferiority studies of successive new
agents ( risk of ‘bio-creep’)
M based on quantitative data applicable to the
comparator being used. Risk of using New1, based
on single AC study vs. Std, as the AC agent for New2
assessment, and so on
New1 might be inferior to original comparator by a
small amount (in itself negligible)
Negligible inferiority added twice (Std-New1; New1New2) may be not negligible ; frequently only 1-2
study with New1 so that meta-analysis of
New1StdPbo gives larger CI, and less assured
efficacy
If ignored, New3 may have little to no efficacy yet
pass the erroneously applied non-inferiority margin
Closing Comments
Planning and conducting a non-inferiority study does not
prohibit achieving a conclusion of superiority if such is
the case. Analytic plan of study can allow for this when
done in the proper manner.
Safety issues also always need considering
Often not as enticing as working on efficacy aspect
Reading and Reporting of Active Control Studies
Planning and analysis aspects unique to AC studies
Great difficulty in assessing study result of these
aspects not well described in publication